xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001920198
007 cr mnu|||uuuuu
008 080107s2007 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002095
Cobb, Sarah Elizabeth.
Structural equation model of exercise in women utilizing the theory of unpleasant symptoms and social cognitive variables
h [electronic resource] /
by Sarah Elizabeth Cobb.
[Tampa, Fla.] :
b University of South Florida,
ABSTRACT: A dramatic decline in physical activity levels occurs from adolescence to young adulthood. Those who were sedentary as adolescents tend to maintain a sedentary lifestyle. Women are particularly vulnerable to the effects of a sedentary lifestyle because of the risk for cardiovascular disease. The purpose of this research was to test a theoretical model of exercise in adolescent and young adult women using the theory of unpleasant symptoms with social cognitive variables and then to test a revised model that was determined a priori. The central hypotheses were that the relationships as depicted in the proposed theoretical models would be reproducible in data from adolescent and young adult women of ages 18 to 25.
Dissertation (Ph.D.)--University of South Florida, 2007.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 206 pages.
Advisor: Mary E. Evans, Ph.D.
t USF Electronic Theses and Dissertations.
Structural Equation Model of Exercise in Women Utilizi ng the Theory of Unpleasant Symptoms and Social Cognitive Variables By Sarah Elizabeth Cobb A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy College Of Nursing University of South Florida Major Professor: Mary E. Evans, Ph.D. Jason W. Beckstead, Ph.D. Janie Canty-Mitchell, Ph.D. Elaine Slocumb, Ph.D. Date of Approval: April 19, 2007 Keywords: LISREL, adolescent girls, young women, physical activity Copyright 2007, Sarah Elizabeth Cobb
Dedication This document is dedicated to my daughters Farah Lee Fl ores and Jennifer Ann Cobb, without whose encouragement and Â“Y OU CAN DO IT, MOM!Â” hollering from the bleachers I never would have survived all this work. This work is in memory of Lieutenant Colonel Ronald St em who exemplified valor in the face of tumultuous times, and taught me tenacit y.
Acknowledgments I thank my brother Lee M. Cobb for his long distance he lp with my inevitable questions, for his many encouraging instant m essages, and for the many files that he located on my behalf. I thank Cecile Lengacher RN Ph.D. for inviting me int o the BSN-PHD program and for her unwavering faith in my abilitie s. I thank Elaine Slocumb RN Ph.D. for her detailed edit ing of my various manuscripts, and her witty laughter that eased my pain during some tough times. I thank Jason W. Beckstead Ph.D. for his gracious persevera nce when instructing me in quantitative methodology, and for h is encouragement in using structural equation modeling for this dissertation resear ch... an inspiring LISRELite and a great professor. I thank Mary E. Evans RN Ph.D. for stepping in as majo r chair advisor when I most needed some guidance, and for her gentle spirit that so often calmed the waters. I thank Janie Canty-Mitchell RN Ph.D. for her inspirati on and dedication to children, and wish her well in her new endeavors. Last but not least, I thank John Ward Ph.D. for providi ng the class that changed my life view about the human body and refute d separation of mind and body. It will definitely affect my future research.
i Table of Contents List of Tables viii List of Figures ix List of Acronyms x Abstract xi Chapter One 1 Introduction 1 Risk of Cardiovascular Disease 1 Impact on Chronic Diseases and Metabolic Syndrome 4 Impact on Musculoskeletal Disorders 6 Maintenance of Physical Activity 8 Theoretical Framework 9 Primary Aims 16 Study Questions 16 Significance 16 Chapter Two 24 Literature Review 24 Theoretical Background 24 Preliminary Studies 25 Qualitative 25
ii Quantitative 25 Factors in the Model of Exercise 27 Psychological Latent Variable 27 PS-1: Self-efficacy for exercise 36 PS-2: Outcome expectations for exercise 36 PS-3: Self-regulation 41 Situational factor 41 S-1: Loneliness 41 S-2: Social support 44 Physiological latent variables 48 PH-1: Exercise capacity 48 PH-2: Perceived health status 50 Ph-3: Anticipated exercise fatigue 52 Unpleasant symptoms 53 US-1: Chronic fatigue 53 US-2: Pain 55 Activity 59 E-1 Exercise 59 Chapter Three 61 Design and Methods 61 Overview of Designs 61 Sample Description and Selection 61 Sampling Frame 61
iii Sampling Size 61 Participants 64 Procedures 65 Data Collection 65 Measures 67 Measures for the Psychological Latent Variable 67 PS-1: Self-efficacy for exercise 67 PS-2: Outcome expectations for exercise 68 PS-3: Self-regulation for exercise 68 Measures for the Situational Factor 69 S-1: Loneliness 69 S-2: Social support for exercise 70 S-3: General social support 70 Measures for the Physiological Factor 71 PH-1: Exercise capacity 71 PH-2: Perceived health status 72 Ph-3: Anticipated exercise fatigue 72 Measures for Unpleasant Symptoms Factor 73 US-1: Chronic fatigue 73 US-2 and US-3 Chronic pain 73 Measure for Activities 74 E-1: Exercise 74 Reliability and Validity of Research Design 75
iv Assumptions 76 Model Identification 77 Data Analysis 78 Parameter Estimation 82 Data Preparation and Screening 84 Missing Data 84 Preliminary Analyses 86 Multivariate normality 86 Measurement Model 88 Validity and reliability of measurement model 89 Structural Models 91 Assessment of structural models 93 Model Modifications 88 Overview of Model Specification Methods 94 Goodness of Fit Indices 96 Testing of Specific Aims 99 Aim 1 99 Aim 2 99 Power Analysis Post Analysis 99 Human Subjects Research 100 Risks to Subjects 100 Human subject involvement 100 Sources of Materials 100
v Potential Risks 101 Adequacy of Protection against Risk 101 Recruitment and informed consent 101 Protection against risk 101 Potential Benefits of the Proposed Research 102 Inclusion of Women, Minorities, and Children 102 Chapter Four 104 Results 104 Overview of Analytic Strategy 104 Participant Characteristics 104 Preliminary Analyses 105 Data quality 107 Original Assessment of Bivariate Relationships 109 Original Assessment of Measurement Model 109 Final Assessment of Bivariate Relationships 114 Final Assessment of Measurement Model 119 Validity and reliability 119 Interrelations among latent factors 124 Goodness of fit 126 Assessment of Structural Models 126 Hypothesis Testing 129 Assessment of Model Fit 129 Aim 1 129
vi Aim 2 132 Fatigue 134 Pain 134 Indirect and total effect of independent variable 134 Squared multiple correlations 135 Model Modifications 137 Model Two Diagnostics 137 Modifications Made 138 Results from Modifications 138 Model Cross-Validation 140 Power Analysis 140 Summary 141 Chapter Five 142 Discussion 142 Aim 1: Testing the Theory of Unpleasant Symptoms 1 42 Fatigue 143 Pain 144 Aim 2: Testing the Alternate Model 2 144 Implications for the Use of the Theory of Unpleasant S ymptoms 144 Implications for Nursing Intervention 147 Limitations of the Study 149 Directions for Future Research 151 Lessons Learned 151
vii List of References 154 Appendices 177 Appendix A: Exercise Self-efficacy Scale 178 Appendix B: Outcome Expectations for Exercise Scale 1 79 Appendix C: Exercise Goals Scale 180 Appendix D: UCLA Loneliness Scale (ULS-8) 181 Appendix E: Social Support for Exercise Scale 182 Appendix F: Multidimensional Scale of Perceived Social Support 183 Appendix G: Rating of Perceived Capacity 184 Appendix H: SF-12 v2 185 Appendix I: Situational Fatigue Scale 188 Appendix J: Chalder Fatigue Scale 189 Appendix K: West Haven-Yale Multidimensional Pain Sca le 190 Appendix L: International Physical Activity Questionna ire 191 Appendix M: Demographic Form 193 Appendix N: Elements of Informed Consent 194 Appendix O: Covariances and Variances for Actual Data ( N = 463) 196 Appendix P: Covariances and Variances for Implied Dat a (N = 463) 200 Appendix Q : Syntax Used for Post-hoc Power Analysis in SPSS 205 Appendix R: Syntax Used to Calculate Delta and Neede d Sample Size 206 About the Author End Page
viii List of Tables Table 1 Definitions of Terms 19 Table 2 Role of Self-efficacy in Structural Equation Models for Exercise 31 Table 3 Diversity Profile of All USF Students 64 Table 4 Summary of LISREL Matrices and Greek Notation 83 Table 5 Targeted / Planned Enrollment Table 10 3 Table 6 Univariate Normality Z Scores 107 Table 7 Regrouping of Indicators and Constructs 11 1 Table 8 Bivariate Correlations 115 Table 9 Measurement Model: Completely Standardized Coefficients 121 Table 10 Composite Reliabilities and Average Varianc e Abstracted 124 Table 11 Standardized Covariances among Latent Varia bles 125 Table 12 Goodness of Fit Indices for Modified Models 139
ix List of Figures Figure 1 The Theory of Unpleasant Symptoms 12 Figure 2 Model of Exercise Utilizing the Theory of U npleasant Symptoms 14 Figure 3 Model of Exercise Altering the Theory of Un pleasant Symptoms 15 Figure 4 Social Cognitive Model of Physical Activity 29 Figure 5 Data Collection Process 66 Figure 6 LISREL Steps 81 Figure 7 Measurement Model 88 Figure 8 Structural Model 1 91 Figure 9 Structural Model 2 92 Figure 10 Measurement Model Results 120 Figure 11 Revised Model 1 Path Diagram 127 Figure 12 Revised Model 2 Path Diagram 128 Figure 13 Completed Structural Model 1 131 Figure 14 Completed Structural Model 2 133 Figure 15 Squared Multiple Correlations 136
x List of Acronyms 1. ANCOVA Analysis of covariance 2. ANOVA Analysis of variance 3. BMI Body Mass Index 4. CAD Coronary artery disease 5. CVD Cardiovascular disease 6. HOMO-IR Homeostasis insulin resistance 7. HS-CRP High sensitivity C reactive protein 8. IL-6 Interleukin-6 9. MANOVA Multivariate analysis of variance 10. MANCOVA Multivariate analysis of variance 11. MET Metabolic equivalent 12. MPA Moderate physical activity 13. MVPA Moderately vigorous physical activity 14, PA Physical activity 15. SEM Structural equation modeling 16. VPA Vigorous physical activity 17. VO 2 max Maximal aerobic capacity
xi Structural Equation Model of Exercise in Women Utilizi ng the Theory of Unpleasant Symptoms and Social Cognitive Variables Sarah Elizabeth Cobb ABSTRACT A dramatic decline in physical activity levels occurs from a dolescence to young adulthood. Those who were sedentary as adolescent s tend to maintain a sedentary lifestyle. Women are particularly vulnerable to the effects of a sedentary lifestyle because of the risk for cardiovascular disease. The purpose of this research was to test a theoretical model of exercise i n adolescent and young adult women using the theory of unpleasant symptoms w ith social cognitive variables and then to test a revised model that was det ermined a priori. The central hypotheses were that the relationships as depicte d in the proposed theoretical models would be reproducible in data from adolescent and young adult women of ages 18 to 25.
1 CHAPTER ONE Introduction Promoting exercise among the United States (U.S.) pop ulation is a national priority. Several of the Healthy People 2010 goals specifically target exercise to increase the proportion of adolescents and a dults who engage in moderate physical activity (U.S. Department of Health and Human Services, 2002). Current recommendations for physical activity di ffer by age. Recommendations by the Centers for Disease and Control ( 2006) are that youth participate in physical activity for 60 minutes at moder ate intensity on most days of the week, preferably daily. Recommendations for adu lts are that they participate in vigorous activity for 20 minutes on at l east three days per week, or engage in moderate activity for 30 minutes on at lea st five days per week (CDC). Furthermore, the exercise does not have to be done all at once; it is beneficial even if the exercise time is divided into portions as sma ll as 10 minutes (CDC). Risk of Cardiovascular Disease There are several reasons why exercise has been emphasize d as a national priority. One of the key reasons is the risk of cardiovascular disease (CVD) or coronary artery disease (CAD) from the combin ed effects of physical inactivity / obesity. Women are particularly vulnerabl e to the effects of a sedentary lifestyle because of the risk for cardiovascular disease (Correa-de-
2 Araujo et al., 2006). In the decade prior to year 20 00, the number of deaths attributable to poor nutrition/physical inactivity incre ased substantially more than the other causes of death (CDC, 2005; Mokdad, Marks, Str oup, & Gerberding, 2004). Mokdad et al. calculated the number of deaths a ttributable to poor nutrition/physical inactivity from the percentage of pe rsons who were overweight or obese; using this method, 400,000 (16.6%) of death s in year 2000 were attributed to poor diet and physical inactivity, which was an increase from the 300,000 (14%) in year 1990. Mokdad et al. were able to use obesity as a proxy for poor physical inactivity because of the high correlat ion between the obesity and poor physical fitness. Recent evidence showed that th e effect of body mass index (BMI) on predicting physical fitness was strong amon g healthy youth ( p < 0.0001) with a decrease of 0.069 minutes treadmill en durance for each unit increase in BMI (Chatrath, Shenoy, Serratto, & Thoele 2002). YouthÂ’s physical activity indices can predict BMI and adult waist circumferences as well. X. Yang et al. (2006) tested a mo del of physical activity and obesity longitudinally from 1980 through 2001 in four cohorts of youth (ages 9, 12, 15, and 18). After following these cohorts for 21 years, X. Yang et al. found a significant total effect that youthful physical activi ty had on adult waist circumference ( = .07, t =4.54, p < .05). Furthermore, youthful BMI accounted for 13% of the variance in the adulthood waist circumfe rence. Interestingly, Wessel et al. (2004) found that among 906 women (mean age 58, SD 12 years) referred for clinically indicated coronary angiography, those who were found to have higher BMI were likely to hav e a history of hypertension,
3 diabetes, dyslipidemia, and higher IL-6 levels and pre valence of metabolic syndrome (Wessel et al.). However, despite having these CAD risk factors associated with higher BMI, neither BMI nor anthropome tric measures (waist circumference, waist/hip ratio, and waist/height ratio) were associated with the risk of mortality or major adverse events ( p >.10). Instead, Wessel et al. found the risk of mortality was associated with poor physical fit ness from physical inactivity, not higher BMI. To summarize, BMI and CAD risk factors are associated with each other, but it is the physical inactiv ity leading to poor physical fitness that is associated with mortality risk, and as noted by X. Yang et al. (2006), youthful physical activity can deter adulth ood obesity significantly. Similar associations of CAD risk factors and BMI were fou nd by McGavock, Anderson, and Lewanczuk (2006). In a study amon g 135 otherwise healthy young adults (mean age for females 28 5 yea rs) categorized into three groups (sedentary, physically active, and endurance train ed), BMI was significantly associated with systolic blood pressure ([BP], ( r = 0.36, p <.01) but was unrelated to large or small artery compliance (McGav ock et al.). The strong association between BMI and CAD risk factors or poor physical fitness, which directly affects mortality risk, is considered evidence that obesity is a risk factor for the adolescents and the young adults und er consideration. Whitlock, Williams, Gold, Smith, and Shipman (2005) f ound in an integrative review of evidence that single BMI measures successfully predicted risk factors in young adulthood in longitudinal studies, particula rly for youth over age 13 (r = > 0.6). To date, BMI is considered the most reliable scre ening test for overweight
4 in childhood for predicting obesity in adulthood. Adol escents who are overweight with BMI > 95 th percentile have a 50% probability of adult obesity ( Whitlock et al.). In summary, youthful physical activity predicts ad ulthood physical activity, which predicts adult waist circumference. Furthermore, ph ysical inactivity at any age is associated with poor physical fitness, which is a prim e indicator of cardiac fitness. Impact on Chronic Diseases and Metabolic Syndrome A second reason for the emphasis on engaging in physical a ctivity is the effect that physical inactivity/low cardiorespiratory fit ness has on chronic diseases and prodromal conditions such as metabolic syndrom e. Physical inactivity has been shown to impact the risk of diabete s mellitus and certain cancers (Warburton, Nicol, & Bredin, 2006) as well as to exacerbate the risk of mortality from any cause whether or not a chronic disease is present (Wessel et al., 2004). One such condition is the metabolic syndrome which is a phenotype that links insulin resistance, hypertension, dyslipidemia, type II diabetes, and other metabolic abnormalities with an increased risk of a therosclerotic cardiovascular disease (R. Weiss et al., 2004). The metab olic syndrome is characteristic for nearly half of severely obese patient s (R. Weiss et al.), and is a major risk factor for coronary artery disease(Council on Sports Medicine and Fitness & Council on School Health, 2006), particularly among women (LaMonte et al., 2005). However the metabolic syndrome is also l inked to physical inactivity (McGavock et al., 2006) not just to BMI.
5 McGavock et al. (2006) studied healthy young adults (age s 20-40) for physiologic differences among sedentary, physically active and endurance trained participants. Using a glucose breath test as a no ninvasive measure of insulin sensitivity, McGavock et al. determined that ther e was a trend toward reduced insulin sensitivity in the sedentary group; fasti ng insulin levels were nearly twice as high in sedentary participants compared t o the endurance-trained participants, and there was a concomitant increase in home ostasis insulin resistance (HOMO IR) levels. These authors believed that sedentary lifestyles lead to cardiac dysfunction and vascular changes by causing a progressive decline in insulin sensitivity. The relationship of obesity to the metabolic syndrome wa s reported by R. Weiss et al. (2004) who studied the metabolic syndrome i n youth ( N = 439); youth were included in the exposed group if their BMI exceed ed the 97 th percentile for their age. The authors found that values for serum gl ucose, insulin, insulin resistance (HOMO IR), IL-6 and systolic BP all increased wi th increasing overweight ( p < 0.001). The overall prevalence for the metabolic syndrome ranged from 38.7% to 49.7% for the moderately obese and the severely obese participants respectively, while there were no cases of me tabolic syndrome among the nonobese participants (R. Weiss et al.). Each h alf-unit increase in BMI (measured in Z scores) significantly increased the risk of the metabolic syndrome ( OR 2.20; 95% CI 1.35 Â– 3.59). At follow-up two years later, eight of the participants who had had impaired glucose tolerance at baseline had
6 developed type II diabetes (R. Weiss et al.). The odds of developing type II diabetes are increased in adolescence if the youth are o verweight. Iannuzzi et al. (2006) also studied the metabolic syndr ome among obese children ( N = 100 obese youth, ages 6 to 14). Obese children with m etabolic syndrome had significant differences relative to nonobe se youth; obese youth had higher insulin levels ( p = 0.014), higher HOMO IR levels ( p = 0.011), and Creactive protein concentrations ( p = 0.021). Using ultrasound parameters for carotid thickness and stiffness, the obese children with met abolic syndrome also had significantly more carotid stiffness than nonobese chi ldren ( p = 0.023). LaMonte et al. (2005) prospectively studied adults ( N = 1,491 women and 9,007 men, mean age 44 9 years) for cardiorespirator y fitness relative to the incidence of metabolic syndrome. Among this group of adu lts, low cardiorespiratory fitness was significantly related to th e development of the metabolic syndrome risk factors. A one metabolic equivalen t (MET) increment in treadmill performance was associated with a 17% reduction in risk of metabolic syndrome for women (LaMonte et al.); in contrast, a sig nificant inverse linear relationship was noted between cardiorespiratory fitness and the metabolic syndrome ( p = 0.02 for women, p < 0.0001 for men). Impact on Musculoskeletal Disorders A third reason for advocating physical activity is that p hysical inactivity/ overweight in youth contributes to an increased risk fo r musculoskeletal disorders such as slipped capital femoral epiphysis, adolescent tibia vara, joint pain especially in the knees, and fractures (Taylor et al., 2 006). Among a total of 355
7 youth (mean age of the overweight = 12.6 2.7) fol lowed prospectively, the prevalence of musculoskeletal complaints was higher for th e overweight group compared to the nonoverweight group ( OR 4.41; 95% CI : 1.3-15.0, p = 0.0096). Taylor et al. noted that the customary increase in bo ne density seen in overweight children is not sufficient to overcome the f orces that are generated when a child falls, for example, and that overweight youth fall with a much greater force than do nonoverweight youth. One of the concerns about high levels of physical activi ty is a possible reduction in bone density. In a prospective cohort stud y among young women followed for 2 years for changes in bone density, neit her body weight nor change in body weight explained the variability in bone de nsity at time 2 (Elgan & Fridlund, 2006). Among those who were underweight (B MI < 19), high physical activity hindered bone density ( = 0.139, SE = 0.04, p = 0.004). However in contrast to underweight women, Elgan and Fridlund fou nd that the bone density at time two among women with a BMI >24 was not affect ed by increased physical activity ( p = 0.689). Thus physical activity should not be restricted among young women with BMI greater than 24 due to fears of chang e in bone density. In prepubescent children, a notable osteogenic effect can be achieved with only a few hours of sports participation (VicenteRodriguez, 2006), physical activity stimulates bone hypertrophy and increases peak mass. In their position statement on osteoporosis and exercise, the American Coll ege of Sports Medicine (1995) noted that habitual inactivity causes ra pid decrease in bone density, whereas the effect of habitual exercise is less ra pid increase in bone
8 density. As women age, it becomes more difficult to main tain the load-bearing stimulus needed for stimulating bone mass (American Colle ge of Sports Medicine). Stimulating bone density growth through re gular physical activity is essential for women particularly as they mature (Borer, 2005). Maintenance of Physical Activity However maintaining regular physical activity for wome n is an issue. Dramatic declines in physical activity levels occur between adolescence and young adulthood (Gyurcsik, Bray, & Brittain, 2004). Wo men who are sedentary as adolescents tend to maintain a sedentary lifestyle (D e Bourdeaudhuij, Lefevre et al., 2005; De Bourdeaudhuij, Philippaerts et al., 2005). Similar findings for young adults were noted by the CDC (2005). In the 18 to 44 age bracket, 32.9% had a sedentary lifestyle in 2003; overall, 37.6% of adults are inactive. According to a recent report by the National Heart Lung and B lood Institute (Krumholz et al., 2005), weight gain over 10 years (defined as an increa sed BMI of 5 kg/m 2 ) was the highest at ages 25 to 34. Therefore physical activit y is even more important as adolescents prepare for young adulthood before the sp urt in BMI occurs. In summary, if weight gain can be avoided before thos e critical years, CVD risk factor levels can be reduced and can obviate the need for costly drug therapy later in life (Krumholz et al., 2005). Physi cal inactivity or sedentary lifestyles have substantial healthcare costs associated with them (Weiss, Froelicher, Myers, & Heidenreich, 2004). However, mor e importantly, physical activity can save lives and increase the quality of life (Jia & Lubetkin, 2005).
9 Despite the growing body of evidence pertaining to t he need for exercise, there are gaps in the literature related to impedime nts to exercise, particularly in the adolescent and young adult population. Impedimen ts such as fatigue and pain were theorized to have an impact on physical activ ity outcomes even for this age population. In addition, there are gaps in the l iterature related to theoretical models of exercise in the adolescent and young adult po pulation. Theoretical Framework Based upon a qualitative pilot study (Cobb, 2005), th e exercise experience of women was often described as having been affected by the symptoms of pain and of fatigue, both of which are key symptoms in the t heory of unpleasant symptoms. In the search for a model to test concerning exercise, the theory of unpleasant symptoms by Lenz, Suppe, Gift, Pugh, and Mi lligan (1995) and revised by Lenz, Pugh, Milligan, Gift, and Suppe (19 97) emerged as a plausible theory that could explain the results of the qualitati ve study. Examples of the use of the theory found in the literature included a cor relational study relating fatigue and exercise among older women who have experienced a myocardial infarct (Crane, 2005); studies relating fatigue and post-part um depression (Corwin, Brownstead, Barton, Heckard, & Morin, 2005), and studie s relating fatigue to various pathologies such as chronic obstructive pulmonary d isease (Reishtein, 2005), end stage renal disease (Liu, 2006; McCann & Bo ore, 2000), and cancer (Redeker, Lev, & Ruggiero, 2000). Other symptoms that have been studied using this middle-range theory include the symptom of nausea (OÂ’Brien, Evans, & White-McDonald, 2002). However these studies mostly used the theory of
10 unpleasant symptoms to explain activity outcomes among older persons with chronic illnesses. To this researcherÂ’s knowledge, the theor y of unpleasant symptoms has never been testing using structural equati on modeling. Thus the possibility of using the theory of unpleasant symptoms was explored further. One key feature of the theory of unpleasant symptoms is that multiple symptoms affect performance outcomes. Originally the t heory of unpleasant symptoms was conceived as a single concept of fatigue durin g postpartum. This single concept eventually merged with the single concept of fatigue during intrapartum to become the framework for the study of f atigue during childbearing. Meanwhile, the single concept of fatigue during intrap artum merged with the single concept of dyspnea in chronic obstructive pulmonary disease and asthma to become the multiple concepts of dyspnea/ fatigue. Thu s three single concepts through collaboration with their various authors merge d into two multiple concepts. These two multiple concepts then were merged i nto a middle-range theory of unpleasant symptoms (Lenz et al., 1995). K ey considerations to the merging of the concepts were that both fatigue and dysp nea were defined by the same subjective symptoms, could be altered by anxiety or depression, and had similar physiological, psychological and situational factor s as antecedents. The symptom experience of either fatigue or dyspnea could in turn influence functional performance (Lenz et al., 1995). The updated theory of unpleasant symptoms (Lenz et al. 1997) asserted that while symptoms can occur in isolation, they often occur simultaneously. Multiple symptoms catalyze each other; these multiple sy mptoms are
11 multidimensional, with duration, timing, intensity, and quality being dimensions of each symptom. In the updated theory of unpleasant sympt oms by Lenz et al. (1997) the physiologic, psychological and situational fact ors are depicted as being related, and performance is depicted as having a reciprocal effect onto the same three factors (see Figure 1). Because the definitio ns of each of the factors described by Lenz et al. included multiple examples, th e factors can be considered as domains, and will be referred to as domai ns in this document.
12 Figure 1 The Theory of Unpleasant Symptoms. Used by permission ( Lenz, E. R., Pugh, L. C., Milligan, R. A., Gift, A., & Suppe, F. (1997). The middle-range theory of unpleasant symptoms: An update. ANS. Advances in Nursing Science, 19 (3), 14-27). Lenz et al. (1997) depicted unpleasant symptoms as media ting the relationship between the psychological, situational, and physiological factors and performance outcome. However they stated that unpleasa nt symptoms moderated the relationship (see relationship among influential factors paragraph two). In their model, each of the influencing factors r elated to each other as well as interacted to influence the symptom experience. Accord ing to Lenz et al. (1997), the psychological domain includes knowledge relat ed to symptom, stress and other affective reactions as well as social support. T he physiological domain
13 includes nutritional balance and both pathological and normal body systems. Of particular interest is that the situational domain inclu des social and physical environment influences such as social support, access to heal thcare, family status, ambient temperature, humidity, and air qualit y (Lenz et al., 1997). It also includes lifestyle behaviours such as physical activity or n utrition. Finally, the outcome component of the theory is that of performance which includes functional performance, functional health status, qual ity of life, and cognitive activity. Examples of functional performance given are physical activity, social activities and interaction, and work. Cognitive activity includes problem-solving as well as lower cognitive functioning (Lenz et al., 1997 ). The theoretical model for this in structural equation format is depicted in Figure 2. Close examination of the conceptual definitions reveal ed some ambiguous boundaries, with social support listed in both psychologica l and situational factors; physical activity listed in both situational and performance factors, and nutrition listed in both physiological and situational factors. These ambiguities as well as theoretical concerns located in the literature r eview prompted the proposed model 2 (see Figure 3), in which unpleasant sy mptoms partially mediate the relationships between the other factors and physical activity. Other reasons for choosing a partially mediated model were th at the literature review revealed different relationships amongst the variable s for social support and the psychological domain, as described in more detail in chap ter 2.
14 The main purpose of this research was to test if the mo dels for the theory of unpleasant symptoms would be reproduced in the dat a from college women of ages 18 to 25. Figure 2 Model of Exercise Utilizing the Theory of Unpleasant Sym ptoms.
15 Figure 3. Model of Exercise Altering the Theory of Unpleasant Symp toms.
16 Primary Aims 1. Assess whether the relationships as depicted by the mod el for the theory of unpleasant symptoms would be reproducible in data f rom women of ages 18 to 25. 2. Assess whether the relationships as depicted by the alt ered model for the theory of unpleasant symptoms would be reproducible i n data from women of ages 18 to 25 with a better fit than the fi rst model. Study Questions 1. Will model 1 be reproducible in data for women ag es 18 to 25? 2. Will the altered model, model 2, be reproducible in the data with a better fit than model 1? Significance Physical inactivity or sedentary lifestyles provoke an eco nomic burden that has burgeoned into an epidemic proportion among the l ast 20 years. Studying national data from hospital discharges of children and adolescents, Wang and Dietz (2002) found that there was a 197% increase for obesity-related diagnoses among those discharged. The frequency with which they found obesity listed as a secondary diagnosis showed that obesity may lead to other conditions, including asthma, adverse pregnancy outcomes, sleep apnea, and gal lbladder disease. When obesity was listed as the principal diagnosis, the av erage length of stay was 13.5 days, more than twice the 6.8 days where obesit y was the secondary diagnosis. This amounted to a cost of $127 million per y ear in 2001 dollars, which is more than a threefold increase.
17 Physical inactivity or sedentary lifestyles have substanti al healthcare costs associated with them for young and middle aged adults as well (J. P. Weiss et al., 2004). J. P. Weiss et al. performed a cost analysi s of healthcare costs and exercise capacity among veterans (mean age 59). There wa s an inverse relationship between exercise capacity (measured in METs) and costs that was independent of age. With each one MET increase in exe rcise capacity, costs were incrementally lower by an average of 5.4% ( p < 0.001). A higher peak MET was significantly associated with lower costs one year later (J. P. Weiss et al.). Therefore exercise, which increases exercise capacity, can re duce long-term healthcare costs. Similar findings exist among those with advancing age. Pronk, Goodman, OÂ’Connor, and Martinson (1999) studied health care ch arges billed to a stratified random sample of 8000 individuals aged 40 years or ol der who had at least one of four chronic diagnoses: diabetes mellitus, heart disea se, hypertension, or dyslipidemia. The health care charges were highly skewed with 86% of the total charges accrued by a quintile of individuals (Pronk et al .). Healthcare costs for sedentary individuals (no physical activity done during a week) were 4.7% higher than for those who were physically active even just one day per week, even after controlling for the chronic diseases (Pronk et al.). Ano ther interesting finding was that females had median charges that were 39% higher t han for males These three studies have shown that physical inactivity cost s our nation millions of dollars. As noted earlier, sedentary rates i ncrease with age, especially among women as they make the transition from adolescen ce to young
18 adulthood. Physical inactivity is a modifiable behavior that has significant impact on our nationÂ’s women in particular. Learning ways to modify behaviors to promote physical activity is vital to our national heal th. In summary, the context of the need for this research w as introduced in this chapter. Key points were that physical activity decre ases with age and that women are particularly susceptible to cardiac events. Beca use of the inverse relationship between cardiac events and cardiac fitness, exercise to increase the cardiac fitness is a valuable tool in prevention of card iac events. In chapter 2, the major concepts of the theoretical mode l are introduced and the preliminary pilot study leading to the inter est in these variables is discussed. In the proposed model, there are thirteen key variables, each of which is discussed in depth in the following chapter. The liter ature review is presented sequentially by latent variables, with each manifest v ariable described. Table 1 provides the definitions of key terms used in the lite rature review.
19 Table 1 Definition of Terms Term Definition Reference Body Mass Index BMI 1. Weight in KG / Height M 2 2. Weight (kg) / height (cm) / height (cm) X 10 000 3. Weight (lb) / height (in) / height (in) X 703 (Council on Sports Medicine and Fitness & Council on School Health, 2006) Metabolic Equivalents (MET) 1 MET = resting metabolic rate; rate of O2 consumption by normal adult at rest 2. 1 MET = 3.5 ml O2 /kg /min (Bulwer, 2004) Obesity 1. BMI z score of 2.0 or more 2. BMI 30 (R. Weiss et al., 2004; Wessel et al., 2004) Moderately obese 1. BMI z score of 2.0 to 2.5 ( R. Weiss et al., 2004) Severely obese 1. BMI z score > 2.5 (R. Weiss et al., 2004)
20 Table 1 (Continued) Term Definition Reference Overweight 1. BMI 95 th percentile for sex and age according to 2000 CDC growth charts (term used by CDC for children and adolescents) 2. BMI 25 Â– 29 adults (Miech et al., 2006; Wessel et al., 2004; Whitlock et al., 2005) At risk for overweight BMI in 85 th to 95 th percentile for age and gender; term used for children and adolescents (Whitlock et al., 2005) Exercise 1. Acute: Any bout of nonhabitual activity 2: Chronic: fitness training May be classified as: 1. Resistance (weight training) 2. aerobic (cardio respiratory training) 3. Flexibility (Stretching) (Bulwer, 2004)
21 Table 1 (Continued) Term Definition Reference Physical activity (PA) Activity besides that which occurs in normal work day, which consumes energy (Bulwer, 2004) Physical inactivity A dichotomous measure indicating respondents who reported both no moderate and no vigorous physical activity over a specified time period of 7 to 30 days (Miech et al., 2006) Sedentary lifestyle Latin for Â“usually sittingÂ” 1. <30 minutes/ day of aerobic exercise on < 3 days/week 2. demanding PA does not exceed 20-minute session, or if occur < 3 times / week (Bulwer, 2004; McGavock et al., 2006) Light PA 1. Physical activity that uses < 4 METs (Bulwer, 2004)
22 Table 1 (Continued) Term Definition Reference Moderate PA 1. Physical activity using 5 METs, or 4 6 METs 2. Physical activity that burns 3.5 Â– 7 calories per minute (Kcal/min) 3. 30 Â– 45 min/day of moderate aerobic exercise 4. PA that burns near 150 Kcal / day or 1000 Kcal/wk (Bulwer, 2004; Centers for Disease Control and Prevention, 2006; K. M. Harris, Gordon-Larsen, Chantala, & Udry, 2006; McGavock et al., 2006) Moderately Vigorous PA (MVPA) 5-8 METS (Nelson & GordonLarsen, 2006)
23 Table 1 (Continued) Term Definition Reference Vigorous PA (VPA) 1. Physical activity using 8 metabolic equivalents (or 6 METs if skipping MVPA) 2. Physical activity that burns more than 7calories per minute (Kcal/min) 3. Endurance trained: > 45 min/day of moderate to intense aerobic exercise 5 days/week (Bulwer, 2004; Centers for Disease Control and Prevention, 2006; K.M. Harris et al., 2006; McGavock et al., 2006) VO 2 max 1. Measure of maximal aerobic capacity as determined by a treadmill test 2. O 2 max, mL*kg -1 min -1 (Kasa-Vubu, Ye, Borer, Rosenthal, & Meckmongkol, 2006; McGavock et al., 2006)
24 CHAPTER TWO Literature Review In this chapter, the key psychological, situational, and physiological variables are discussed in depth as they relate to exer cise among young women. Self-efficacy, outcome expectations for exercise and selfregulation (goals) are discussed as indicators for the psychological factor of the mo del. Loneliness, social support for exercise, and general social support are discussed as indicators for the situational factor of the model. Lon eliness was selected as an indicator for the situational factor based upon Lenz et al. (1997) description of the situational factor as including the relationships wi th others as well as with the physical environment. Exercise capacity, physical health sta tus, and anticipated fatigue in different activity situations are discussed as indicators for the physiological factor of the model. The unpleasant sympt oms of chronic pain and chronic fatigue are discussed. And finally, the concept o f exercise is discussed. Understanding the contribution of each of these variabl es to the overall model is one key to understanding the proposed model of exerci se in women. Theoretical Background As mentioned in the previous chapter, the choice of usin g the theory of unpleasant symptoms as the theoretical basis for this res earch emerged as a consequence of seeking a theoretical model to help explai n the findings from a
25 qualitative study. The preliminary studies leading up to testing the theory of unpleasant symptoms via structural equation modeling are discussed next. Preliminary Studies Qualitative. In a pilot qualitative study (Cobb, 2005) college-age d adolescent women ( N = 4) from ethnic minority groups and of ages 18 to 25 we re interviewed individually about their exercise experien ces. Cultural differences were noted between Black women from the Caribbean Isla nds and Black women of African-American origin from the United States. How ever unpleasant symptomatology was a consistent reason across both cultures f or stopping exercise, with fatigue and pain being the two most fre quently mentioned symptoms. Although all the participants were university students who were knowledgeable about the benefits of exercise, exercise wa s not a priority with them. The question of how to promote physical activity among those who already knew the benefits intrigued this investigator and led t o a desire to research the influence of unpleasant symptoms on exercise in more de pth. Quantitative. In a pilot study (Cobb, 2006) young collegiate female s ( N = 41, M age = 24.29, SD = 3.3, range 22 Â– 37) were queried about their exe rcise habits, unpleasant symptoms (defined as fear of pain, c hronic fatigue, and loneliness), positive aspects (defined as benefits of exe rcise, perceived health status, and perceived exercise capacity), pros/cons of decision s about exercise, and the need for cognition when making choices. Data re vealed a large range of activity with a mean of 3549 MET per minute per week, which indicated that the mean activity level was within the range categorize d as high physical activity
26 levels (IPAQ, 2005). Their average perceived exercise ca pacity was 10.29 metabolic equivalents, which indicated that their percei ved exercise capacity was slightly higher than that of women from the same deca de of life as measured by the scaleÂ’s authors (Wisen, Farazdaghi, & Wohlfart, 200 2). According to the authors, the predicted and the objectively tested meta bolic equivalents were 11.4 and 11.2 respectively, and the self-rating of perceived exercise capacity was 9.2 ( SD = 1.5). Interestingly the female students in the pi lot study reported higher levels of loneliness than national norms ( M = 50.6, SD = 7.6 pilot versus M = 34.5, SD = 18.2 national) as reported by Hays and DiMatteo (19 87). In this study, 43.9% of the students were classified as lonely according t o the cutoff point given by the authors. And finally, the mean summative fatig ue score was 30.19 ( SD = 5.12), which was slightly above the cutoff point of 28 designated as the point of fatigue by the authors (Chalder et al., 1993). Fifty -one percent of the students were classified as fatigued. And finally, the perceived rating of exercise capacity was the only variable to even approach significance as a predictor of metabolic minutes per week ( f (1, 59) = 3.28, p = .069). In summary, the qualitative and quantitative data b oth showed that collegiate females of ages 18 to 25 do have fatigue an d pain, although the relationship of the unpleasant symptoms with exercise is not clear. Structural equation modeling with an appropriate sample size is warranted for testing these relationships.
27 Factors in the Model of Exercise The theory of unpleasant symptoms utilizes five concepts : a) psychological, b) situational, c) physiological factors, d) unpleasant symptoms as the mediating latent variable, and e) activity as the final variable. Indicators for these five latent variables were selected based upon the research for each variable. Psychological Latent Variables Self-efficacy, the primary construct from the social cogn itive theory of behavioral change, was used for the psychological factor. Using literal dictionary definitions, self-efficacy is the awareness of oneÂ’s abili ty to be effective and to control oneÂ’s actions and outcomes (Merriam-Webster, 200 7). In KearÂ’s concept analysis of self-efficacy, three characteristics emerged: a) self-concept, b) control, and c) cognitive processes. Antecedent conditions were social experiences, efficacy expectations, and mastery experiences. According to BanduraÂ’s social cognitive theory (Bandura, 1986, 1994; Bandura, Adams, & Beyer, 1977) perceived self-efficacy is defined as personsÂ’ beliefs about their capabilities to produce certain lev els of performance that influence events that affect their lives. People with high confidence in their capabilities approach difficult tasks differently than tho se who doubt their capabilities. People with high confidence view the diff icult tasks as challenges to be mastered, whereas people with low confidence shy awa y from difficult task. There are four key sources of self-efficacy: Mastery exper ience, vicarious experience, social persuasion, and alteration of somati c and emotional states
28 (Bandura, 1994). Mastery experiences boost personsÂ’ conf idence in their capabilities to succeed and provide a sense of resilience A vicarious experience is observation of someone elseÂ’s modeling a behavior (Ba ndura et al., 1977). Repeated observations of successful performances boost pe rsonsÂ’ confidence in their own capabilities to succeed, particularly when the social models possess similar characteristics to the persons (Bandura, 1994). Soci al persuasion is done through verbal assurances that they possess the capabiliti es to master given activities. And finally, by altering personsÂ’ negative emotional proclivities and interpretations of their physical states, their stress is re duced and self-efficacy is boosted (Bandura, 1994). The core determinants of self-efficacy are knowledge, p erceived selfefficacy, outcome expectations, goals, perceived facilita tors, and impediments to the changes one seeks (Bandura, 2004). Indicators for psy chological factor of this study were chosen to reflect three of these dimensi ons of self-efficacy: a) self-efficacy for exercise, b) outcome expectations for exe rcise, and c) goal setting for exercise. Figure 4 depicts a model of exercise based upon those three dimensions of self-efficacy as described by Bandura (1997 ; 2004). In their studies, E.S. Anderson, Wojcik, Winett, and Williams (200 6) and Rovniak, Anderson, Winett, and Stephens (2002) provided a mo re complex depiction of the SCT model of exercise that included social support a s well as interrelations between self-efficacy, outcome expectations, and goal sett ing.
29 Figure 4. Social Cognitive Model of Physical Activity. Social cognitive theory or self-efficacy have been used i n several structural equations models to explain physical activity ( E. S. Anderson et al., 2006; Dishman et al., 2005; McNeill, Wyrwich, Brownson Clark, & Kreuter, 2006; Motl, Dishman, Saunders, Dowda, & Pate, 2007; Motl et al., 2002; Resnick, 2001; Resnick & Nigg, 2003; Rovniak et al., 2002). Tabl e 2 summarizes the more recent structural equation models concerning self-efficacy a nd physical activity. A wide age range was selected purposefully because self-eff icacy changes as one matures (Bandura, Caprara, Barbaranelli, Gerbino, & Pastorelli, 2003). Adolescents go through an especially taxing phase in wh ich they have to deal with puberty changes, enlarged peer networks, and emoti onally invested partnerships (Bandura et al., 2003). Common througho ut most of these studies was that self-efficacy had large total effects on vario us modalities of exercise,
30 and often mediated the relationships between social sup port and other situational variables. Important for this research is that Rovniak et al. (2002) found the largest total effect on physical activity ( = .71, p < .001) among adolescents and younger adults. Following Table 2 is a literature rev iew of each of the key constructs utilized in the structural equation models.
31 Table 2 Role of Self-efficacy in Structural Equation Models for Exercise Author Independent Variables Outcome Statistic for SE Sample Theory Social support; social pressure, SE a (McNeill et al., 2006) Walking Total Effect on walking = 0.269, t = 6.74 Black vs. White adults age 18-65 N = 1090 SEM + SCT Social support; social pressure, SE a (McNeill et al., 2006) MPA Total Effect on Moderate Intensity activity = 0.353 Black vs. White adults age 18-65 N = 1090 SEM + SCT (McNeill et al., 2006) Social support; social pressure, SE a VPA Total Effect on vigorous intensity exercise = 0.443 Black vs. White adults age 18-65 N = 1090 SEM + SCT Note. SE = Self-efficacy; PA = physical activity; SEM = Socia l ecological model; SCT = Social cognitive theory; SET = Self-efficacy theo ry; HPM = Health Promotion Model; TTM = Transtheoretical model MPA = Moderate physical activity; VPA = vigorous physical activity
32 Table 2 (Continued) Author Independent Variables Outcome Statistic for SE Sample Theory (McNeill et al., 2006) Physical Environment; Intrinsic Motivation; SE Walking = 0.269, Black vs. White adults age 18-65 N = 1090 SEM + SCT (Rovniak et al., 2002) Social Support SE selfregulation & outcome expectations PA Physical activity Total effect on PA = 0.71 p < .001 Student Age 18-28 N = 353 (244 women) SCT (Resnick, 2001) Chronic illness Physical health SE PA Current Exercise R 2 = .24 of SE explained by illness & health Older adults N = 201 Note. SE = Self-efficacy; PA = physical activity; SEM = Socia l ecological model; SCT = Social cognitive theory; SET = Self-efficacy theo ry; HPM = Health Promotion Model; TTM = Transtheoretical model MPA = Moderate physical activity; VPA = vigorous physical activity
33 Table 2 (Continued) Author Independent Variables Outcome Statistic for SE Sample Theory (E.S. Anderson et al., 2006) Social Support SE Selfregulation & outcome expectations PA. Physical activity Total effect on PA = 0.12 p <.05 Church Age 18-92 ( M = 52.73, SD = 14.56) N = 999 SCT SE PA; SE intentions PA MPA Direct effect on MPA = 0.240 p <.0001 SE PA; (Motl et al., 202b) VPA Direct effect on VPA = 0.201 p <.0001 Teen Girls (M = 13.57, SD = 0.67) N = 1,797 SET Note. SE = Self-efficacy; PA = physical activity; SEM = Socia l ecological model; SCT = Social cognitive theory; SET = Self-efficacy theo ry; HPM = Health Promotion Model; TTM = Transtheoretical model MPA = Moderate physical activity; VPA = vigorous physical activity
34 Table 2 (Continued) Author Independent Variables Outcome Statistic for SE Sample Theory (Shin et al., 2005) Prior behavior perceived SE PA Commit to plan for exercise Total effect on planned exercise = 0.08 t = 8.40 p <.01 Adult with chronic disease (M = 53.57, SD = 13.9) N = 400 HPM SE Selfmanagement PA Direct effect 0.59 SE outcome expectations PA Direct effect 0.46 (Dishman et al., 2005) (Dishman et al., 2005) (Dishman et al., 2005) SE perceived barriers PA Physical activity Direct effect -.45 Teen females N = 309 6 th + 8 th graders (M age 11.5, SD .6)Â’ + N = 296 M 13.5, SD .6 TTM + SCT Note. SE = Self-efficacy; PA = physical activity; SEM = Socia l ecological model; SCT = Social cognitive theory; SET = Self-efficacy theo ry; HPM = Health Promotion Model; TTM = Transtheoretical model MPA = Moderate physical activity; VPA = vigorous physical activity
35 Table 2 (Continued) Author Independent Variables Outcome Statistic for SE Sample Theory SE enjoyment PA Direct effect 0.47 (Dishman et al., 2005) (Dishman et al., 2005) SE PA With PA r = .38, p < .05 Health & Social Support SE PA Direct effect .50 (Resnick & Nigg, 2003) (Resnick & Nigg, 2003) Health & Social Support Stages of change PA Exercise SE to stage = 0.42 & stage to PA = .26 Older adults (M age 86.1, SD 5.9) TTM + SCT Note. SE = Self-efficacy; PA = physical activity; SEM = Socia l ecological model; SCT = Social cognitive theory; SET = Self-efficacy theo ry; HPM = Health Promotion Model; TTM = Transtheoretical model MPA = Moderate physical activity; VPA = vigorous physical activity
36 PS-1: Self-efficacy for exercise. Self-efficacy expectation is the belief that one can successfully do the behavior required to produ ce outcomes (Resnick, 2005). A synonymous term is task self-efficacy, which is th e belief in oneÂ’s effectiveness in doing specific task (Zimmerman & Cleary, 2005). As noted in Table 2, task self-efficacy, or self-efficacy expectation, has been demonstrated by many authors to be the key construct in models for ph ysical activity outcomes. Moritz, Feltz, Fahrbach, and Mack (2000) conducted a me ta-analysis of relations of self-efficacy measures to sport performance The average correlation between self-efficacy and sport performance was significant ( r = .38, z = 25.80, p < .001). Moritz et al. also found that familiarity w ith performance tasks was associated with larger correlations to self-efficacy ( r = .36) compared to novel tasks ( r = .31); this supported BanduraÂ’s (1994) position that mastery experiences lead to higher self-efficacy. Using a prospective design, Rovniak et al. (2002) also t ested the SCT model of social support leading to self-efficacy, outcom e expectations, and selfregulation, which in turn lead to physical activity (de picted in Figure 3) among 283 undergraduate students. Self-efficacy had a strong total effect on physical activity ( direct/total = .71, p <.05). PS-2: Outcome expectations for exercise. Outcome expectancy is the expectation that a given course of action will produce certain outcomes as well as values for those outcomes (Bandura, 1994). These out comes are perceived as either risks or benefits. Physical outcomes include grati fying and aversive effects of the behavior and the associated losses and ben efits (Bandura, 2004).
37 Social outcomes involve approval or disapproval that the behavior elicits among peers or family. Personal outcomes involve oneÂ’s self-eva luated reactions to oneÂ’s health behavior and health status (Bandura, 2004 ). Using a prospective design, Rovniak et al. (2002) tested the SCT model of social support leading to self-efficacy, outcome expectati ons, and self-regulation, which lead to physical activity, among 283 undergraduat e students. Rovniak et al. did not find significant total effects of outcome expectations on physical activity, and expectations did not predict levels of phy sical activity. Similarly, E.S. Anderson et al. (2006) tested a revise d SCT model among 999 adults recruited from area churches (age range 18 Â– 92, M = 52.73, SD = 14.56 years). Positive outcome expectations had a negati ve direct effect on physical activity, and a small but positive indirect effe ct on physical activity, producing a non-significant total effect (E.S. Anderson et al.). However there have been multiple studies documenting t he evidence for exercise benefits, which are synonymous with positive outc ome expectations for exercise (Allison et al., 2005; C. Anderson, 2003; De B ourdeaudhuij & Sallis, 2002; Deforche, De Bourdeaudhuij, Tanghe, Hills, & De Bode, 2004; Enthoven, Skargren, Carstensen, & Oberg, 2006; Grubbs & Carter, 2002a; A. H. Harris, Cronkite, & Moos, 2006; Landers, 2006; McDevitt, Snyder Miller, & Wilbur, 2006; Nelson & Gordon-Larsen, 2006). Perceived benefits of exercise noted by adolescents include increased stamina and muscle strength, and improved muscle tone (Grubbs & Carter, 2002). Psychological benefi ts include decreased stress or anger, increased self-confidence, greater self-di scipline, and better
38 feeling (Allison et al., 2005). Of key interest for t his proposed model is that exercise has been shown to increase slow wave sleep and tot al sleep time, which are beneficial for replenishing the bodyÂ’s energy stores and offsetting fatigue (Landers, 2006). A recent analysis of the National Longitudinal Study o f Adolescent Health (Nelson & Gordon-Larsen, 2006) showed other benefits. A dolescents who exercised ( N = 11,957; M age 15.8, SD = 11.6 years) were less likely to have risky health behaviors such as having sex without birth control, smoking, drinking and driving drunk. Those who exercised 5 hours per week at moderately vigorous physical activity (MVPA) were less likely to have low self-esteem ( ARR = .83, CI = .80 .86), more likely to achieve grades of Â“AÂ” in the hard sciences, and were more likely to sleep greater than or equal t o eight hours per night (Nelson & Gordon-Larsen). In another study of adolescents (Deforche et al., 2004), the sample ( N = 90, mean age = 14.6, SD = .9 years) was categorized by weight status. MANOVA was used to analyze differences in attitudes tow ards exercise among the three weight groups. Perceived benefits that were statistically significant across the groups were pleasure ( F = 8.1, p < .001), which was higher among nonobese adolescents; looking better ( F = 3.2, p < .05), which was higher among obese adolescents; and losing weight ( F = 8.6, p < .001), which was higher among obese adolescents. Benefits that were viewed the same across all three groups were social contacts, competition, feeling better, and improving health and physical conditions (Deforche et al., 2004).
39 Anderson (2003) sampled collegiate women ( N = 397, mean age 23, SD = 6.99) to determine motives for exercise as well as reason s for quitting. Mental motives cited by those who met CDC guidelines for exer cise included centering (time to be alone, 32%), relieving tension and stress ( 75%), and improving mental performance (30%). Those who did not meet CD C guidelines for exercise cited the same benefits but had a reduced rate: center ing (20%), relieving tension and stress (61%), and improving mental perform ance (23%). A similar pattern was noted throughout the entire list of cited benefits with the exception of exercising because it was a school requirement (cited by 5% of the Â‘noÂ’ exercise group). De Bourdeaudhuij and Sallis (2002) investigated the c ontribution of perceived benefits in explaining variance in physical a ctivity of moderate to vigorous intensity among three age groups, one of which was age group 16 -25 (mean age 21, SD = 2.9). For females, perceived benefits showed R 2 of 3% in a regression analysis that ultimately explained 13% of th e variance in physical activity for females. Competition benefits for males ( = .14) and health benefits for females ( =.13) were the most significant benefits (De Bourdeaudh uij & Sallis). K.M. Harris et al. (2006) followed a cohort ( N = 424) of depressed adults across 10 years to examine factors that influence the natu ralistic course of depression. At intake the mean age was 39.9 ( SD = 14.1). The effect of exercise on global depression was a 2.24-point drop in depression for each increment of
40 physical activity (effect = -2.24, SE = .64, p < .001). This study showed that physical activity may help to reduce concurrent depression (K.M. Harris et al.). Similar findings were found in a qualitative analysis of 34 participants (16 men, 18 women, age range 18 Â– 50) with serious mental illnesses including schizophrenia, mood disorders, and anxiety disorders (McDe vitt et al., 2006). These participants were drawn from a larger sample ( N = 2,216) housed in two midwestern rehabilitation centers. Key themes that emer ged about exercise benefits were primarily mental health benefits (Â‘hap py feelingÂ” and Â“can sleep at nightÂ”). In contrast with common mental illness symptoms (anergia and anhedonia), physical activity was seen as a way to Â“becom e more involved with lifeÂ” (McDevitt et al., p. 53). Regular exercise can also affect level of disability (Ent hoven et al., 2006). Enthoven et al. queried patients with low back pain f ive years after their initial entry into an experimental design comparing chiropracti cs and physiotherapy for those with low back pain (Enthoven et al.). The main o utcome variable was disability. Logistic regression was used. Among other pred ictors, those who had lower exercise levels at baseline had more disability both at baseline and at the five year follow-up ( OR 3.35, 95% CI = 1.48 Â– 7.58, p < .01). One of the longer term benefits of exercise is a reduction in residual disab ility after having a backpain event (Enthoven et al.). In summary, the benefits of exercise range from mental benefits (feeling better, less fatigue, less depressed) to physical benefi ts (weight loss, less
41 debilitation after injuries). Outcome expectations for exercise include benefits, and will be used as the indicator. PS-3: Self-regulation. Self-efficacy beliefs also involve self-regulation. Self-regulation encompasses both goals and plans. The str onger the perceived self-efficacy, the higher the goals people use for thems elves (Bandura, 2004). Those goals may either be proximal ones or distal ones. An example of proximal goals is intentions (Bandura). Exercise self-regulation encompasses several skills, includin g planning, organizing, and managing oneÂ’s exercise activities. It i s important because motivation is not enough to sustain exercise behavior (R ovniak et al., 2002). As noted earlier, as women transition from adolescence thro ugh young adulthood, maintenance of physical activity becomes more and more di fficult. Using a prospective design, Rovniak et al. tested a structural equ ation model of selfefficacy, outcome expectations and self-regulation lead ing to physical activity among 283 undergraduate students. Self-regulation had a strong total effect on physical activity ( direct/total = .48, p <.05). Situational Factors S-1: Loneliness. Loneliness also has a significant impact on physical activity. Loneliness is defined as a continuum ranging fr om alienation to connectedness that is a pervasive, depressing, and debilita ting condition (Killeen, 1998). In a study among 1,297 adolescents ( N = 630 females, 654 males, mean age 15.3, SD = 2.9), Page and Tucker (1994) used loneliness as a dep endent measure with exercise frequency as an independent varia ble. Exercise frequency
42 was inversely associated with loneliness ( r = -.13, p < .001). Furthermore, MANCOVA testing showed significant differences among the exercise frequency groups relative to loneliness (WilksÂ’ Lambda (12, 2560) = 0.9657, p < .0001); those who exercised zero times per week had the highest least sq uares means for loneliness (mean least squares = 9.07), and those who exe rcised more frequently had lower loneliness scores (Page & Tucker). T hese authors offered a possible explanation for the findings that physical exer tion resulted in reduced levels of loneliness: Regular exercise in groups may foste r reduced loneliness, but biochemical mechanisms of the brain may also explain the findings, because regular physical activity increases levels of brain norepi nephrine and serotonin which promote feelings of well being (Page & Tucker). Storch et al. (2007) tested a cross sectional model of loneliness as a mediating variable between peer victimization and ph ysical activity among overweight youth ( N = 100, mean age 12.9 2.8). Using Baron and KennyÂ’s (1986) guidelines for mediation, loneliness met the cr iteria for being a mediator (Storch et al.). Loneliness exacerbated the difficulty that overweight youth had engaging in exercise. Mahon, Yarcheski, and Yarcheski (1998) tested a cross sectio nal model of loneliness as a mediating variable between perceive d social support and positive health practices, which included physical activity among young adults ( N =70, men = 42, women = 28; mean age 24.93; SD = 2.50; range 22-34). Using Baron and KennyÂ’s (1986) guidelines for mediation, l oneliness met the criteria for being a mediator. Mahon et al. (1998) also found an inverse relationship
43 between loneliness and positive health practices such as ex ercise ( r = -.54, p < .001). This same research group (Mahon, Yarcheski, & Yarcheski, 2 001) later studied a younger age population ( N = 127, mean age 12.9, SD = .63; 55 girls, 72 boys) and found through regression of positive health p ractices on loneliness that loneliness was significant ( B = -.27, p =.01). In a later study, Mahon, Yarcheski, and Yarcheski (2004) u sed a crosssectional, correlational design to test a model of lonel iness as a mediating variable in a younger age group as well ( N = 134, mean age 12.9, SD = .58; 70 girls, 64 boys). In this younger age group, loneliness w as a partial mediator between social support and positive health practices. Maho n et al. (2004) also found an inverse relationship between loneliness and po sitive health practices such as exercise ( r = -.50, p < .001). That finding indicated that although loneliness was a dominant mediator in the relationship of social support and positive health practices during young adulthood (Maho n et al., 1998), it was not during adolescence (Mahon et al., 2004). However even as a partial mediator, loneliness limited the extent to which the adolescents w ere motivated to carry out positive health practices such as exercise (Mahon et al., 2 004). Yarcheski, Mahon, Yarcheski, and Cannella (2004) did a m eta-analysis of 37 studies published since 1983 pertaining to predictors o f exercise. Yarcheski et al. found that loneliness had the largest effect size ( = -0.48) as a predictor of positive health practices across all the studies. These stud ies were among healthy adolescents and young adults.
44 In summary, loneliness has been found to be a significa nt predictor of exercise, a mediating variable affecting exercise outcomes, as well as an outcome variable affected by exercise. S-2: Social support. Bandura (1997) discussed the coaching influences on development and maintenance of self-efficacy as a key m eans of social support. Â“The task of developing resilient self-efficacy in athlet es rests on the managerial efficacy of coaches (p. 397)Â”. Effective coaching support includes carefully graded mastery experiences with gradually increasing pre ssure situations. At the same time, the effective coaches avoid placing players pr ematurely in situations that are set-ups for premature failure; precipitous rem oval of the athlete when he/she gets into trouble only undermines their sense of efficacy. Bandura (1997) also noted that perceived social pressure to become more p hysically active accounted for exercise involvement although at a lesser l evel than self-efficacy beliefs, expected benefits, and satisfaction with goals a chieved. Several authors examined the role of social support ( Allen, Markovitz, Jacobs, & Knox, 2001; Barrera, Toobert, Angell, Glasgow & Mackinnon, 2006; Callaghan, 2006; Cerin, Taylor, Leslie, & Owen, 2006 ; Marquez & McAuley, 2006; McNeill et al., 2006; Motl et al., 2007; Ward et al., 2006). Motl et al. (2007) examined the cross-sectional relationships of environment al factors, social support, and self-efficacy on exercise in 12 th grade girls ( N = 1,655; M age = 17.7 years, SD = .06). They targeted this particular grade level be cause they had found a sharp decrease in physical activity by the time g irls reach 12 th grade. Motl et al. (2007) specified three social functions to m easure social support: a)
45 guidance, b) nurturance, and c) reassurance of worth. Social support had a direct effect on physical activity ( = .28) as well as an indirect effect through self-efficacy. Targeting perceived social support among a dolescent girls was a useful means of indirectly and directly increasing physical activity (Motl et al.). McNeill et al. (2006) measured two aspects of social supp ort in their structural equation model of exercise: a) emotional sup port and b) informational support. Both of these factors had indirect effects on wa lking outcomes that were mediated by motivation and self-efficacy. However, cont rary to their hypothesis, the association between social support and self-efficacy w as not significant. Allen et al. (2001) specifically targeted hostile pe rsons in their analysis of coronary artery risk development in young adults (CARDI A) study data. Allen et al. found gender and racial differences in the effects of social support on physical activity outcomes. Hostile Black women exercised significant ly less than other subgroups, even in the presence of high social support. I n contrast hostile White women with high social support exercised significantly mo re ( p = .02) than those with low social support, as did men of both races. The ag e range was 18 to 30, N = 5,115 ( n = 2,287 women). Allen et al. noted that there was a ccumulating evidence for the protective effect of social support on exercise, despite the racial differences in women. Similar protective effects of social support on exercise as well as on other health care behaviors were found by Callaghan (2006). Of 254 participants (ages 14 Â– 19), the mean score on exercise was significantly di fferent between those with high social support versus low social support ( t = 4.10, p < .001).
46 Cerin et al. (2006) used a technique to test for med iational analyses that was described by MacKinnon, Krull, and Lockwood (2000). In a small randomized controlled trial ( N = 52 with 48 women, age 45 to 78), Cerin et al. found that social support was a mediator of walking bot h immediately after an intervention and at four weeks later (MacKinnon et al. test = 1.144, p = .020). Social support has also been found to mediate the eff ects of other interventions. Barrera et al. (2006) studied 279 wome n with type 2 diabetes by using an intervention that emphasized cohesion among t he participants and the mobilization of social resources to change lifestyle beha viors. Barrera et al. found that social embeddedness mediated the effect of l ifestyle intervention on physical activity. Social support also has been shown to correlate well wit h physical activity in different ethnic groups. Marquez and McAuley (2006) found that among Latino adults ( N = 153, M age = 29.4, females n = 86) social support from friends correlated significantly with the physical activity classifi cation ( r = .20, p < .05). However social support from the family was not a signifi cant correlation. In contrast, Ward et al. (2006) studied physical activi ty correlates in adolescent girls ( N = 1162, M age 14.6, 45% African American) and found that physical activity status (active versus inactive) was significa ntly associated with family support. However among African American girls, i t was true only for the active girls who were of normal weight status, versus any activity level for White girls (Ward et al.). Ward et al. stated that family support was relevant to all
47 adolescent girls, irrespective of weight status, and that interventions should focus on social-cognitive variables unique to different races a nd activity levels. However social support has not always been a consistent fa ctor or predictor of physical activity. Von Ah, Ebert, Ngamvitr o, Park, and Kang (2004) studied a convenience sample of 161 college students ( M age 19.7 4.09 years) to determine predictors of health behaviors including p hysical activity. They assessed social support by two methods: a) assessment of the n umber of available others, and b) assessment of satisfaction with perceived social support. Neither of these two indicators had a significant impact on health behaviors, which the authors attributed to measurement issues (Von Ah et al.). In summary, social support was often included in structural equation models for physical activity. Some of these studies measure d both the indirect effects of social support on exercise, through self-effica cy, as well as the direct effects of social support on exercise, while others measur ed only the indirect effects of social support on exercise through self-effica cy. Other studies found evidence that social support also functioned as a mediato r in models between interventions and physical activity. And finally, social support has been demonstrated to correlate with physical activity in a wi de range of age groups and ethnic groups. However, it has not always been a consi stent determinant of health-related behaviors.
48 Physiological Latent Variables Physiological factors are the antecedents that are often reflected in, and diagnosed by, the presence of unpleasant symptoms (Lenz et al., 1997). This concept includes normal bodily function and the individu alÂ’s level of energy (Lenz et al. 1997). The indicators selected for this concept a re perceived exercise capacity, anticipated fatigue from exercise, and perceive d health status. PH-1: Exercise capacity. Exercise capacity is a clinical measurement of maximal oxygen uptake. Wisen, Farazdaghi, and Wohlf art (2002) developed a scale that allows patients to select the most strenuous act ivity that they could sustain for 30 minutes, with corresponding metabolic equi valents (METs). By definition, one MET is the measurement of resting oxyg en uptake (VO 2 ) with the patient in a sitting position; a higher level of activ ity uses up a higher amount of oxygen. Wisen et al. demonstrated that healthy women (age 21-79) were able to accurately predict their maximal MET level as confirmed by ramp testing. The MET level can be converted to VO 2 by the use of an age-adjusted formula (Wisen et al.). Being able to accurately predict VO 2 from patientsÂ’ self-report of their perception of exercise capacity is valuable. This se lf-perception of exercise capacity is theorized by this PI to be a perception that positively impacts the clientÂ’s decision to exercise. Functional exercise capacity or physical fitness can be measur ed in other ways as well. Researchers from the National Heart, Lung, and Blood Institute WomenÂ’s Ischemia Syndrome Evaluation enrolled women ( N = 936) in a prospective multicenter cohort study (Wessel et al., 2004) They used the Duke
49 (DASI) self-report measure of functional capacity where women reported their ability to perform various exercise activities; these wer e used to estimate METs. The exercise capacity scores significantly differentiated be tween the low fitness women (N =631, DASI < 25) and the high fitness women ( N = 275, DASI > 25). The DASI functional capacity score was inversely related w ith serum levels of high sensitivity C-reactive protein ([hs CRP], r = -.19, p < .001) and IL-6 ( r = .14, p < .001). The DASI functional capacity scores remained sig nificant independent predictors of adverse events including mort ality (Wessel et al., 2004). Perception of functional capacity also affects other life events. Patients ( N = 545) enrolled in a multicenter comparison of drug effe cts on functional capacity were asked to rate their perceived health perceptions o n a visual analogue scale (VAS) of 10 cm with anchors of 0 on the left, correspo nding to death and 10 on the right, corresponding to perfect health (Havranek et al., 2001). These researchers defined perceived health as being determined by a high level of physical functional capacity as well as a low level of emo tional distress. Perceived functional capacity predicted cardiac events i n patients with cardiac failure ( RR with each VAS decile = .74, p = .001, 95% CI .61-.88), and predicted cardiac events more than did exercise treadmill time (Ha vranek et al.). In summary, self-report measures of exercise capacity have been found to be reliable estimates of actual function as measured via treadmill testing, and as strong predictors of cardiac events.
50 PH-2: Perceived health status. Perceived health status often is envisioned as being synonymous with quality of life (QO L). However in a metaanalysis, Smith, Avis, and Assmann (1999) examined 12 QO L studies to determine if QOL is a different construct from health st atus. The authors then used structural equation modeling to test a model of determinants of QOL that included biologic/physiologic status as the exogenous varia ble leading to symptom severity and through to quality of life. They determined that from the patientÂ’s perspective, QOL and health status are two d ifferent constructs (Smith et al.). The continuum for health states ranging from death to optimal functioning corresponds more closely to perceived health than it doe s for QOL. Quality of life focuses more on psychological functioning than physical he alth status (Smith et al.). Therefore, perceived health status is defined bei ng part of the physical concept of perceived health. Perceived health status is frequently measured when study ing health disparities. Researchers from Tennessee (Ahmed et al., 20 05) used the national Health Interview Survey data from 1999 to 2000 ( N = 23,459 men) to examine health disparities using logistic regression. Those who per ceived better health status had an increased likelihood of engaging in leisur e time physical activity; however racial/ethnic disparities were noted even after accounting for sociodemographic characteristics. Recently Chen, James, and Wang (2007) compared the he alth promotion practices across two cultures: Taiwanese ( N = 265) adolescents and American ( N = 285) adolescents from San Diego (age range 12 to 15). The researchers used
51 the Adolescent Health Promotion (AHP) scale based upon PenderÂ’s model of health promotion and OremÂ’s self-care deficit theory, w hich has 40 items with six dimensions including exercise behavior. In general, the American adolescents had better perceived health status and total AHP scores ( 2 = 10.6, p < .01) than the Taiwanese adolescents, indicating cultural dispariti es still exist. Perceived health status has a medium effect on exercise o utcomes; in their meta-analysis of predictors of positive health pra ctices, Yarcheski et al. (2004) noted that predictors of positive health practice s included perceived health status ( = .37). Using the physical component of the Medical Outcomes Stud y Short Form-36 ([SF-36], Ware & Sherbourne, 1992) Finnish r esearchers (Leino-Arjas, Solovieva, Riihimaki, Kirjonen, & Telama, 2004) foll owed a cohort of 902 industrial employees (mean age 34.6 at baseline) for 2 8 years to analyze trends in physical activity and perceived health status. Those w ho engaged in vigorous physical activity at baseline and at the 5 year follow -up had a decreased risk of poor physical functioning (age-adjusted OR = 0.34, 95% CI = .22 .53). Those who reported vigorous physical activity at either of th e time points (but not both) had a decreased risk of poor physical functioning as well although not as much of a decreased risk (age-adjusted OR = .57, 95% CI = .33 .98). Another interesting finding was that while total vigorous leisu re physical activity did not vary between white-collar and blue-collar workers, blue collar workers with only moderate leisure physical activity fared well on the SF -36 scores, possibly
52 indicating the protective effect of their on-the-job p hysical labor (Leino-Arjas et al., 2004). Similar findings among much older adults ( N = 316, mean age 69, SD 4.12) were found by Lee and Laffrey (2006). They te sted a theoretical model that included three constructs as predictors of physical activity (i ndividual characteristics, interpersonal influence, and environment ). Individual characteristics included oneÂ’s cognitive appraisal of perceiv ed health status, which was queried by a single item Â“how would you rate your overall health at this time?Â” Scores ranged from one to four, with four meaning greater perceived health. They found that perceived health status influe nced physical activity indirectly ( = .032, p <.01) such that those with greater perceived health status had fewer barriers to physical activity (Lee & Laffrey) PH-3: Anticipated fatigue from exercise. Anticipated fatigue as a result of exercise can be a barrier to exercise participation and o ften occurs in healthy individuals. Fatigue is not necessarily a symptom of dise ase (C. M. Yang & Wu, 2005). Among healthy college age students in Florida, ( N = 147, ages 18 24), the statements Â“exercise tires meÂ” and Â“I am fatigued by exerciseÂ” were rated as the first and third top barriers to exercise (Grubbs & C arter, 2002). C. Anderson (2003) sampled collegiate women ( N = 397, mean age 23, SD = 6.99) to determine motives for exercise as well as rea sons for quitting. Anderson found that fatigue ranked third as the prima ry reason for quitting; 17% of those who met CDC guidelines for exercise ( N = 174), and 26% of those who did not meet the CDC guidelines (N =217) cited fatigu e as a reason for quitting.
53 In summary, fatigue has been cited as the primary barr ier to exercise as well as a reason for quitting. Fatigue can be exacerbat ed by exercise even without causing significant functional impairment (C. M Yang & Wu, 2005). Anticipated fatigue is theorized to be a perception of physiological status that will impact the physical activity, mediated by existing fatig ue as an unpleasant symptom. Unpleasant Symptoms US-1: Chronic fatigue. Ream and Richardson (1996) defined fatigue: Â“Fatigue is a subjective unpleasant symptom which incorpor ates total body feelings ranging from tiredness to exhaustion creating a n unrelenting overall condition which interferes with individualsÂ’ ability to function to their normal capacityÂ” (p. 527). Fatigue is a significant problem fo r adolescents, and can be attributed to medical or psychiatric disorders, syndromes o f unknown etiology, and lifestyle choices such as exercise (Mears, Taylor, Jorda n, Binns, & Pediatric Practice Research, 2004). To study characteristics of fatigue among adolescents, Mear s et al. (2004) collected data for a one-year period on adolescents visit ing a primary care clinic. They determined the prevalence of chronic fatigue synd rome like illness (4.4%) and of prolonged fatigue of greater than one month (8%). Symptom predictors of prolonged fatigue included the adolescentsÂ’ reporting t hat exercise worsened their fatigue; among the fatigued group, exercise wor sened fatigue in 38.2%, and among the not fatigued group, exercise worsened fatigu e 10.5% in (Mears et al.).
54 Using data that were obtained from a sub-sample of the United States National Longitudinal Study of Adolescent Health, Rhe e, Miles, Halpern, and Holditch-Davis (2005) interviewed 20,745 adolescents abo ut 10 symptoms, and asked them to rate the frequency of having experienced t he symptoms during the past 12 months. Over 20% ( N = 3,962) reported having experienced fatigue and fatigue was the third most prevalent symptoms. Fatigue also was associated with other symptoms. The definition of fatigue used in this study was Â‘Â”tiredness with no reasonÂ”. Striking gender differences were noted: 1 5.96% ( N = 1,495) of boys and 25.38% ( N = 2,467) of girls reported fatigue ( OR = 1.79, 95%, CI = 1.621.98; Rhee et al., 2005). Another interesting findin g was that the probability of recurrent fatigue increased in a linear fashion with ea ch increase in year of age. However in the same study reported elsewhere (Rhee, 2005) no significant differences occurred between racial groups when reporti ng prevalence of fatigue. Other authors studied the prevalence rates of fatigue among healthy adolescents ( N = 3,467; 1,718 boys and 1,749 girls, mean age 14.7, SD 1.4) from the Netherlands (ter Wolbeek, van Doornen, Kavel aars, & Heijnen, 2006). These researchers found the prevalence rates for fatigue among the girls was 20.5% and among the boys was 6.5% ( p < .001). Of those who reported fatigue, fatigue lasting for 1 month was reported by 80.0% of the girls and 61.5% of the boys ( 2 = 17.80; p < .001). In contrast to the study by Mears et al. (20 04), ter Wolbeek et al. (2006) found that exercise was not a sig nificant predictor of fatigue. Instead, ter Wolbeek et al. found that a de creased participation in sports
55 was related to fatigue in both girls ( t = 6.80, SD = 4.17, p <.001) and boys ( t = 7.76, SD = 4.48, p < .001). As shown, fatigue is a common unpleasant symptom among healthy adolescents. The impact of fatigue on exercise outcomes h as been studied as well (C. Anderson, 2003; Grubbs & Carter, 2002; Y. H. Kim, 2006). In a 10-year longitudinal study of a large biracial cohort of girls, Y.H. Kim et al. reported that fatigue (Â“IÂ’m too tiredÂ”) was the second most frequent ly cited barrier to exercise. These results were obtained in a multicenter prospective study of obesity development in 2,379 girls who were followed annuall y from ages 9 or 10 to ages 18 or 19 (Y.H. Kim et al.). There were no significant differences in the amounts of sleep obtained; the fatigued girls averaged 8.3 hours per night of sleep, and the nonfatigued girls averaged 8.6 hours per night ( p = .77 .85). US-2: Chronic pain. Pain is a limiting factor to exercise as well. Melzack (2001) defined pain as a multidimensional experience p roduced by multiple influences which include genetic and sensory influences, and modulated by psychological stress and other cognitive events. Melzack posi ted that a neuromatrix translates cognitive, sensory, and affective inputs into outputs such as pain perception and stress signals. Thus cognitive, senso ry, and affective beliefs all contribute to the perception of pain. Fear of movement or reinjury among patients with m uscular skeletal injuries can lead to longstanding pain or disability (C ook, Brawer, & Vowles, 2006). After a painful experience has occurred, some pe ople catastrophize the experience, which perpetuates fear, avoidance, and disu se (Lethem, Slade,
56 Troup, & Bentley, 1983; Slade, Troup, Lethem, & Ben tley, 1983). However the prevalence of pain among healthy adolescents was not fu lly documented until recently. Using data that were obtained from a sub-sample of the National Longitudinal Study of Adolescent Health, (Rhee et al ., 2005) interviewed 20,745 adolescents about 10 symptoms and asked them to rate th e frequency of having experienced the symptoms during the past 12 months. Ove r 28% ( N = 5,301) reported having experienced headaches; over 27% ( N = 5,038) reported having experienced musculoskeletal pain; and over 17% ( N = 3,331) reported having experienced stomachaches. All of these were commonly associa ted with fatigue as well as with other symptoms. Striking gender differ ences were noted for all three symptoms: 20.73% ( N = 1,801) of boys and 37.43% ( N = 2,236) of girls reported headaches ( OR = 2.29, 95%, CI = 2.06-2.54; Rhee et al.). Similar findings were noted for musculoskeletal pain. The proba bility of recurrent musculoskeletal pain increased in a curvilinear/quadratic f ashion; the pain peaked at ages 16 to 17 and decreased to age 22. The impact of pain on exercise outcomes has been document ed by several researchers (Allison, Dwyer, & Makin, 1999; C. An derson, 2003; Bigal, Liberman, & Lipton, 2006; Gyurcsik et al., 2004; Parks, Housemann, & Brownson, 2003; Poulton, Trevena, Reeder, & Richard, 2002). Bigal et al. (2006) studied the influence of baseline weight status on the prevalence, severity, and disability of migraines. The sample consisted of 30,215 p articipants of ages 18 to 89 ( M = 38.7), of whom 45% were overweight, obese, or mo rbidly obese (Bigal et
57 al.). Among those who were morbidly obese, physical acti vities exacerbated the pain more than for the normal weighted ( OR = 1.7, CI = 1.2 2.2). Parks et al. (2003) queried 1,818 adults to study barri ers to exercise across different settings (urban, suburban, or rural) and two incomes (lower or higher). Those who were urban, lower income reported being afraid of injury as a barrier to exercise significantly more than the others ( 2 = 17.80, p < .005). These findings by Parks et al. (2003) have been corrobo rated by other researchers using younger adolescents. Allison et al. (19 99) used a two-stage cluster sample of 1,041 high school students (9 th and 11 th graders) to study perceived barriers to exercise across three settings: a) ph ysical education classes, b) sports at school, and c) non-school sponsored recr eational sports. Discomfort and injury both emerged as perceived barrier s and both items loaded onto the same factor in a principal components analysis of the perceived barrier items (Allison et al., 1999). Gyurcsik et al. (2004) examined barriers to vigorous phy sical activity among 132 students ( M age = 17.84, SD = .46 years) in their freshman year at a university in Alberta. Eighteen of the students ident ified injury as a barrier to exercise in the intrapersonal barriers domain. Anderson (2003) sampled collegiate women ( N = 397, M age 23, SD = 6.99) to determine motives for exercise as well as reason s for quitting. Of those who met CDC guidelines for exercising, 9% cited a medi cal/injury/physical condition or symptom as a reason to quit exercising, wher eas among those who
58 did not meet CDC guidelines for exercising, 13% cited a medical/injury/physical condition or symptom as a reason to quit exercising. Poulton et al. (2002) followed a birth cohort of p articipants to age 26 ( N = 980, 499 males) and assessed them regularly for physical a ctivity in New Zealand. Some study members began declining the sub-max imal exercise bike test because they feared discomfort. Therefore the resea rchers added questions about Â“How much discomfort do you anticipateÂ” and then Â“How much discomfort did you actually experienceÂ” during the bike test. The researchers then separated the participants into under-predictor, accurate predictor and overpredictor groups. A 3 (group level) X 2 (gender) ANOVA was done for each physical health measure (Poulton et al.). Those in the over predictio n group had worse physical health, had higher BMI, and lower VO 2 max scores (Poulton et al.). Thus fear of discomfort can have devastating effects even in the mid-t wenties age group. In summary, pain often deters persons from exercising due to fear of injury, discomfort, or more pain. Research as shown that healthy adolescents and young adult s can experience negative symptoms such as fatigue or pain and yet a gap in the literature still exists for the knowledge about negati ve symptomatology related specifically to exercise among healthy students. Three fact ors have been posited to affect oneÂ’s predisposition to, or manifestation of, unpleasant symptoms: a) psychological, b) situational, and c) physiological. The r eactions to the unpleasant symptoms are theorized to mediate the relat ionship between the antecedent factors and physical activity as the outcome.
59 Activity E-1: Exercise. Exercise is defined as an activity for developing the mi nd or the body (U.S. Department of Health and Human Servi ces, 2002). For the purposes of this model, exercise is working the muscles to d evelop cardiovascular fitness by increasing the bodyÂ’s maximum cap acity to consume oxygen (Noakes, 2000). Exercise is also working the muscles t o obtain mental health benefits. Although there is a definite semant ic difference between exercise and physical activity, both are used interchangeably in this dissertation. Exercise can be whole body or can be of isolated muscles. E xperienced cyclists similar in age ( M age 28.5), height and weight, years of cycling experience (5 3) and forced vital capacity ( M = 5144 888) were randomized to respiratory muscle endurance training or control/plac ebo groups (Holm, Sattler, & Fregosi, 2004). After training, the exper imental group showed a significant increase in pulmonary ventilation rate afte r training, and no improvement was seen in the control/placebo group. The training group also had a significant increase in VO 2 ( p < .027). In summary, chapter 3 summarized the literature review including the theoretical background to the study, the preliminary stu dies leading up to the choice of the theoretical model, and the key factors tha t are used in the model of exercise. Physical indicators include anticipated capacity to exercise, health status, and anticipated exercise fatigue. Psychological ind icators include exercise self-efficacy, anticipated exercise outcomes, and self-reg ulation. The unpleasant symptoms include chronic fatigue and pain. The unpleasa nt symptoms of fatigue
60 and pain are evident in the lives of health adolescent s and young adults. However little is known about how all these variables i ntertwine, and whether or not the psychological, situation, or physical factors are m ediated by the unpleasant symptoms. In the following chapter, the desig n and methods are discussed in depth, including a description of each of the key indicators used for the variables of interest.
61 CHAPTER THREE Design and Methods Overview of Research Design A non-experimental, cross sectional design was used with da ta collected from a sample of 463 adolescent and young adult women attending the University of South Florida (USF). An Internet survey approach using DillmanÂ’s (2007) tailored design recruitment method was used to c ollect study design variables. Threats to validity were minimized by using established reliable and valid instruments to assess the study variables and by usin g a computer random generator (SPSS) to select those to invite from among all the eligible participants. Sample Description and Selection Sampling frame. The sampling frame used in this study consisted of a listing of female USF students between ages 18 and 25 obtained from the office of the registrar (University of South Florida, 2006a) This age range was chosen as a target because it is the time of transition into the age bracket where most weight gain occurs (National Heart Lung and Blood Insti tute working group, 2006). Sample size. Calculations were undertaken to determine the required number of responses for analysis to test the proposed th eoretical model using structural equation modeling (SEM). The proposed structu ral model consisted of
62 32 parameter estimates and 59 degrees of freedom. Using the power calculations proposed by MacCallum, Browne, and Saguwar a (1996), a minimum sample size of 187 was needed to achieve a power of .80 with 60 degrees of freedom. Since Marsh, Balla, and McDonald (1988) suggest ed that parameter estimates are unstable in samples of less than 200, the g uidelines of Bentler and Chou (1987), which were a ratio of 5:1 or 10:1 respon ses to estimated parameters, were applied. The optimal sample size using these ratios was 160 to 320. Therefore 320 were selected initially as the sampl e for this study. However, the sample size was double checked by another method. MacCallum, Widaman, Zhang, and Hong (1999) conducted a Monte Carlo study using 100 data sets to generate a matrix of rati o of variables to factors and communality level by sample sizes. Highest communality levels were obtained with 20:3 ratios of variables to factors which remained constant at 100% across all levels of the sample sizes. At 10:3 ratios of variab les to factors, the communalities of the studies did not reach 95% (Â‘goodÂ’) until the sample sizes exceeded 200. At higher ratios, wide and high communal ities were obtained with smaller sample sizes of 60 to 100. For this analysis, ther e was a ratio of 13:5, which is approximately comparable to a ratio of 8:3. A ccording to the matrix given by MacCallum et al. (1999), at 10:3 ratios a sample si ze of at least 400 was needed to reach good communality (defined as being in the .92 to .98 range). Because this study did not reach the necessary ratio of 1 0:3, a sample size of greater than 400 was thought to be needed, and 500 w ere sought.
63 The registrarÂ’s list of age-eligible female students co ntained seventeen thousand names; therefore, the population at USF was more than adequate to meet the sampling size (See Table 3). It was anticipated that the racial/ethnic distribution of responses would closely correspond to the distribution of USF students, as indicated by the data in Table 3 from USF (University of South Florida, 2006b).
64 Table 3 Diversity Profile of all USF Students Undergraduate Graduate Total # enrolled % enrolled # enrolled % enrolled # enrolled % enrolled Total 19,931 100.0% 5,473 100.0% 27,263 100.0% Race/Ethnicity African Am 2,877 14.4% 405 7.4% 3,460 12.7% Hispanic 2,295 11.5% 466 8.5% 2,890 10.6% Asian 1,256 6.3% 232 4.2% 1,612 5.9% Am Indian 75 0.4% 22 0.4% 103 0.4% Alien 477 2.4% 526 9.6% 1,047 3.8% White 12,479 62.6% 3,755 68.6% 17,550 64.4% Not reported 41 0.2% 25 0.5% 89 0.3% Gender Male 7,836 39.3% 1,905 34.8% 10,318 37.8% Female 12,054 60.5% 3,543 64.7% 16,856 61.8% Not reported 41 0.2% 25 0.5% 89 0.3% Participants Participants were female students recruited via email a t the University of South Florida during the spring 2007 semester. Inclusio n criteria for the study were the following: a) female and b) between the ag es of 18 and 25. All the
65 invited studentsÂ’ email addresses were placed into a lot tery for two separate cash prizes of $100 each. No student was paid or given extra credit for participating. Procedures Following institutional review board (IRB) review an d approval, the survey instruments were entered into an Internet-based softwa re program called Ultimate Survey (Prezza Technologies, 2007). This prog ram is designed to send out invitations to a list of email addresses and to provide the recipient of the email with a link to the online survey. The sample wa s randomly selected from the electronic file of all 17,000 eligible female stud ents of ages 18 to 25 using SPSSÂ’ random selection syntax. Email addresses from thi s selection process were transferred to Ultimate Survey, which was capable of tracking responses and deleting respondentsÂ’ email addresses from the invi tation list whenever subsequent reminders were sent. A demographic question con firmed the age and asked the participant not to continue if they were out of the stated age range of 18 to 25. Data Collection The elements of DillmanÂ’s (2007) total design method, revised for email/ Internet surveys, guided the data collection process. Pot ential participants could receive a maximum of four email contacts; the second con tact was four days after the first, and the third and fourth contacts wou ld follow in 5-day increments. Data collection spanned two weeks in spring 2007 (See Fi gure 5). However due to upcoming scheduling constraints (midterm exams) for man y students, the fourth contact was eliminated.
66 Figure 5. Data Collection Process. Participants were able to complete the online questionn aire on a computer in any location that afforded them access to the Interne t. However an occasional student reported problems opening the link from their home computer, which was resolved by amending their firewall. The email addre sses were all campus emails; however many students had their campus emails fo rwarded to an offsite email system, which resulted in a number of undeliverab le emails. Questionnaire items were not randomized due to constra ints of the Ultimate Survey system. The order of the questionnair es was as follows: The demographic profile, the International Physical Activit y Questionnaire (IPAQ), Exercise Self Efficacy Scale, the Rating of Perceived Ca pacity scale, the Social Support for Exercise Scale, the Situational Fatigue Sca le, the West Haven-Yale Multidimensional Scale, the Exercise Goals Scale, the Out come Expectations for
67 Exercise Scale, the Multidimensional Scale of Perceived S ocial Support, the UCLA-8 Loneliness Scale, the SF-12, and the Chalder Chronic Fatigue Scale. To assess the extent to which the participants were attend ing thoughtfully rather than responding randomly, four items were created and interspersed randomly throughout the questionnaire. This strategy provided validity to some of the questions (for example, the question Â‘Are you a lonely person?Â” was inserted as a validity check for the UCLA-8 Loneliness Scale). It a lso allowed for easier identification of respondents who were not attentive so that they could be excluded from analyses. After all data were collected, data were exported fro m the Ultimate Survey to SPSS version 11.4 (2002) on a dedicated compute r. Data transfers were completed in one bulk export. Measures Measures for the Physiological Latent Variable PS-1: Self-efficacy for exercise. Self-efficacy for exercise was operationally defined as the confidence that one has to exercise when other things get in the way. The empirical indicator for thi s was the Exercise SelfEfficacy Scale created by Shin, Jang, and Pender (2001) for adults with chronic diseases. It was chosen because it included both pain and fatigue situations according to specifications given by Bandura for rating exercise self-efficacy. It is an instrument with three factors (situational/interpe rsonal, competing demands, and internal feelings) with a standardized CronbachÂ’s co efficient of .94. These three factors explained 96.4% of the variance. The par ticipants rated their
68 confidence to exercise regularly three times per week und er a given situation using a percentage scale from 1% (can not do it) to 10 0% (certainly can do it). Exercise self-efficacy was an indicator for the latent va riable PSYCHOLOGICAL with the label PS1 used in the figures. The scale is pr ovided in Appendix A. PS-2: Outcome expectations for exercise. An outcome expectation for exercise was operationally defined as the belief that o ne can do the behavior required to produce the outcomes of physical activity. T he empirical indicator for this variable was the Outcome Expectations for Exercise Sca le-2 (Resnick, 2005). It is a 13-item scale that has two subscales: Posit ive outcome expectations and negative expectations that are scored s eparately, with the negative expectations being reverse scored. Confirmator y factor analysis showed a fair fit to the data ( 2 = 167.3, df = 64, p < .05; RMSEA = .08). Alpha coefficients of the two subscales were .93 and .80 respecti vely. The Outcome Expectations Scale Â– 2 explained 66% of the variance i n outcome expectations. The Outcome Expectations Scale Â– 2 is a revision from the first Outcome Expectations Scale, which included only the positive exp ectations. The negative expectations were added specifically to capture the outc omes of fatigue or pain expected to result from exercise. The Outcome Expectatio ns Scale was an indicator for the latent variable PSYCH and was label ed as PS2 in the figures. The scale is provided in Appendix B. PS-3: Self-regulation for exercise Exercise goals were operationally defined as the setting of goals in advance, self-monito ring, and problem solving, which are part of self-regulation (Rovniak et al., 200 2) The empirical indicator for
69 this was the Exercise Goals Scale (Rovniak et al., 2002). The scale exhibited good internal consistency ( = .89) and test-retest reliability ( t test-retest = .87). The Exercise Goals Scale was an indicator for the latent vari able PSYCH and was labeled as PS3 in the figures. The scale is provided in Appendix C. Measures for the Situational Factor S-1: Loneliness. Loneliness was operationally defined as the feeling of being alone even in the midst of others. The empirica l indicator for this variable was the UCLA-8 Loneliness Scale which is a revision from the original UCLA-20 and the UCLA-4 (Hays & DiMatteo, 1987; Russell, Peplau & Ferguson, 1978). When tested among college students ( M = age 21, range 17-48, SD = 4.5), it had an overall coefficient of 0.8996 and the standardized item of 0.90 (Hartshorne, 1993). Mahon, Yarcheski, T, and Yarcheski, A. (1995) validated the use of the scale among adolescents ages 12 to 21. Stateme nts in the questionnaire are evaluated on a 4-point Likert scale f rom strongly disagree (1) to strongly agree (4). Positively worded items are rev erse scored to negatives, so that for each item a high score (4) indicates the loneli est (Hartshorne). According to the recommendation of Hartshorne, one item that wa s problematic (item 17: Â“I am unhappy being so withdrawnÂ”) was revised to read Â“I am unhappy and withdrawnÂ”. Raw scores were transformed into a 0-100 sca le (Mahon et al.). Normative measurements revealed that in the United St ates, the mean score for the UCLA-8 Loneliness Scale after transformat ion to a 0 Â– 100 scale was 35.4 ( SD 19.2, range 0 Â– 100) reported by Hays and DiMatteo (1987). However in the pilot study for this research (Cobb, 20 06), after transformation the
70 mean was 50.6 ( SD 7.69, range 40 to 71). The UCLA-8 Loneliness Scale wa s an indicator for the latent variable SITUATIONAL and wa s labeled S1 in the figures. The scale is provided in Appendix D. S-2: Social support for exercise. Social support for exercise was operationally defined as the support received for par ticipating in regular physical activity from the people closest to you. The empirical indicator for this was the Social Support for Exercise Scale (Reis & Sallis, 2005; S allis, Grossman, Pinski, Patterson, & Nader, 1987). There are two subscales, each with the same questions but referencing different sources of social suppor t. Each subscale has 13 items. Scores were computed by summing the responses fo r each scale. The CronbachÂ’s coefficients ranged from 0.81 to 0.87 for the friend scale (Reis & Sallis). Courneya, Plotnikoff, Holz and Birkett (2001) used the same questionnaire but changed it to a single item Â“How much support do you receive for participating in regular physical activity from the people closest to you?Â” rather than asking the same series of questions with references fi rst to friends and then to family. A combination of the two approaches was use d, with all 13 items from one subscale asked in reference to Â‘the people closest to youÂ’. The Social Support for Exercise Scale was an indicator for the late nt variable SITUATIONAL and was labeled as S2 in the figures. The scale is provid ed in Appendix E. S-3: General social support. General social support was operationally defined as an exchange of resources between at least t wo individuals intended to enhance the well being of the recipient. The empirical indicator of this was the Multidimensional Scale of Perceived Social Support (Zim et, Dahlem, Zimet, &
71 Farley, 1988). This instrument specifically addressed the subjective assessment of social support adequacy from three specific sources: fam ily, friends, and significant other/ special person (Zimet et al,). Each of these groups was measured by four items, with a total of 12 items on t he total scale. For the Significant Other Subscale, CronbachÂ’s coefficient was 0.91, with that of the total scale being 0.88. This research used just the fou r items from the Significant Other Subscale. This Multidimensional Scale of Perceived Social Support was an indicator for the latent variable SITUATIONAL and wa s labeled as S3 in the figures. The scale is provided in Appendix F. Measures for Physiological Factor PH-1: Perceived exercise capacity. Perceived exercise capacity was operationally defined as the most strenuous activity an d the corresponding metabolic equivalents (METs) that one could sustain for 3 0 minutes. The empirical indicator for this was the one-item Rating of Perceived Capacity (RPC) scale (Wisen et al., 2002). The scale is a progressive scale from 1 to 20 METs with corresponding activity descriptions. The scale can be used to mathematically calculate predicted physical capacity for e xercise. The RPC scale was validated against the ramp cycle test, and refer ence values for METs are available for each decade of life (Wisen et al.) In the pilot study for this research (Cobb, 2006) the mean was 10.29 ( SD = 3.69; range 5 Â– 20). The Rating of Perceived Capacity scale was an indicator for t he latent variable PSYCHOLOGICAL and was labeled as PH1 in the figures. T he scale is provided in Appendix G.
72 PH-2: Perceived health status. Perceived health status was operationally defined as oneÂ’s perception of overall health. The em pirical indicator for this variable was the SF-12 (Ware, Kosinski, & Keller, 1996) Test-retest reliability of the SF-12 summary measure was 0.890 in the United Stat es. Coefficients ranged from 0.760 to 0.774 in the initial analysis. The short er version of the scale was able to reproduce more than 90% of the variance in t he SF-36 measure in the general US population (Ware et al., 1996). The SF-1 2 has been validated for different populations, including young adult homeless persons ( M age 37.40). CronbachÂ’s for this group ranged from 0.82 for physical health t o 0.79 for mental health (Larson, 2002). The SF-12 was an indicat or for the latent variable PHYSIOLOGICAL and was labeled as PH2 in the figures. T he scale is provided in Appendix H. PH-3: Anticipated fatigue. Anticipated fatigue was operationally defined as the fatigue that is anticipated from doing various fut ure activities. The empirical indicator for this was the Situational Fatigue Scale (C M. Yang & Wu, 2005), which was specifically designed to measure both mental and physical fatigue while taking the situational demands of various activiti es into consideration. It has two subscales. Four items comprise the Physical Fatigue Subsca le, with a CronbachÂ’s of 0.88. Nine items comprise the Mental Fatigue Subsca le, with a CronbachÂ’s of 0.89. Overall, the CronbachÂ’s was 0.90. The Situational Fatigue Scale was an indicator for the latent variable PHYSIOLOGICAL and was labeled as PH3 in the figures. The scale is provided in Appendix I.
73 Measures for Unpleasant Symptoms Factor US-1: Chronic fatigue. Chronic fatigue was operationally defined as the lessening of either mental or physical energy that has b een ongoing for at least a week. The empirical indicator for this variable was Chal der Fatigue Scale (Chalder et al., 1993). This is an 11item scale with two primary factors: physical and mental fatigue. CronbachÂ’s reliability of the Chalder Fatigue Scale was 0.845 for the physical fatigue items, 0.821 for the me ntal fatigue items, and 0.8903 overall. Subsequent testing by Morriss, Wearde n, and Mullis (1998) revealed that scoring may be done on a dichotomous basi s and still retain the overall reliability. When used in this pilot study (Co bb, 2006) the standardized was .8629 and the mean was 30.19 ( SD = 5.12; range 20 Â– 46). The Chalder Fatigue Scale was an indicator for the latent variable UNPLEASANT SYMPTOMS and was labeled as US1 in the figures. The sca le was provided in Appendix J. US-2 and US-3: Chronic pain. Chronic pain was operationally defined as an ache, discomfort, soreness, or throbbing that that w as ongoing for at least a week. The empirical indicator for this variable was the West Haven-Yale Multidimensional Pain Inventory by Kerns, Turk, and Ru dy (1985). The first part of the scale is comprised of 20 items, each rated on a Li kert-type scale but with varying response patterns depending upon the nature of the question. It is a subjective assessment of pain descriptions and how it affect s the participantÂ’s life. From those 20 questions are five subscales, two of which were used for this study (the Pain Severity Subscale, with factor loadings ranging from .68 to .80,
74 and the Negative Mood Subscale, with factor loadings ra nging from 0.59 to 0.87). These two subscales were used as indicators for the latent v ariable UNPLEASANT SYMPTOMS and were labeled US2 and US3 re spectively. The scale is provided in Appendix K. Measure for Activities E-1: Exercise. Physical activity was measured using the short form of t he International Physical Activity Questionnaire ([IPAQ], Craig et al., 2003; IPAQ, 2005). Exercise was operationally defined as the use of physical activity to expend energy, which was measured by intensity, frequen cy, and duration of the exercise. The empirical indicator for this variable was t he International Physical Activity Questionnaire short form which assessed walking, m oderate-intensity activities and vigorous-intensity activities. The IPAQ pr ovided separate scores for each of the levels of activity. The total minutes per week in physical activity was computed by summing the frequency in minutes by duratio n in days. Data were converted into metabolic equivalents per minute per we ek (METs min -1 / week) by weighting each type of activity by its energy requ irements defined in METs. The weights were as follows: a) 8 for vigorous intensity activity, b) 4 for moderate-intensity activity, and c) 3.3 for walking. Te st Â– retest SpearmanÂ’s reliability coefficients for the IPAQ short form when t ested in the United States ranged from .81 to .88 (Craig et al.). The pooled for the short form was .76 by 1,974 persons across 12 countries (Craig et al.). When us ed in the pilot study for this research (Cobb, 2006) the mean was 4036.27 (SD = 4 297, range 198 Â– 23,460). In the pilot study, participants reported di fficulty with estimating the
75 hours/minutes in each activity level, and therefore th ere were several who selected Â‘DonÂ’t knowÂ” as a response. To enhance the resp onse rate of those questions, the Â‘DonÂ’t KnowÂ’ response option was deleted for this research. Scoring instructions for the IPAQ (IPAQ, 2005) were to discard participants who reported more than 3 hours per day of vigorous activit y or of moderate activity; therefore, the response options for the Â‘hours per d ayÂ’ question was limited to a drop-down menu of four options (0 Â– 3 hours). Likewise, instructions for the IPAQ were to discard those who reported more than 16 hours p er day cumulative in all activities. Accordingly, the menu of options was limited to 12 hours maximum for the walking, and to 16 hours maximum for sitting. The IPAQ responses were all provided in drop-down menus to eliminate the Â‘fill-i n-the-blankÂ’ question format. These changes were anticipated to increase the overall r esponse rate for the IPAQ, and to minimize outliers. Physical activity was represented by a latent variable EXERCISE with EX1 as the label for its sole indicator, the IPAQ. Because i t was a single indicator, the measurement error for EX1 was fixed at .25, which was derived from the testretest reliability of .75 reported by Craig et al (2 003). The IPAQ questionnaire is provided in Appendix L. Reliability and Validity of the Research Design The purpose of this study was to determine the valid ity of two theoretical models of exercise utilizing the theory of unpleasant sy mptoms and social cognitive variables. The cross-sectional approach to data co llection was most
76 appropriate at this early stage in the development of the model to isolate the relationships among the variables. Structural equation modeling (SEM) was the appropriat e choice of analytic techniques available to test the theoretical models that were proposed a priori. Structural equation modeling, using the maximum likeli hood estimation procedure, is a full information technique in that al l model parameters are estimated simultaneously and a change in one parameter during the iteration process could result in a change in other parameters in t he model (Diamantopoulos & Siguaw, 2005). Additionally, SEM m odels measurement error as part of the parameter estimation process and is there fore more germane to testing the model than the use of path analysis, which ca rries an assumption of measurement of variables without error. The reliability of the research design was ensured thr ough the consistent application of procedures for data collection, correctio n, and analyses. The integrity of the research was also enhanced by specifying more that one theoretical model apriori and by making model modific ations only if theory-driven not data driven, thereby helping reduce error from o ver analyzing the data. Assumptions The proposed study was based on the assumption that an a dequate sample would be obtained. The use of DillmanÂ’s (2007) revised total design method that included a total of four contacts with pot ential respondents was projected to yield a response rate of 34%. This respon se rate was based on
77 studies by Leece et al. (2006) that addressed certain desi gn features of the letters that were sent in Internet surveys. Model Identification Prior to beginning analysis, the number of parameters to be estimated in the model was calculated and compared to the number of data points. To be testable, the model needed to have fewer parameters t han data points. Using Bentler and ChouÂ’s formula (1987), there were 91 dat a points in the variance/covariance matrix, which met the criterion of h aving more data points than parameters to be measured. The following formul a was used to calculate the number of data points: p* = p (p + 1) / 2, where p was the number of variables and p* was the num ber of data points. The calculations for this research were as follows: p = 13(13 + 1) / 2 = 91 data points This satisfied the requirement to exceed the 32 param eters for the model An alternative formula for checking identification is th e following formula t s / 2, where t is the number of parameters to be estimated, s is the number of variances/covariances amongst the observed variables calculat ed as ( p + q )( p + q + f1), where p is the number of y-variables and q is the number of x variables (Diamantopoulos & Siguaw, 2005).
78 In this case, the model was over-identified (having mo re data points than parameters). Had the opposite been true, the model cou ld not have been tested reliably. Generally under-identified models produce unreliable statistics (Bentler & Chou, 1987) because the p -values for the model might have been too low as a result of under-identification. Data Analysis Structural equation modeling is a causal model in which the paths in a graphic model are expressed as a series of algebraic equat ions (Boyd, Frey, & Aaronson, 1988). Theoretical variables, which are not observable but are presumed to exist, are known as latent variables. Measur able and observable variables known as manifest variables are used as indicat ors for the theoretical constructs. Karl Joreskog created a software program for th e analysis of linear structural relations and named it LISREL by its acronym (Boyd et al.). This was the software program chosen for this analysis. The analytic strategy followed the steps outlined by Di amantopoulos and Siguaw (2005) for structural equation modeling. The t erm LISREL is an acronym for linear structural relationships, and is the name o f the computer software used for covariance structure analysis. Covariance structure ana lysis is a multivariate statistical technique which combines confirmatory factor an alysis and modeling to analyze hypothesized relationships among latent varia bles and manifest indicators. The typical full covariance structure model co ntains two parts: a) the measurement model and b) the structural model. The a nalysis seeks to confirm that the hypothesized relationships across latent variab les and their manifest
79 indicators are consistent with empirical data. This is done by comparing the covariance matrix implied by the structural equation (h ypothesized) model to the actual covariance matrix derived from the empirical da ta. The goal of SEM is to explain the patterns of covarian ce observed among the study variables (Kelloway, 1998). In essence, the m odel explains if two or more variables are related. Path diagrams depict the m odels; a simple path represents the direct relationship between two variabl es and a compound path represents the product of two or more paths. In turn, t he sum of the simple and compound paths linking two latent variables produces the correlation that links the two variables. Decomposition of the correlations pr oduces the beta weights (standardized regression coefficients). These structural r elations are represented by structural equations, which in turn are combined to produce the implied correlation matrix (Kelloway, 1998). Theref ore examination of the bivariate correlations is a necessary preliminary step. The manifest indicators are reflective, meaning that t hey are simply the observed characteristics of an underlying construct (Diamant opoulos & Siguaw, 2005). It is the underlying constructÂ’s relationships tha t define the value of each X. Recalling from the methods chapter that X 1 = 1 1 + n 1, if correlations amongst the manifest reflective indi cators for any given latent variable are not related, then that reflects a misspecified or poorly conceptualized underlying concept.
80 Preliminary analyses included identification of values outside the range of permissible responses and listwise deletion of outliers, asse ssment of univariate and multivariate normality, and examination of biva riate relationships among the indicators. Next the measurement model was assessed as describ ed in the methods section, including validity and reliability of the model. Once an acceptable fitting measurement model was obtained, the full structural models as well as associated mediating variables were tested as descri bed in the methods section. Model modifications were attempted but not re tained, and model crossvalidation was not feasible for this single-sample set o f data.
81 Figure 6. LISREL Steps (adapted from Diamantopoulos & Siguaw, 2005, p.7).
82 Parameter Estimation As noted earlier, structural equation modeling (SEM) i s a method of doing a covariance structure analysis. The implied covariance m atrix is one which implies certain predictions for the variances and covarian ces of the variables in the model. Written in matrix notation, the model-ba sed covariance matrix is as follows: where is a vector containing the model parameters. The covari ance matrix is expressed as a function of the model parameters. If the model is correct and if the parameters are known, the population covariance ma trix would be exactly reproduced by the data. The observed sample variances and covariances contained in matrix S are compared to the model-based covariance matrix; the difference between the two matrices is known as the resi dual matrix. The aim of SEM is to minimize this difference (Diamantopoulos & S iguaw, 2005). The model equations are written as a set of matrices t hat correspond to different components of the model. These matrices are d enoted in Greek notation. These matrices and their corresponding model components are described in Table 4. LISREL matrix notation involves designating numbers as functions: the value of 1 in the equation tells LISREL to estimate the parameter for that matrix element; the value of 0 in the equa tion tells LISREL to Â‘fixÂ’ or Â‘constrainÂ’ that matrix element to zero (Diamantopoul os & Siguaw, 2005). = b 2 VAR(X) + VAR (e) bVAR(x) VAR(x)
83 Table 4 Summary of LISREL Matrices and Greek Notation Matrix Title LISREL Notation Matrix Symbol (Element Symbol) Model Components Lambda X LX r x ( x ) Paths from latent X variables to their indicators Lambda Y LY r y ( y ) Paths from latent Y variables to their indicators Theta Delta TD ( ) Variance-covariance matrix between error scores for X variables Theta Epsilon TE n ( n ) Variance-covariance matrix between error scores for Y variables Phi PH ( ) Variance-covariance matrix for the latent X variables Gamma GA ( ) Causal paths from latent X to latent Y variables Beta BE ( ) Causal paths from latent X to latent Y variables
84 Table 4 (continued) Matrix Title LISREL Notation Matrix Symbol (Element Symbol) Model Components Psi PS ( ) Variance-covariance matrix of residual terms for latent Y constructs Data Preparation and Screening After all data were exported from the Ultimate Surv ey to SPSS (SPSS, 2002), error-checking procedures were undertaken. First, the frequency distributions of all collected variables were examined to identify values outside the permissible range of response options. Individual re cords with outliers were identified, errors corrected, and the entire record e xamined for data entry accuracy. The process of checking the frequency distribution of all study variables continued iteratively until no values outside the permissible range of response options were identified. The second error-checking procedure involved selection o f a random sample of 10 of the records in the database. Data in each entry were checked against the source document in Ultimate Survey to ver ify successful data export directly into SPSS. Missing Data The design of the Internet survey gave participants a visual indicator of their progress in the survey. Missing data was minimize d by visually presenting only one question matrix at a time. Conditions were set to restrict any
85 unnecessary questions from appearing to the participant by using skip patterns. For example, if they responded that they did VPA on zero days of the week, they did not receive the subsequent two questions dealing wit h hours per day and minutes per day spent in VPA. However, there were no forced responses, so missing data were anticipated. A number of strategies w ere undertaken to assess and/or intervene with missing data. First, SPSS w as used to count the number of missing responses for individual items included in the survey batch. Since the analysis plan included variables expressed as a total subscale score of a measure and variables expressed as a single indicator of the respective latent variable, different strategies were necessary to deal wi th missing data, based on how the individual items were used in the planned ana lysis. For variables that were expressed as a total subscale sco re, the pattern and quantity of missing data was assessed for each individu al item comprising the respective subscales as well as the aggregate responses for all items included in all subscale calculations. If less than 10% of the responses were missing from an item comprising a specific subscale, and th e pattern of missing data was determined to be missing at random, missing dat a were supplanted by the mean of that item. If greater than 10% of the d ata were missing for a single item included in the calculation of a subscale score, the i tem was excluded from the calculations used to determine the subscale score. There is an application program for manipulating dat a, transforming data, computing covariance matrices, and performing explorator y analyses called PRELIS (precursor to LISREL). Using a graphical interfa ce, users can define
86 variable properties, insert variables, or delete cases ( DuToit, DuToit, Mels, & Cheng, 2005). Data from SPSS or from Excel can be imp orted into PRELIS and then the data can be cleaned. This mechanism was used as a safety check for cleaning done in SPSS; the output matrices were the sam e using either program. Ultimate Survey had the option of exporting data as a comma delimited file with an SPSS code book of variables and value labels. This option was chosen. After all data screening and missing data procedu res were completed, the mean subscales scores were calculated to come up with t he indicators for the latent variables to be tested in the theoretical model s outlined in Figures 2 and 3. Next PRELIS (Joreskog & Sorbom, 2005) was used to constr uct the covariance matrix used to test the theoretical model as depicted in Figures 2 and 3. Preliminary Analyses Multivariate normality. In this analysis, the multivariate normality of the data was assessed as specified by Diamantopoulos and Siguaw (2005). Assessment of univariate and multivariate normality was done through PRELIS, which is a program used for preprocessing the raw data. One of the assumptions of parameter estimation using the maximum likelihood ( ML) estimation method is that departures from multivariate normality are not too severe. While ML estimation is robust to minor violations, severe ones ren der the ML estimation questionable. Multivariate normality assumption is also needed for interpretation of standard errors and chi-square statistics (Diamantopoulo s & Siguaw). The tests for univariate normality for continuous variables were assessed. The univariate tests examined each variable individually and calculated a z-score
87 coefficient of skewness and kurtosis; significant p values indicated departures that were significantly different from zero. The mul tivariate measures of skewness and kurtosis were also measured. Skewness has to do with the symmetry of the distribution, whereas kurtosis has to do with the peakedness of the distribution. Skewness is 0 and kurtosis is 3 with a n ormal distribution (Olsson, Foss, Troye, & Howell, 2000). According to Curran, West, and Finch (1996), ML estimat ion is robust even at higher levels of skewness and kurtosis, given larg e sample sizes. Curran et al. found that ML was more likely to detect a specifi cation error given increasing departures from normality. At moderate univ ariate skewness of two and at kurtosis of seven, Curran et al. found 6% bias a nd 100% rejection of the model using chi-squared as the statistic with N of 500. A t severe univariate skewness of three and kurtosis of twenty-one, Curran et al found 18% bias and 100% rejection of the chi-square with N of 500. Anoth er finding was that as the severity of the nonnormality increased, the greater th e corresponding loss of power. Therefore one must plan to include additional subjects in the study to compensate for loss of statistical power from nonnormal d ata (Curran et al.). Once multivariate normality was assessed, a two-step appro ach was used to test the proposed theoretical model. First, the measurement model as depicted in Figure 7 for each latent variable was teste d to determine the fit of the model to the data.
88 Figure 7. Measurement Model. Based on the assessment of each measurement modelÂ’s fit to the data, appropriate modifications were undertaken to improve m easurement model fit. The first step was undertaken based on the recommendatio n of Kelloway (1998) that if the final model does not fit the data, measur ement model misfit could be ruled out as a source of the misfit of the model to the data, and attention could be focused on improving model fit through the modificatio n of structural parameters. Measurement Model A measurement model is one in which the posited relati ons of the observed variables to the underlying constructs is specifie d (J. C. Anderson & Gerbing, 1988). When building measurement models, t he use of multiple indicators is preferred because the meaning given to the underlying construct is less ambiguous with more details; therefore at least two indicators are desired
89 and at least four are preferred (J. C. Anderson & Gerb ing). If an indicator estimates only one construct, it is unidimensional and lo ads on only the one construct; however if it is multidimensional and loads on more than one construct, it is correlated with the other indicators and becomes problematic in interpretation of meaning (J. C. Anderson & Gerbing). The relationship between an indicator and its underlying construct can be expressed algebraically: X = + where X is a vector of observed variables, is a matrix of factor loadings relating the observed measures to the underlying construct and is a vector of random measurement error (J. C. Anderson & Gerbing). A lternatively the patterns could have been specified as follows and mainta ined the same measurement model: Y = r + n There were five latent variables and thirteen indicat ors for those constructs. There were 35 parameters to be estimated, using 56 deg rees of freedom in the measurement model. Validity and reliability of measurement model. Evidence for validity of the indicators used to represent the constructs was assessed by met hods described by Diamantopoulos and Siguaw (2005). First, all indica tor loadings were examined for significance (at p < .05 or better), as indicated by significant tvalues. The error variances were examined next; insigni ficant error variances may indicate specification error (Diamantopoulos & Sigua w). Because of the difficulty in comparing the validity of different in dicators, which use different
90 scales and which possibly had different reference scales from othersÂ’ analyses, the magnitudes of the completely standardized loadings were also inspected. Evidence for reliability of the indicators used to rep resent the constructs was assessed also by methods described by Diamantopoulos a nd Siguaw (2005). First the square multiple correlations ( R 2 ) were assessed because they showed the proportion of variance in each indicator th at is explained by its underlying latent variable (Diamantopoulos & Siguaw) and a higher R 2 denotes higher reliability. Next a composite reliability valu e for each latent variable was calculated to assess construct reliability using the follow ing formula: c = ( n ) 2 / [( ( n ) 2 + ( )] where c was the composite reliability, was the indicator loading, was the indicator error variances of the s or s, and was the summation over the indicators of the latent variable (Diamantopoulos & Si guaw). A c value of greater than 0.6 provided evidence that the indicators were r eliable measurements of the construct. And finally, a complementary measurement of c omposite reliability was calculated, which was the average variance extracted ( v ). This showed the amount of variance that was captured by the construct in relation to the amount of error variance. It was calculated by the following f ormula: v = ( n 2 ) / [ n 2 + ( ) ] where n was the indicator loading, was the indicator error variances of the s or s, and was the summation over the indicators of the latent v ariable (Diamantopoulos & Siguaw). It was desirable for the va lue of v to be at least
91 0.50 or above to show that a substantial amount of th e variance in the indicators was captured by the construct versus that accounted for by measurement error. Structural Models Once an acceptable fitting measurement model for each l atent variable was obtained, the full models were tested using structura l equation modeling implemented through LISREL (Joreskog & Sorbom, 2005). Structural parameters, the relationships between latent variables, were expressed as a series of equations and these equations transformed into an instruction set for the analyses. Figure 8. Structural Model 1. Path diagram depicting the structural relations for the theory of unpleasant symptoms
92 As depicted in Figure 8, there were three latent exog enous variables and two latent endogenous variables. The structural model had 32 parameters that had to be estimated and 59 degrees of freedom. A structural model is one which specifies the posited causa l relations of the estimated constructs. The structural relationship can b e expressed as an equation as follows: r = r + + where r represents the vector of endogenous constructs, represents the vector of exogenous constructs, represents the matrix of coefficients for the effects of the endogenous constructs on one another, represents the matrix of coefficients for the effects of the exogenous constructs on the endogenous constructs, and represents the vector of residual errors in the equatio ns and random disturbance terms.
93 Figure 9. Structural Model 2. Path diagram depicting the structural relations for the hypothesized model that altered the theory of un pleasant symptoms As depicted in Figure 9, there were two latent exogen ous variables and three latent endogenous variables. The structural mode l had 32 parameters that had to be estimated and 59 degrees of freedom. Assessment of structural models. Assessment of the structural models involved determining where the theoretical relations hips specified in the models were indeed supported by the data. This involved thr ee steps: a) examine the signs of the parameters to see if they matched the hypot hesized direction, b) examine the magnitudes of the parameters to determine if they were significantly different from zero, and c) examine the R 2 to determine how greatly it explained the joint power of the hypothesized antecedents (Diama ntopoulos & Siguaw, 2005)
94 Assessment of the fit of the model to the data was eval uated using comparative fit indices as recommended by Beckstead (2002a ; 2002b; 2005; 2006).and through other authors (J. C. Anderson & Gerb ing, 1988; Diamantopoulos & Siguaw, 2005; Jaccard & Wan, 1996; K elloway, 1998). If model modifications were necessary, these were undertak en only if theoretical and statistical evidence can justify such a modification. Model Modifications Overview of Model Specification Methods Modifications can be to the measurement model or the structural models. The measurement model can be modified by changing the patterns of the loadings or by changing the measurement error matrices. The structural model can be modified by changing the path coefficients from fixed to free or vice versa, or by altering the relationships of the correlations o f the disturbance terms. Reducing the parameters to be estimated produces a more parsimonious model, which inevitably results in an increase of the degrees of freedom and the chisquare statistic (Diamantopoulos & Siguaw, 2005). Howeve r first adding parameters to be estimated, although at the cost of pa rsimony, will decrease the chi-square statistic and improve model fit. The recommend ed method is to first improve the fit of the model prior to improving par simony (Diamantopoulos & Siguaw). Model modifications in covariance structure analysis can be problematic because the stability or consistency of model modificatio ns over repeated samples is threatened (R. C. MacCallum, Roznowski, & Necow itz, 1992).
95 Another concern is the issue of cross-validation, or how we ll that modified model fits an independent sample from the same population (R C. MacCallum et al., 1992). Because of the capitalization on chance, using t he data-driven process of model specification reduces the generalizability of the model to other samples and to the population (R.C. MacCallum et al., 1992). Modifications of an initial model to improve fit ha s too often been done when sample sizes were too small, when too many modifi cations were used, and modifications were not justified on substantive ground s (R. C. MacCallum et al., 1992). MacCallum et al. drew repeated samples from a l arge population and demonstrated that unless n is quite large, the fit of the final model becomes dependent on matters of sampling. Therefore MacCallum et al. (1992) heartily endorsed a different method of finding an adequate fit to the model. Based on their advice, two models were planned a priori. The testing of the specific aims incor porated testing both of the models that were selected a priori based on the liter ature of the theoretical concepts. The central hypothesis of this research was that th e relationships as depicted in the proposed theoretical models (see Figure s 1 and 2) would be reproducible in data from women of ages 18 to 25. Th ese hypotheses are represented algebraically as = ( ), where represents the observed population covariance matrix, is the vector of model parameters, and ( ) represents the covariance matrix implied by the model (Kelloway, 1998).
96 Goodness of Fit Indices The LISREL program provides several goodness-of-fit ind ices. The indices used in this analysis are discussed. The minimum fit funct ion chi-square, the root meant square error of approximation (RMSEA), the nor med fit index (NFI), the non-normed fit index (NNFI), the comparative fit ind ex (CFI), the goodness of fit index (GFI), the adjusted goodness of fit index (AGFI ), and the parsimony goodness of fit index (PGFI) are introduced here. The minimum fit function chisquare is unlike the more familiar use of the chi-squared statistic. With structural equation models, the goal is to equate the estimated covariance matrix implied by the model and the population covariance matrix ga thered from the empirical data. Equality between those two matrices indicates a p erfect fit. Departures from this perfect fit are determined by various fit indices a nd by examining the residual discrepancies between the observed and implied covariance s (Ratner, Bottorff, & Johnson, 1998). A small nonsignificant chi-squared prov ides evidence that the specified model and the empirical data are congruent ra ther than different. The chi-squared statistic is sensitive to sample size; therefore when using sample sizes large enough to support using LISREL, the chi-squa red statistic is often rejected as a function of the sample size (Boyd et al., 1988; Ratner et al.). Marsh et al. (1988) noted three types of indices; the stand-alone indices will be discussed first. The stand-alone indices include the chi-squared test statistic, the 2 / df ratio, LISRELÂ’s root-mean-square residual (RMR), GFI, and adjusted GFI. As noted above, the 2 is sensitive to sample size; this is because the formula for 2 involves N in the calculations (Marsh et al.). In contr ast, the
97 RMSEA focuses on the discrepancy between and ( ), while taking df or model complexity, into account. Values indicative of go od fit are those under 0.05; values between 0.05 and under 0.08 indicate a r easonable fit; values between 0.08 and 0.10 are of mediocre fit; and value s > 0.10 indicate poor fit (Diamantopoulos & Siguaw, 2005). While others label t he values differently (see Kelloway, 1998), generally the value of less than 0. 05 is desired. Accordingly, LISREL provides a test of significance of the RMSEA that indicates whether the RMSEA is significantly different from 0.05. The 90% con fidence intervals are also provided; thus reporting RMSEA is advantageous (Kellow ay). The RMR by Joreskog and Sorbom is the square root of th e mean of the squared residuals; its range depends upon the type of ma trix used in the approximations. If correlation matrices are used, the r ange is 0 to 1; however if covariance matrices are used, the range starts at zero but can exceed one, with no upper bound noted (Marsh et al., 1988); therefore the interpretation of the RMR is more difficult. Accordingly, LISREL provides a su mmary measure of the standardized residuals (the residuals divided by their e stimated standard errors); this summary measure, the standardized RMR, is indicative of acceptable fit if it is less than 0.05 (Diamantopoulos & Siguaw, 2005). The GFI is another commonly reported index. It is base d on the ratio of the sum of the squared discrepancies to the observed vari ances, thus as the observed variances increase, so does the GFI. It ranges f rom 0 to 1, with values greater than 0.9 indicating that the data fits well ( Kelloway, 1998). The GFI is an absolute fit index in that it directly assessed how well the predicted covariance
98 ( ) from the parameter estimates reproduces the sample c ovariance from the empirical data. According to Kelloway, GFI is generall y recommended as the most reliable measure of absolute fit. The GFI normall y ranges from 0 to 1; higher values indicate better fit, with values of at l east 0.90 preferred (Diamantopoulos & Siguaw, 2005). The GFI is independ ent of sample sizes and it is possible for it to be negative (Marsh et al., 198 8). The AGFI is similar to the GFI in that it adjusts the GFI for degrees of freedom thus penalizing the use of additional parameters. It too generally ranges from 0 to 1 but can be negative (Marsh et al.). Comparative or relative fit indices show how much bett er the model fits compared to a baseline model. The comparative fit indi ces do not compare against a perfect model; instead, they compare to a kno wn poor model (usually the null or independence model, see Kelloway, 1998). The NFI, NNFI and the CFI are all relative fit indices, with CFI being the one most often reported in the literature (Diamantopoulos & Siguaw, 2005). The NNFI range starts at zero and can exceed the value of one, whereas the NFI and CFI range from 0 to 1. In both, higher values indicate better fit, with values o f at least 0.90 preferred (Diamantopoulos & Siguaw; Kelloway, 1998). The NFI sh ows the percentage improvement over the baseline null/independence model ; with an NFI of .90, the model is 90% better fitting than the null/independen ce model. Its counterpart is the PNFI, in which lower values are expected in relati on to the NFI (Kelloway). These indices are provided in the next chapter for resul ts.
99 Testing of Specific Aims Aim 1 In the first model, the unpleasant symptoms domain was posited to be the sole mediator variable between the independent psychol ogical, situational and physiological factors and the outcome activity factor. The first aim was to test if this model would be reproducible in data from women o f ages 18 to 25. Aim 2 The second aim was to determine if modifying the mode l to emphasize the psychological domain as a partial mediator between the exogenous variables and both unpleasant symptoms and physical activity woul d provide a better fit than the model without the added mediation. Based up on the prior research in the social-cognitive models of exercise, it was anticipate d that model 2 would be reproducible in the data with improved goodness of fit indices. Power Analysis Post Analyses Using the method described Diamantopoulos and Siguaw ( 2005), power analysis was done. This power value indicated the probab ility that a false null hypothesis, or an incorrect H 0 would be rejected, where the null hypothesis was specified as H 0 : Â– ( ) = 0 or as its equivalent H 0 : = ( ). MacCallum et al.(1996) provided the syntax in the appendix of the ir article for calculating posthoc power. Kim (2005) provided the syntax in the appe ndix of the article for calculating the needed sample size based upon the non-ce ntrality delta for the anticipated model. This analysis was also done post-hoc to validate the power analysis.
100 Human Subjects Research Risks to Subjects Human subject involvement Data were collected from a randomly selected sample of 464 active students who were enrolled in the University of South FloridaÂ’s information system. The sample were fem ale of any ethnicity but primarily Caucasian, African American or Hispanic according to the ethnic profile of USF, and ranged in age from 18 to 25. See Table 3 for the ethnic profile of USF. Sources of materials. Data for this study were provided by students through completion of an Internet-based survey using a university-provided program called Ultimate Survey. Survey questions were put into Ultimate Survey using various formats as needed. Formats includ ed dichotomous yes/no questions, matrices of questions all using the same scale, in dividual questions with rating scales, multiple choice options, and options t o fill in their own answers. Selected demographic data were obtained to assist with interpreting results. Invitations were sent out to email lists of par ticipants. Each participant received a link to the Ultimate Survey URL. The surve y was designed to allow each participant to take the survey only once and the p articipantÂ’s email address was automatically deleted from the invitation list as e ach survey was completed. This was done automatically by the Ultimate Survey sof tware mechanism. Mailing list database access was limited through password p rotection to the PI. Potential risks. The anticipated risks to subjects were minimal and involved psychosocial concerns. If a subject had experienced particularly strong
101 fatigue or had experienced feelings of pain related t o exercise, feelings of uneasiness might have returned when the participant com pleted the survey. This risk was anticipated to be minimal and transient, and wa s no greater than those experienced during a recall of the events to a colleagu e at school. Adequacy of Protection against Risks Recruitment and informed consent Subjects were recruited through a direct emailing of the URL link to the actual survey. A waiver of signed consent was obtained from the IRB since a signed consent document would have been the only permanent link of a subject to their responses. The required elements of informed consent were delivered in the cover page incl uded in the survey batch online. Protection against risk While the risks to participants were anticipated to be minimal, there was a potential likelihood that some subjects would experience transient feelings of unpleasantness as they recalled thei r exercise experiences. Participants were notified of this potential risk throug h the cover letter. The collected data were anonymous in that no personal i dentifying information was collected. The actions of deleting the participantsÂ’ survey number from the invitation list upon completion of t he survey and of completing all data collection prior to commencing data transfer w as an additional safeguard to protect the anonymity of responses. Participants wer e reminded in the cover letter and throughout the survey forms to avoid provi ding any information that could potentially identify them in their responses. I f identifying information was discovered at the time of data entry, this information was obliterated.
102 Potential Benefits of the Proposed Research Participants did not derive any direct benefits from t heir participation in this study. An incentive in the form of a chance at winning one of two separate $100 checks was offered to all those invited via the initial contact letter and reiterated in subsequent letters. Participants may have derived som e personal satisfaction with participating in a study of an important topic t o the general health of the public. Nursing professionals, health service administrato rs, and policy makers are anticipated to derive the indirect benefit from th e results of this study since these results added to the body of knowledge related t o exercise science and began to fill a gap in the knowledge about gender-spe cific processes leading to positive health practices. Inclusion of Women, Minorities, and Children Women were the focus of the study. It was anticipated t hat the racial/ethnic distribution of responses would closely correspond to the distribution of USF students, as indicated by the data from USF (2006b, see Table 5). Participants between the ages of 18 to 21 qualified as children according to the guidelines published by National Institutes of Health (1998). Adolescents were included in this study, therefore children are in cluded.
103 Table 5 Targeted/Planned Enrollment Table Targeted/Planned Enrollment: Number of Subjects Ethnic Category Total Hispanic or Latina 50 Not Hispanic or Latina 446 Ethnic Category: Total of all subjects 496 Racial Categories American Indian/Alaska Native 4 Asian 40 Native Hawaiian or other Pacific Islander 2 Black or African American 70 White 380 Racial Categories: Total of all Subjects 496
104 CHAPTER FOUR Results Overview of Analytic Strategy This chapter presents the results of the research. Samp le characteristics of participant are presented first, followed by a descri ption of the preliminary analysis. These included assessment of data quality, bivar iate relationships, and the measurement models. Problems initially encountered with the fit of the measurement models are addressed, as are the steps under taken to deal with these problems. These are followed by hypothesis testi ng, in which each research aim is addressed sequentially. Finally the powe r analysis is presented. Participant Characteristics Five hundred nineteen female students completed the study. The mean age of the participants was 21.57 (SD = 2.01; range 18 Â– 25). Of the 480 participants who completed the racial demographics, 76.9% ( n = 399) were Caucasian/White, 9.8% ( n = 51) were African Black or Caribbean Black, 0.2% ( n = 1) were Native Indian or Alaskan Indian, 0.4% ( n = 2) were Hawaii or Pacific Islanders, 5.2% ( N = 27) were Asian, and 6% ( n = 27) identified themselves as other. For ethnicity, 11.1% ( n = 58) were Hispanic. Among those ( n = 39) who identified their ethnicity as other, 10.2% ( n = 4) were African, 30.7 ( n = 12)
105 identified themselves as Â‘AmericanÂ’, 41.0% ( n = 16) were West Indian, and 17.9% ( n = 7) were of mixed heritage. Preliminary Analysis Data quality. Five hundred nineteen students completed the study, whi ch was a response rate of 9.0% from among 5733 deliverab le emails distributed. Another 56 (10.7%) were deleted listwise from analyses due to missing data and/or invalid or implausible responses. Specifically, 5 % ( n = 3) provided data with more than 25% of the responses missing; 7% ( n = 4) reported exercise hours or minutes but not the days per week; 22% ( n = 12) reported days of exercise but no hours or minutes; 1% ( n =1) reported implausible high amounts of time spent exercising (greater than 16 hours of exer cise per day); 5% ( n = 3) reported implausible low amounts of time spent exercisi ng (0 minutes per week); 56% ( n = 31) did not answer the single-item question about exer cise capacity; and 4% ( n = 4 ) reported implausible answers for the loneliness scal e which demonstrated a probable response bias on reverse scored it ems. The data from one participant were notable for more than one of th e aforementioned errors, summing to 57 erroneous observations among 519 females. Of the 463 participants whose data lacked discernible e rrors and were therefore included in the data analyses, 79% ( n = 364) were Caucasian/White, 9% ( n = 43) were African Black or Caribbean Black, 1% ( n = 1) were Native Indian or Alaskan Indian, 1% ( n = 2) were Hawaii or Pacific Islanders, 5% ( n = 23) were Asian, and 6% ( n = 27) identified themselves as other. The average age for
106 the sample was 21.57 years (SD = 2.01 years). Ethnic i dentities included 11% ( n = 51) Hispanic, 79% ( n = 368) Non-Hispanic, and 9% ( n = 44) others. Available data for the 56 participants who were exclu ded listwise are reported to address concerns about respondent bias. Signi ficance testing of comparisons between the included versus non-included parti cipants are summarized. The average age of the excluded participan ts was 21.30 (SD = 1.94 years). Excluded and included participants were similar by all racial categories and by age. The data from both groups were compared. The only variable on which these groups differed significantly was the total health status scale which was significantly lower among the 56 females whose data were excluded ( M = 74.02, SD = 35.99, t = 2.29, df = 516, p = .022). Thus it appeared that the 56 females who were excluded from the analysis were genera lly comparable to the 463 females were included in the analyses. In this data, 13 variables had severe univariate skewness and one had severe univariate kurtosis, as assessed by Curran et al. (1 996) criteria for z scores. Table 6 provides the tests for univariate normal ity in this study. Severe multivariate skewness was present but no severe multivari ate kurtosis was present, again using Curran et al.Â’s criteria.
107 Table 6 Univariate Normality Z-Scores Variable Skewness p value Kurtosis p value SE1 4.369 0 -1.93 0.054 SE2 -0.707 0.48 -4.718 0 SE3 -0.376 0.707 -6.567 0 EXP1 -9.777 0 6.074 0 EXP2 -8.511 0 4.882 0 EXP3 -7.808 0 3.654 0 G1 2.822 0.005 -5.099 0 G2 2.593 0.01 -5.064 0 G3 4.328 0 -1.72 0.085 L1 10.907 0 5.448 0 L2 11.87 0 6.404 0 SS1 1.688 0.091 -14.333 0 SS2 2.423 0.015 -6.316 0 SS3 2.739 0.006 -23.137 0 AGE 1.011 0.312 -8.92 0 Note SE = Self-efficacy; EXP = Expectations; G = Goals; L = Loneliness; SS = Social Support;
108 Table 6 (Continued) Variable Skewness p value Kurtosis p value SFTOT -7.662 0 3.746 0 RPC 4.826 0 3.15 0.002 US11 3.767 0 -0.124 0.901 US12 4.248 0 -1.085 0.278 US21 8.08 0 2.368 0.018 US22 3.045 0.002 -1.309 0.191 PA 9.942 0 5.174 0 Note. SE = Self-efficacy; EXP = Expectations; G = Goals; L = Loneliness; SS = Social Support; SFTOT = Perceived health; RPC = Rating perceived capacity; US = Unpleasant Symptom; PA = Physical Activity One possible solution to these violations of normality could have been to use a different method of estimation such as WLS. Weig hted lease squares estimation would have required an absolute minimum sam e size equal to k ( k -1) / 2 variables, where k is the number of variables (Diamantopoulos & Siguaw, 2005), therefore this sample size was adequate. Howeve r since the use of WLS has been found problematic even in sample sizes of 1000 the solution was to depend upon the robustness of ML estimation to departur es from normality. Accordingly, the ML method was selected as the most appro priate one to use in this set of data. The findings of Olsson et al. (2000) su pport this method, as ML was better at detecting misspecification errors at higher nonnormality.
109 Original Assessment of Bivariate Relationships Bivariate correlational analysis was used to make an init ial assessment of the relationships amongst the constructs, whereupon signif icant problems were noted. Each latent construct had up to three indicator v ariables, and while the indicator variables for any given construct were significa ntly correlated, the magnitudes of the correlations were not strong enough to demonstrate a single underlying construct. The correlations within any given construct were under 0.500 magnitudes. This indicated that the constructs were too broad and had to be narrowed. Original Assessment of Measurement Model Due to the indicator-construct links as originally posited, the initial measurement model failed to pass the criteria for anal ysis. For example, at least one of the lambda values was negative (an impossible an swer). Thus while the goodness of fit indices for the measurement model at f irst appeared to be of mediocre fit, these values could not be trusted due to t he illogical lambdas. Failure to pass the original assessment of the measureme nt model meant that further analysis could not be done. Hence the the oretical constructs were reviewed, and while keeping the same indicator variabl es already collected from the participants, the structural models were rearranged For example, the psychological latent variable was too broad; it was split into three different constructs: a) self-efficacy, b) expectations, and c) goals, which are congruent with the constructs of BanduraÂ’s self-efficacy theory (Ban dura, 2004). The three scales that originally had been combined as indicators for the psychological
110 latent variable were further subdivided so that one sca le (Exercise Self-efficacy Scale by Shin et al., 2001) provided the indicators fo r self-efficacy, one scale (Outcome Expectations for Exercise scale by Resnick et al., 2 001) provided the indicators for expectations, and one scale (Exercise Goals S cale by Rovniak et al., 2002) represented goals. Most of the subscales were b ased upon factor analyses provided by the authors. While the overall th eory of unpleasant symptoms remained unchanged, it now had three latent and narrower psychological variables instead of its broad psychological o ne. Similar changes were made for the other constructs. See Table 7 for a summary of the changes, and Figures 15 and 16 for graphic depiction of the chan ges in the theoretical models. The changes in the theoretical models also required chang es in the aims of this research. Originally the intent had been to t est model 1 (the theory of unpleasant symptoms as depicted by Lenz, 1995; 1997) and then to test model 2, which altered the theory of unpleasant symptoms to permit partial mediation. Conceptually, the broad factors as depicted in the theor y of unpleasant symptoms became theoretical domains. For instance the p sychological domain contained three factors; the situational domain contai ned two factors, the physiological domain contained three factors, and the unpleasant symptoms contained two factors. Accordingly, the measurement mode l and the structural models were altered.
111 Table 7 Regrouping of Indicators and Constructs Original Construct Original Indicators New Construct New Indicators PSYCHOLOGICAL PS1 SELFEFF SE1 PS2 SE2 PS3 SE3 EXPECT EXP1 EXP2 EXP3 GOALS G1 G2 G3 Note. SELFEFF = Self-efficacy; EXPECT = Expectations. PS1 = Exercise Selfefficacy Scale; PS2 = Outcome Expectations for Exercise Sca le; PS3 = Exercise Goals Scale. SE1 = ((q11 + q12 + q13 + q16 + q17 + q 18)/6) of the Exercise Self-efficacy Scale; SE2 = ((q4 + q8 + q10 + q14 + q15 )/5) of the Exercise Selfefficacy Scale; SE3 = ((q1 + q2 + q3 + q5 + q6 + q7 + q9)/7) of the Exercise Selfefficacy Scale. EXP1 = sum (q1 to q3); EXP2 = sum (q4 to q6); EXP3 = sum (q7 & q8) of the Outcome Expectations for Exercise Scale. Th ese were reverse scored into a positive direction to be consistent with the other indicators in same construct.G1 = sum (q1 to q3); G2 = sum (q4 to q6); G3 = sum (q7 to q10) of the Exercise Goals Scale.
112 Table 7 (Continued) Original Construct Original Indicators New Construct New Indicators SITUATIONAL S1 LONELY Lonely1 S2 Lonely2 S3 SOCSUPP SS1 SS2 SS3 PHYSIOLOGICAL PH1 AGE Age PH2 PH3 HEALTH SFTOT RATEXCAP Excap Note. LONELY = Loneliness; RATEXCAP = Rating of Exercise Cap acity. S1 = UCLA Loneliness Scale (UCLA-8); S2 = Social Support for Exercise Scale; S3 = Multidimensional Scale of Perceived Social Support L1 = sum (q2, q3, q11) of the UCLA-8; L2 = sum (q14, q17, q18) of the UCLA-8. SS1 = = sum (q1 to q3) of the Social Support for Exer cise Scale; SS2 = sum (q4 to q6) of the Social Support for Exercise Scale, and SS 3 = q13 of the Social Support for Exercise Scale. SFTOT = sum of transposed f actors from SF12-v12* *each dimension was altered by reducing the number of q uestions; EXCAP = Rating of Perceived Capacity.
113 Table 7 (Continued) Original Construct Original Indicators New Construct New Indicators UNPLEASANT SYMPTOMS US1 FATIGUE Fatsub1 US2 Fatsub2 US3 PAIN Painsub1 Painsub2 EXERCISE EX ACTIVITY PA Note. US1 = Chalder Fatigue Scale; US2 = the West Haven-Ya le Multidimensional Pain Scale. US11 = ((sum (q1 to q3)) + (sum (q6 to q8))) of the Chalder Fatigue Scale; US12 = q9 of the Chalder Fati gue Scale; US21 = (sum (q1, q7, q12)) of the West Haven-Yale Multidimensiona l Pain Scale; US22 = (sum (q6, q18, q20) of the West Haven-Yale Multidimen sional Pain Scale. EX and PA both = International Physical Activity Questi onnaire in its entirety.
114 Final Assessment of Bivariate Relationships Bivariate correlational analysis was repeated with the newly narrowed constructs. All of the indicators within each given constru ct were correlated at a magnitude of at least 0.600 except for one indicator (painsub2). All of the correlations were in the anticipated direction as well. Based on this new bivariate correlational analysis, the decision was made to continue assessing other aspects needed for the preliminary analyses. Consult Tab le 8 for the new bivariate correlations with corresponding means and stand ard deviations.
115 Table 8 Bivariate Correlations SEI SE2 SE3 EXP1 EXP2 EXP3 SE1 1 SE2 0.742** 1 SE3 0.755** 0.762** 1 EXP1 0.264** 0.242** 0.279** 1 EXP2 0.342** 0.315** 0.354** 0.764** 1 EXP3 0.252** 0.209** 0.263** 0.770** 0.717** 1 G1 0.406** 0.353** 0.371** 0.215** 0.333 0.235 G2 0.366** 0.365** 0.351** 0.229** 0.327** 0.255** G3 0.411** 0.383** 0.403** 0.246** 0.353** 0.273** L1 -0.025 -0.055 -0.088 -0.065 -0.065 -0.108* L2 -0.077 -0.102* -0.154** -0.141** -0.143** -0.141** SS1 0.086 0.027 0.043 0.043 0.060 0.184** SS2 0.182** 0.109* 0.149** 0.092* 0.111* 0.225** SS3 0.127** 0.069 0.095* 0.109* 0.134** 0.168** Note SE = Self-efficacy; EXP = Expectations; G = Goals; L = Loneliness; SS = Social Support; Correlation significant at the 0.05 level (2 tailed ) **Correlation significant at the 0.01 level (2 tailed )
116 Table 8 (Continued) SEI SE2 SE3 EXP1 EXP2 EXP3 AGE 0.078 0.084 0.063 0.092* 0.105* 0.017 SFTO 0.237** 0.253** 0.230** 0.119* 0.169** 0.097* RPC 0.358** 0.361** 0.382** 0.173** 0.295** 0.187** US11 -0.246** -0.208** -0.256** -0.106* -0.114* -0.127** US12 -0.192** -0.158** -0.190** -0.101* -0.061 -0.139** US21 -0.066 0.000 -0.032 -0.035 -0.040 -0.011 US22 -0.202** -0.171** -0.199** -0.081 -0.082 -0.070 PA 0.246** 0.250** 0.224** 0.091 0.188** 0.140** Means 21.712 26.241 35.428 12.652 12.469 8.240 SD 13.645 11.886 17.153 2.378 2.414 1.735 G1 G2 G3 L1 L2 SS1 G1 1 G2 0.758** 1 G3 0.727** 0.786** 1 L1 -0.101* -0.095* -0.114* 1 L2 -0.095* -0.090* -0.134** 0.613** 1 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ). Correlation significant at the 0.05 level (2 tailed ) **Correlation significant at the 0.01 level (2 tailed )
117 Table 8 (Continued) G1 G2 G3 L1 L2 SS1 SS1 0.154** 0.161** 0.149** -0.183** -0.154** 1 SS2 0.248** 0.263** 0.285** -0.144** -0.141** 0.800** SS3 0.147** 0.163** 0.202** -0.118* -0.128** 0.596** AGE 0.051 0.003 0.029 -0.033 -0.043 -0.050 SFTO 0.132** 0.174** 0.173** -0.114* -0.230** 0.000 RPC 0.270** 0.227** 0.260** -0.064 -0.081 0.077 US11 -0.157** -0.133** -0.189** 0.240** 0.340** -0.095 US12 -0.101* -0.102* -0.103* 0.254** 0.325** -0.079 US21 -0.019 0.029 0.021 0.079 0.096* 0.043 US22 -0.062 -0.042 -0.128** 0.252** 0.414** -0.050 PA 0.205** 0.223** 0.247** -0.013 -0.087 0.059 Means 7.786 7.562 9.108 10.437 10.394 8.423 SD 3.445 3.286 3.960 4.800 5.304 3.764 Note. SE = Self-efficacy (SE rescaled by 10 -1 ). ; EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12v2 (using approximately half the questions in each dimensi on); RPC = Rating perceived capacity; US = Unpleasant Symptom; PA = Physica l Activity. (PA rescaled by 1000 -1 ). Correlation significant at the 0.05 level (2 tailed ) **Correlation significant at the 0.01 level (2 tailed )
118 Table 8 (Continued) SS2 SS3 AGE SFTOT RPC US11 SS2 1 SS3 0.647** 1 AGE -0.010 -0.029 1 SFTO 0.016 0.060 0.008 1 RPC 0.151** 0.114* -0.046 0.299** 1 US11 -0.148** -0.050 -0.041 -0.484** -0.176** 1 US12 -0.103 -0.017 -0.008 -0.301** -0.124** 0.620** US21 0.090 0.063 -0.023 -0.434** -0.081 0.303** US22 -0.058 -0.036 0.023 -0.333** -0.095* 0.426** PA 0.125** 0.045 -0.152** 0.143** 0.227 -0.134 Means 8.149 2.695 21.583 67.351 8.909 14.998 SD 3.367 1.451 2.037 13.331 3.042 4.305 US12 US21 US22 PA US12 1 US21 0.647** 1 US22 -0.010 -0.029 1 PA 0.016 0.060 0.008 1 Means 2.11 0.390 0.803 4.096 SD 0.844 0.379 0.370 3.813 Note. SE = Self-efficacy (rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12v2 (using
119 approximately half the questions in each dimension); R PC = Rating Perceived Capacity; US = Unpleasant Symptom; PA = Physical Activit y (times 1000 -1 ). Correlation significant at the 0.05 level (2 tailed ) **Correlation significant at the 0.01 level (2 tailed ) Final Assessment of Measurement Model Validity and reliability. Evidence for validity of the indicators used to represent the constructs was assessed by methods described by Diamantopoulos and Siguaw (2005). First of all, indi cator loadings were examined for significance (at p < .05), as indicated by significant t Â– values. The measurement model with standardized values is depicted in Figure 10. All of the lambda parameters that were freed for estimation wer e significantly different than zero. Because of the difficulty in comparing the vali dity of different indicators, which use different scales, the relative magnitudes of t he completely standardized loadings were also inspected. Standardizat ion is advantageous in that it facilitates recognition of improper estimates (D iamantopoulos & Siguaw). The factor loadings or x are displayed in Table 9. All x values (completely standardized) were .68 or above with the one exceptio n, and as expected from the bivariate correlational analysis, that was for the pain indicators. These x values indicated that 20 of the22 indicators loaded hig hly on their respective latent factors.
120 SELFEFF1 0.26 SELFEFF2 0.26 SELFEFF3 0.23 EXPECT1 0.20 EXPECT2 0.26 EXPECT3 0.28 GOAL1 0.30 GOAL2 0.20 GOAL3 0.23 LONELY1 0.57 LONELY2 0.12 SSEX1 0.29 SSEX2 0.09 SSEX3 0.53 AGE 0.00 SFTOT 0.00 EXCAP 0.00 FATSUB1 0.18 FATSUB2 0.53 PAINSUB1 0.72 PAINSUB2 0.57 METMWEEK 0.00 SELFEFF 1.00 EXPECT 1.00 GOALS 1.00 LONELY 1.00 SOCSUP 1.00 AGE 1.00 HEALTH 1.00 RATEXCAP 1.00 FATIGUE 1.00 PAIN 1.00 ACTIVITY 1.00 Chi-Square=306.26, df=158, P-value=0.00000, RMSEA=0 .045 0.86 0.86 0.88 0.89 0.86 0.85 0.84 0.90 0.88 0.66 0.94 0.84 0.95 0.68 1.00 1.00 1.00 0.90 0.69 0.53 0.66 1.00 Figure 10. Measurement Model Results. Note. The correlations amongst the latent variables were not shown in an effort to maximize the visibility of the diagram. Lambdas an d theta-deltas are completely standardized. SE and SELFEFF = self efficacy ; EXP and EXPECT = expectations; G = goals; L and LONELINESS = loneliness, SS and SOCSUPP = social support; SFTOT = perceived health scale; RPC and RATEXCAP = rating perceived capacity; US = Unpleasant symptoms; PA = physical activity
121 Table 9 Measurement Model: Completely Standardized n x Coefficients Indicator x Latent Variable SE1 a .862 SELFEFF SE2 .862 SE3 .880 EXP1 a .893 EXPECT EXP2 .858 EXP3 .850 G1 a .839 GOALS G2 .896 G3 .877 L1 a .656 LONELY L2 .936 SS1 a .841 SOCSUPP SS2 .952 SS3 .685 AGE a 1.00 AGE Note SE and SELFEFF= Self-efficacy; EXP and EXPECT = Exp ectations; G = Goals; L = Loneliness; SS and SOCSUPP = Social Support a used as marker indicator for that construct, with scale set to 1
122 Table 9 (Continued) Indicator x Latent Variable SFTOT a 1.000 HEALTH RPC a 1.000 RATEXCAP US11 a .905 FATIGUE US12 .685 US21 a .528 PAIN US22 .658 PA a 1.000 ACTIVITY Note. SFTOT = perceived health status (SF12v2 portions); RPC and RATEXCAP = Rating of perceived capacity; US = Unpleasant Symptom ; PA = Physical Activity. a Scale was set to 1 on this indicator. Next the error variances were examined; nonsignificant error variances may indicate specification errors (Diamantopoulos & Sigua w, 2005). The of the loneliness subscale 2 was the only non-significant error variance among the 22 indicators. Next the reliability of the indicators used to repre sent the constructs was assessed. First the squared multiple correlations ( R 2 ) were assessed. The proportions of variance in each non-marker indicator th at was explained by its underlying latent variable ranged from .279 (pain s ubscale 2, as expected from its lambda), to .907 (social support for exercise subscale 2, as expected from its error variance) with 13 of 18 non-marker indicators ha ving R 2 greater than .70. With the exception of the pain subscale 2, all of the R 2 were at least .400.
123 Finally the composite reliability value for each late nt variable and its related average amount of variance extracted was calcula ted according to the formulas by Diamantopoulos and Siguaw (2005) given i n the method section. A composite reliability ( c ) greater than .60 provided evidence that the indicator s were reliable measures of the construct. Next the avera ge variance extracted ( v ) was calculated to reveal the amount of variance that wa s captured by the construct in relation to the amount of error variance. A value for v of at least .50 or above showed that a substantial amount of the vari ance in the indicators was captured by the construct versus that accounted for by mea surement error. Table 10 provides both the composite reliabilities and the average variance extracted for each of the constructs. As expected from the reported values of x, the composite reliabilities were above .60 with one exception, the latent variable of pain. Likewise, the amount of vari ance extracted for each of the constructs exceeded the desired .50 with the same exception pain. In summary, the composite reliabilities and the composite average variances extracted for the constructs were reliable. Only pain was slightly below th e desired limits.
124 Table 10 Composite Reliabilities and Average Variance Extracted Latent Variable c v SELFEFF .901 .766 EXPECT .900 .752 GOALS .903 .760 LONELY .784 .652 SOCSUPP .869 .612 AGE 1.000 a 1.000 a HEALTH 1.000 a 1.000 a RATEXCAP 1.000 a 1.000 a FATIGUE .780 .644 PAIN .522 .355 ACTIVITY 1.000 a 1.000 a Note. c = Composite reliability ; v = Amount of variance abstracted; a Scale was fixed to 1 on the single indicator of this latent vari able. SELFEFF = Self-efficacy; EXPECT = Expectations; SOCSUPP = Social Support; RATEX CAP = Rating of Exercise Capacity. Interrelations among latent factors Standardized covariances among the latent variables were examined in the measurement mo del as well, and are presented in Table 11. All of the correlations were in the direction hypothesized.
125 Table 11 Standardized Covariances among Latent Variables (N = 46 3) Variable SELFEFF EXPECT GOALS LONELY SOCSUPP SELFEFF 1 EXPECT 0.37 1 GOALS 0.497 0.356 1 LONELY -0.135 -0.171 -0.135 1 SOCSUPP 0.157 0.160 0.298 -0.176 1 Variable SELFEFF EXPECT GOALS LONELY SOCSUPP AGE 0.085 0.086 0.028 -0.046 -0.022 HEALTH 0.275 0.147 0.186 -0.239 0.019 RATEXCAP 0.423 0.246 0.285 -0.088 0.147 FATIGUE -0.303 -0.149 -0.195 0.42 -0.155 PAIN -0.246 -0.112 -0.078 0.507 -0.006 EXERCISE 0.276 0.154 0.259 -0.086 0.115 Variable AGE HEALTH RATEX USYM1 USYM2 ACTIVITY AGE 1 HEALTH 0.008 1 RATEXCAP -0.046 0.299 1 FATIGUE! -0.04 -0.52 -0.192 1 PAIN 0.009 -0.612 -0.148 0.699 1 ACTIVITY -0.152 0.143 0.227 -0.147 -0.035 1 Note. SELFEFF = Self-efficacy; EXPECT = Expectations; SOCSU PP = Social Support; RATEXCAP = Rating of Exercise Capacity
126 Goodness of fit. The fit of the measurement model was evaluated using several criteria as described in the methods chapter. Fo r the first criterion, that of the chi-squared statistic, the measurement model was reje cted ( 2 = 312.855, df = 158, p < .001). However, other fit indices suggested that the model adequately fit the data (RMSEA = 0.0451; CFI = 0.978; GFI = 0. 943; AGFI = 0.909; RMR = 0.0378; and PGFI = 0.589). These data tentatively suggested that the rejection of the model was primarily attributable to the large r sample size. In addition, the ratio of 2 to df was 1.98, which met the conventional criterion of the r atio of 2 to df being under two. Assessment of Structural Models Because of the restructuring of the latent variables described earlier in this chapter, the structural models were respecified to accommo date 11 latent variables. These changes were depicted in Figures 11 and 12 below. Corresponding to changes in the hypothesized structural m odels, the aims of the study were expanded to include the increased number of latent variables but otherwise remained the same.
127 Figure 11. Revised Model 1 Path Diagram. Note The theory of unpleasant symptoms; path diagram depict ing the structural relations among 11 latent variables. Shaded boxes outl ine the original psychological, situational, physiological, unpleasant sympt oms, and activity domains as described by Lenz et al. (1997). 1 = SELF-EFFICACY; 2 = EXPECTATIONS; 3 = GOALS; 4 = LONELINESS, 5 = SOCIAL SUPPORT; 6 = AGE; 7 = HEALTH and 8 = RATING OF EXERCISE CAPACITY; 1 = Fatigue; 2 = PAIN (Pain); 3 = ACTIVITY. All s are correlated.
128 Figure 12. Revised Model 2: Path Diagram. Correlations amongst s not shown for clarity of the diagram.
129 Hypothesis Testing The central hypothesis of this research was that the rel ationships as depicted in the proposed theoretical models (see Figure s 11 and 12) would be reproducible in data from women of ages 18 to 25. Assessment of Model Fit Aim 1 In the first model, the unpleasant symptoms domain was posited to be the sole mediator variable between the independent psychol ogical, situational and physiological factors and the outcome activity factor. The first aim was to test if this model would be reproducible in data from women o f ages 18 to 25. Using the 2 statistic as the criterion, the first model was rejected ( 2 = 400.120, df = 167, p < .001). However, other fit indices provided evidence that model 1 adequately fit the data (GFI = 0.926, AGFI = 0.889, CVI = 0.966, RMSEA = 0.0554, and standardized RMR = 0.049). The completed structural model in Figure 13 contains th e standardized path coefficients ( and ) and disturbances ( ). The disturbances communicate the proportion of unexplained variance (1 Â– R 2 ) in the endogenous variables or sources of influences on the endogenous variables depicted in the model. In model 1, FATIGUE had a significant total effect on ACTIVITY ( t = 2.784, = 0.178). In contrast to fatigue, PAIN did not hav e a significant effect on ACTIVITY. Next the squared multiple correlations for the Y variables were examined for model 1. Only two of five indicators for the endogenous variables explained at least 70% of their latent variables. Re spectively, fatigue subscale 1
130 and physical activity explained 81.4% and 98.3%. Next t he squared multiple correlations for the X variables were checked. All of the squared multiple correlations for the X variables were above 70% with th e exception of two x variables: loneliness subscale 1 and social support for exercise subscale 3. For the entire SEM, PAIN had the most variance explained ( R 2 = 56.1%) and FATIGUE had the second most amount of variance expla ined ( R 2 = 41.8%). Unfortunately however, model 1 only explained 3% of the variance for ACTIVITY.
131 Figure 13. Completed Structural Model 1. Note Path coefficients and disturbances are completely stand ardized; Correlations amongst s not shown for clarity of the diagram.
132 Aim 2 The second aim was to determine if altering the model from a fully mediated model to a partially mediated model would improve the fit of the model. Based upon the prior research in the social-cognitive mod els of exercise, it was anticipated that model 2 would be reproducible in the data with improved goodness of fit indices. As in model 1, model 2 was statistically rejected ( 2 =341.520, df = 159, p = .000). The ratio of the 2 to the df was 2.14. The other fit indices showed that model 2 fit the data adequately (GFI = 0.938, AGFI = 0.901, CVI = 0.973, RMSEA = 0.0493, and standardized RMR = 0.10). The completed structural model in Figure 14 contains th e standardized path coefficients ( and ) and disturbances ( ) for model 2. The disturbances communicate the proportion of unexplained variance (1 Â– R 2 ) in the endogenous variables or sources of influences on the endogenous varia bles depicted in the model.
133 Figure 14. Completed Structural Model 2. Note. Path coefficients and disturbances are completely standa rdized. All s are correlated. Statistically significant
134 Fatigue In model 1 fatigue had a significant total effect; h owever, in model 2 after controlling for effects of other variables, FA TIGUE had a non-significant total effect on ACTIVITY ( t = 1.038; = 0.068). Pain In model 2, PAIN still had a non-significant total effect on ACTIVITY ( t = 1.637, = 0.209). The direction of the relationship of PAIN on ACTIVTY was just the opposite than that which had been anticipated. It had been hypothesized based upon the model that PAIN would have a negative effect on ACTIVITY and would be of small magnitude. Instead it had a positive effect of moderate nonsignificant magnitude. This led to the suspicion that th ere might be a suppressor variable inflating the effect of pain. According to T abachnick and Fidell (2000) either one of two criteria indicates that a suppressor va riable is present: a) the absolute value of the simple correlation of the IV an d DV is smaller than the beta weight for the IV, or b) the signs of the simple correl ation and the beta weight are opposite. Both of these criteria were met for PAIN as t he IV on ACTIVITY. PAIN was negatively correlated with ACTIVITY (r = 0.014, = 0.209). Indirect and total effects of independent variables. The total effects of the eight KSI on FATIGUE in model 2 were examined. Thre e were significant: a) SELF-EFFICACY ( t = -2.885), b) LONELINESS ( t = 6.209), and c) HEALTH ( t = 8.876). Next the total effects of the eight KSI on PA IN were examined. As with FATIGUE, SELF-EFFICACY, LONELINESS and HEALTH all h ad strong effects on PAIN. However, the strongest total effects of KSI o n either FATIGUE or PAIN were those of HEALTH on FATIGUE and PAIN.
135 Next the non-standardized and completely standardized matrices in model 2 were examined for their indirect and total e ffects on ACTIVITY; all eight of the indirect effects of the IVs on ACTIVITY were non -significant. However three of the eight total effects of the IVs on ACTIVI TY were significant (SELF EFFICACY, GOALS, and AGE), with AGE having the larg est total effect ( t = 3.817, = 0.169) followed by SELF-EFFICACY ( t = 2.624, = 0.159) then GOALS ( t = 2.272, = 0.132). This change in significance from non-signifi cant indirect effects to significant direct effects provided evi dence that the mediating effects of FATIGUE and PAIN were too small in these data from this population to support the fully mediated model of unpleasant symp toms. One curious finding was that for four of the variable s, the total effect on ACTIVITY was smaller than the indirect effect. The only way this can happe n is for a reversal of signs to occur, causing a direct effect t hat is the largest of all three effects. The four variables were SELF-EFFICACY, LONELINESS, AGE, and HEALTH. SELF-EFFICACY and AGE each had significan t total effects on ACTIVITY. This also provided evidence that the mediati ng effects of FATIGUE and PAIN were too small in these data with this popula tion to support the full mediational model depicted by Lenz et al. (1997) in the theory of unpleasant symptoms. Squared multiple correlations. Model 2 was the better fitting model of the two models for the theory of unpleasant symptoms. The squared multiple correlations amongst the Y and X variables were checked. These results were essentially the same as found in model 1, with pain subsc ale 2 (mental pain)
136 explaining the least amount of variance in its latent variable ( R 2 = 29.3%) and metabolic equivalents per min per week explaining the most ( R 2 = 98.3%). No major differences were noted for the X variables from those found in model 1, with all the R 2 being greater than 0.700 with the same two exception s, loneliness subscale 1 and social support for exercise subscale 3. For t he entire SEM, PAIN had the most variance explained ( R 2 = 57.3%).and FATIGUE had the second most amount of variance explained ( R 2 = 40.3%). The R 2 for ACTIVITY had a larger change than anticipated between model 1 and mo del 2. As seen in Figure 15, the R 2 went from 3% to 16% between model 1 and model 2. P art of this unusual increase in R 2 perhaps is explained by the inflated effect of PAIN d ue to the presence of a suppressor variable. Without further testing to isolate the specific suppressor variable, it is difficult to interpret. This finding warrants further research. 0.418 0.561 0.029 0.403 0.573 0.1610 0.1 0.2 0.3 0.4 0.5 0.6 0.7 FATIGUE PAIN ACTIVITY R2 Model 1 R2 Model 2 Figure 15. Squared Multiple Correlations.
137 Model Modifications Based upon the methods described by Diamantopoulos and S iguaw (2005), model modifications were examined as a way to further improve an already well-fitting model (model 2). Model modific ations were undertaken only if they were theoretically driven, not purely data driv en. As noted by Diamantopoulos and Siguaw, data driven modification s capitalize too much on chance. Model Two Diagnostics Focusing first on improving the model fit as suggested, the standardized residual statistics and model indices were examined. Of all the elements in the residual covariance matrix, the stem-leaf plot showed 11 data elements with absolute values greater than 4.00. The majority of th e residuals were clustered between -2 and + 2. Of those larger residuals, 7 were positive and 4 were negative. The residuals ranged from -6.6 to 5.8. Larg e positive residuals indicate the need for adding paths to correct underfitting of t he model, and large negative residuals indicate the need for eliminating paths to co rrect overfitting of the model (Diamantopoulos & Siguaw, 2005). Next the Q plot of the normal probability of the r esiduals was examined. It showed a slight shallow departure from the expected 45 degree angle with nonlinearity on one end (as expected from the univariate analysis). Model 2 modification indices (MI) and standardized expe cted parameter changes (SEPC) were examined next. A modification inde x reflects the potential decline in 2 value if a previously fixed parameter is freed to be estimated.
138 Modification indices greater than 3.84 ( df = 1, = .05) are considered large. Among these data, the largest modification index for the r Y s was that for adding path from FATIGUE to the pain subscale 2 (MI = 24.970 SEPC = 0.109. The largest modification index for the r X s was that for expectations subscale 3 to SOCIAL SUPPORT (MI = 20.870, SEPC = .139). Adding a beta path from FATIGUE to PAIN and one from PAIN to FATIGUE would change the 2 value by 27.797 each, with an SEPC of 0.028. Modifications Made The addition (freeing) of model parameters was consider ed. The largest MI was that for adding paths between FATIGUE and PAI N. This made sense theoretically according to the theory of unpleasant symp toms. Because one of the stipulations of SEM is to have not have any non-re cursive paths, both of these alterations could not be done simultaneously. The refore, each path was added separately. Results from Modifications Freeing the path from PAIN to FATIGUE did alter th e model ( 2 = 28.665; df = 1); likewise freeing the path from FATI GUE to PAIN altered the model with the same results ( 2 = 28.665; df = 1). Table 12 provides the details of the modification results. Because of the minimal difference in the goodness of fit indices, the modifications were not retained.
139 Table 12 Summary of Goodness of Fit Indices for Modified Models Model Parameter 2 df RMSEA GFI AGI SRMR Model 2 341.520** 159 0.0493 0.938 0.901 0.0548 Pain to Fatigue 312.855** 0.0451 0.943 0.909 0.037 8 2 a 28.665 1 Fatigue to Pain 312.85 0.0451 0.943 0.909 0.0378 2 a 28.165 1 Note a 2 = Change in 2 from model 2. RMSEA = Root mean square error of approximation; GFI = Goodness of fit index; AGFI = Ad justed goodness of fit index; RMR = Root mean residual; PGFI = Parsimony goo dness of fit index p < .05 ** p < .001
140 Model Cross-Validation Comparisons for ECVI are made amongst the saturated, in dependent, and estimated models of the same overall model, not across est imated models. Having a smaller value for the estimated model is desi rable, but this was not true in this case (ECVI = 1.137; ECVI saturated = 1.031; ECVI independence = 15.494). Because data were collected from only one sampling of the population, further validation was not feasible at this point in time. Power Analysis Post hoc power analysis was done according to the syntax provided by McCallum et al. (1996). See appendix R for the SPSS syntax used to calculate the power. This power analysis syntax used specified condit ions of alpha = 0.05, RMSEA of null hypothesis = 0.05; RMSEA of alternate h ypothesis = 0.08, df, and sample size to calculate the post-hoc power. This power w as the power to reject the H 0 given that the H 0 is false. For this study, the power to reject the H 0 given that the H 0 was false for the structural model with 159 degrees of freedom was 1.00. Thus, the probability that the incorrect H0 wou ld be rejected was of ample size. Prior to the study, different sets of guideline s had been used to project the needed sample sizes. A minimum of 400 participants was n eeded and at least 500 were sought. The final number of participants afte r listwise deletion and exclusion of inappropriate data was 463. Appendix R p rovides the syntax used to show that the sample size needed to reach power 0.8 0 at alpha 0.05 was 125 participants; this syntax was based upon the non-centralit y parameter delta
141 calculated to be equal to 50. Thus this study was well-p owered and rejection of the 2 was expected based upon the excessive sample size. Summary This chapter focused on the results of the research and s ummarized the data. Preliminary analyses including assessment of data q uality for outliers and normality, bivariate relationships, and measurement m odels were done. Problems noted with the original indicator-construct lin ks were discussed. Models 1 and 2 were revised after the indicator-construct lin ks were re-arranged. After these changes, hypothesis testing was done. Of the two rev ised models, model 2 had the best evidence of fit. Although modifications were attempted for a third model, and even though the results were better, those results were so minimal overall that the decision was made to not retain the t hird model. Implications for these findings are discussed in depth in the following ch apter, as are plans for future research, suggestions for others, and a brief discussi on of lessons learned.
142 CHAPTER FIVE Discussion This chapter discusses key findings and possible explanati ons associated with those findings, limitations to the study, implicat ions for community health, directions for future research, and lessons learned. Each aim and each research question for this research is discussed sequentially. Findin gs that are different from established findings in the literature are highli ghted in the discussions. Aim 1: Testing the Theory of Unpleasant Symptoms The first aim of this research study was to test the theo ry of unpleasant symptoms as described by Lenz et al. (1997) and to ascert ain whether the implied model would be reproducible in the data from the collegiate women of ages 18 to 25. As this appeared to be the first time t hat the theory of unpleasant symptoms has been tested using structural equation model ing, there were no prior studies with which to compare results. Given that physical activity was used as the performance outcome, it was hoped that the use o f SEM would give further credence to the theory of unpleasant symptoms As indicated by the data, model 1 (the original theory of unpleasant symptoms) adequately reproduced the implied covariance matrix among collegiate women of a ges 18 to 25. However, it only explained three percent of the variance in activi ty.
143 As noted in chapter 2, there were several concepts wit hin the theory of unpleasant symptoms that had been tested previously via S EM, such as selfefficacy, self-regulation, and outcome expectations for t he psychological domain, social support for the situational domain, and age and health status for the physiological domain. Pain had been studied using SEM a s well. It was the combination of these concepts in the theory of unpleasan t symptoms that was unique for this study. Fatigue. The relation of fatigue to physical activity ( = -.178) in model 1 was not surprising. These results were consistent with the conceptual model and were consistent with Garber and Friedman (2003) who fo und that fatigue was inversely correlated with physical activity among patien ts with idiopathic ParkinsonÂ’s disease. The finding of high levels of fatigue (11.8% passed th e screening threshold for fatigue) was surprising, given that the f atigue questionnaire was a chronic fatigue questionnaire geared to physical fatigue as well as emotional fatigue. One possible explanation is that the chronic f atigue scale had a time reference of fatigue within the past month. Another possible explanation is that although the chronic fatigue scale inquired about fatig ue in the past month, there may have been some crossover into thinking about fatigu e that resulted from exercise. This finding of greater than normal fatigue among collegiate females ages 18 to 25 is a finding that warrants further resear ch. Another significant finding for model 1 was that fati gue was affected by self-efficacy for exercise ( = -.174). One plausible explanation for the
144 relationship between self-efficacy and fatigue is that by nature of its definition, self-efficacy is the confidence in oneÂ’s ability to perfo rm despite barriers. This interrelationship between symptom expression and psychol ogical factors highlights the integrated mind/body system and warrant s further research in young adult women. Loneliness also had a strong association with fatigue ( = 0.330). This relationship was puzzling. Documentation of a relation ship between fatigue and loneliness had not been found in the literature. One possible explanation for the direct relationship between FATIGUE and LONELINESS is that the relationship is a spurious one. And finally, as expected, there was an inverse effect of HEALTH on FATIGUE ( = 0.408). Pain. The unusual finding of a positive but tiny effect of PAIN on ACTIVITY was not expected; however since the effect was n ot significantly different from zero, the finding of a positive effect was deemed the function of sampling error. LONELINESS also had a strong association with PAIN ( = 0.447). This relationship was puzzling. This relationshi p between PAIN and LONELINESS had not been anticipated. As with FATIGUE, one possible explanation for the direct relationship between LONE LINESS and PAIN is that the relationship is a spurious one. And finally, as exp ected, there was an inverse effect of HEALTH on PAIN ( = 0.496). This is consistent with other research. Aim 2: Testing the Alternative Model 2 Model 2 was a partially mediated model that fit sign ificantly better than model 1. This model explained over 16% of the varian ce in activity. As expected,
145 there were significant relationships between SELF-EFFI CACY and ACTIVITY ( = 0.180) and between GOALS and ACTIVITY ( = 0.110); however, the relationship between EXPECTATIONS and ACTIVITY was no t as strong ( = 0.024). As expected from the literature, there was an inverse effect of AGE on ACTIVITY ( = -.179). According to Krumholz et al. (2005) there is a negative relationship between age and activity. The decline in physical activity starts in high school and worsens during young adulthood, which is just before the decade of highest weight gain for women. However, giv en that this study was done among a restricted age range (age 18 to 25), fin ding this was the most significant relationship with ACTIVITY was a surprise. O ne possible explanation is that the older study participants are more likely to be in graduate classes or to be employed, which would leave them less time to exer cise. As noted in chapter 4, there were three major findin gs that complicated the interpretation of these data. First, the pain eff ect was puzzling. Second, the pain effect provided evidence for a possible suppressor variable when controlling for other variables that were moderately correlated w ith PAIN. Third, the reversal of directional signs between indirect and direct effects i n four variables was explained by a larger direct than total effect for th ose variables. The presence of a larger direct effect than a total effect shows that t he mediated effect, the indirect effect, is too small to be of consequence. Thu s the question came as to whether the mediators (FATIGUE and PAIN for UNPLEASA NT SYMPTOMS) are even needed. The psychological domain (SELF-EFFICACY, EXPECTATIONS,
146 and GOALS) can be expanded to include SOCIAL SUPPORT, HEALTH, and LONELINESS without detriment to the model. Implications for Use of the Theory of Unpleasant Sympt oms The model for the theory of unpleasant symptoms as de picted by Lenz et al. (1995; 1997) showed unpleasant symptoms as fully me diating the relationships between psychological, physiological, and sit uational factors. The data for this study fit the model adequately, and hig hlighted the importance of unpleasant symptoms in this age group. The direct effects of LONELINESS ( =.330) on FATIGUE and HEALTH ( = -.496) were moderately strong; in particular, the direct effect of LONELINESS on FATIGUE was interesting for this population. Likewise, LONELINESS and HEALTH also had m oderately strong indirect effects on PAIN ( = .447 and = -.496 respectively). SELF-EFFICACY had a significant and strong direct effect on FATIGUE ( t = -3.003, = -0.174) and on PAIN ( t = -2.139, = -0.162). Again, these findings highlight the impor tance of the mind/body integration. However, the model 1 as a whole only explained 3% of the variance in ACTIVITY. In stark contrast, model 2 as a whole explaine d 16% of the variance in ACTIVITY. Allowing the other variables to bypass t he unpleasant symptoms of FATIGUE and PAIN by having direct effects on ACTIVITY substantially improved the fit of the model. Another point worth noting is that the definitions fo r the factors in the theory of unpleasant symptoms were sometimes ambiguous ; for example, social support is listed as both a psychological and a situational factor by Lenz et al.
147 (1995). In this study, SOCIAL SUPPORT only explained a small portion of the model (indirect effect = 0.014). In summary, the theory of unpleasant symptoms (model 1) was adequate when tested in this population. However, the direct ef fects of PAIN and FATIGUE on ACTIVITY were non-significant in model 1. After con trolling for the psychological, situational and physiological variables in m odel 2, unpleasant symptoms still did not influence exercise activity. The e ffects of SELFEFFICACY, LONELINESS, and HEALTH were significant on both FATIGUE and PAIN. However, the evidence from model 2 showed that there were nonsignificant indirect effects of all eight exogenous varia bles on ACTIVITY via unpleasant symptoms, yet when allowed to bypass the unpl easant symptoms, the direct effects on ACTIVITY were significant for SELF -EFFICACY, GOALS, and AGE. Thus it appears that the social cognitive mode l of exercise as described by Bandura (1997; 2004) is a more parsimoniou s model for explaining individual differences in exercise, at least in this popu lation and age range. Implications for Nursing Intervention A complex model of psychological, situational, and phy siological predictors of exercise in the presence of unpleasant sympt oms of pain and loneliness was tested among collegiate women of ages 18 to 25. In addition to studying more established links among the psychological variables and exercise, this study also examined the previously unexplored media ting role of unpleasant symptoms as posited by Lenz et al. (1995; 1997).
148 As the data showed, the theory of unpleasant symptom s fit the observed data but explained little (3%) of the variance in ex ercise activity. For the first time the relationships of fatigue and of pain to exercise w ere documented in this population. It has implications for those working with y oung adults in sports, schools, and in healthcare. First of all, the prevalence of fatigue and pain needs to be acknowledged even among active college women. Re cent evidence provided by Rimes et al. (2007) reveals that among ad olescents, the point prevalence rate for fatigue was 34%; this study used the same measure of fatigue (Â‘over the last month, have you been feelin g much more tired and worn out than usual?) as was used by Rimes et al. It is impo rtant to note that this rate did not include those with chronic fatigue or with clin ical evidence of chronic fatigue syndrome. The finding that the psychological variables (SELF-EFF ICACY, EXPECTATIONS, and GOALS) partially mediated the rel ationships of the other variables with ACTIVITY is not surprising given the comp lex integration of the mind/body system. However, this has strong implications fo r healthcare providers who are using exercise prescriptions as part of their trea tment plans. Incorporating interventions to increase self-efficacy for exercise will facilitate the promotion of exercise as a treatment modality for fati gue. Incorporating interventions to increase self-efficacy for exercise will also facilitate the promotion of exercise for any number of conditions such as obesity (see Fabricatore, 2007). The findings from this research also have implications for public policy. Fuemmeler, Baffi, Masse, Atienza, and Evans (2007) s urveyed 1139 participants
149 in the US in 2004 and found that women favor requir ing healthcare companies to reimburse for obesity treatment and preventive program s. Suggestions for policy changes included tax incentives to employers to provide e xercise facilities. This current research among collegiate women provides evidence that the exercise outcome is affected by psychological, situational, and phy siological factors as presented in the theory of unpleasant symptoms. Rather than merely providing exercise facilities, an implication from this research is th at all the factors need to be considered simultaneously. Limitations to the Study There are a number of limitations to this current stu dy. First, this was a cross-sectional design and causation cannot be established. Second, the Internet-based sampling method only reached those stu dents who elected to read their emails from strangers. Although it allowed for reaching a large number of participants within a very narrow timeframe, the Internet-based sampling method was fraught with problems. Even though the Ult imate Survey system allowed only one response per participant, there was n o way to validate who the respondents were. The entire survey was self-report, an d due to the nature of the online survey, some of the questions had to be altere d in their format from the original survey authorsÂ’ designs. For instance, the IPAQ questionnaire is based upon a Â‘fill in the blankÂ’ question format. Although the Â‘fill in the blankÂ’ or open response format was allowed in the Internet survey, t rial runs with the Ultimate Survey revealed that responses from the open format were not exported directly into SPSS, and required coding of responses o ne by one. Because this
150 was not feasible, the decision was made to offer the num ber of hours and the number of minutes for exercise as a drop-down menu. Another limitation was the failure to include some i mportant variables such as BMI, the existence of co-morbidities (either mental or physical), medication usage, sleep patterns, and hours spent in class or work. S elf-report of weight and height would have provided the needed parameter s to calculate BMI. Screening for mental co-morbidities such as anxiety or depression, both of which are known to impact fatigue levels (Rimes et al., 2007) would enhance the study. Other limitations were present as well. The ethnic pr ofile of the respondents did not closely reflect that of the universit y students as expected, and the ethnic profile did not match that of the surro unding community. This may limit generalizability of the findings. Using a strati fied sampling method is one way to remedy this in future studies. Another limitation of this current study was the origi nal selection of manifest indicators for the latent variables. Data were obtained from all the participants and when bivariate correlational analyses w ere done, there was not the needed magnitude of correlation among indicators f or the same latent variable. Some of the data that were collected were n ot used as a result. For instance, the situational fatigue scale as an indicator o f anticipated fatigue for the physiological factor was not reliable in this sample, wi th a CronbachÂ’s alpha of only .33. Had a more comprehensive pilot study been d one, some of these problems may have been averted.
151 Directions for Future Research As noted in the limitation section, several variables that could have affected the outcome of physical activity were not colle cted or examined. Future research is warranted to explore the unique findings of this present research in more depth. For instance, the effect of fatigue on exe rcise needs to be studied to ferret out the difference between anticipated versus ch ronic fatigue, the effects of sleep deprivation, antecedent anxiety and/or depression and work schedules. Another direction for research is to explore the phenom enon of pain in this population. Pain did not have a significant effect on exercise for this study. Future research should differentiate between chronic pain, anti cipated pain from exercise, and catastrophizing pain. Future research stud ies could explore methods to enhance self-efficacy for exercise. Extendin g this research to younger adolescents would be warranted, particularly sin ce a strong effect of age on exercise was found. Lessons Learned For those who want to use an Internet-based survey, study measures should be piloted on line to determine if question fo rmats have to be altered. For this study, SPSS was used to randomly select potential pa rticipants from a list of students. Because of the low response rate (9%), a second round of participants was randomly selected by SPSS from the same list of stude nts. It was necessary to double check for duplicity of names in the second rand omized list compared to the first randomized list. Sending duplicate invitati ons to a few participants was
152 averted by weeding out the duplicate names; however, for thousands of names, this required use of valuable time. Another lesson learned (for those budding LISRELites) is that no shortcuts can be taken. When the textbooks such as Diamantopoulos an d Siguaw (2005) mandate that bivariate correlation be done first, it is futile to run the measurement model until the relationships within latent variables have been established. It is quite possible for LISREL to give a reasonable fit of the model to the implied data, and yet inspection of the data reveals oddities s uch as negative variances, negative lambdas, squared multiple correlations greater than one, lambdas greater than one, or correlations that donÂ’t make sense For instance, the correlations among indicators for any given latent vari able should be of sufficient magnitude to warrant being considered indicators of the same concept, and they also need to be in the same direction as the other indi cators in the concept. For example, the scale for loneliness was designed in such a w ay that a higher score indicated higher loneliness (more Â‘badÂ’); when this was paired with social support, in which a larger number was a Â‘goodÂ’ amount of social support, a negative lambda was produced. Because of the reverse cod ing that had to be done (sometimes to reverse the original reverse coding), meticulous notes of all coding were necessary. It was helpful to keep one syntax f ile just for data cleaning and coding purposes, and to keep one syntax file for the actual analyses in SPSS. Another lesson learned is that data are not exported e xactly as intended by Ultimate Survey. For instance, one of the Likerttype scales had points
153 ranging from 0 to 5; when it was imported into SPSS, the discovery was made that Ultimate Survey coded the first response option as 1, the second response option as 2, and so forth. This meant having to recode all the ones into zeros and so on. For the expectations scale, where a smaller numbe r meant higher expectations, Ultimate Survey still coded the first re sponse as 1 (when it should have been 4). Maintaining a code book is essential. The data editor from the original data was never save d; each time the data were needed, the file was opened and all the recodes w ere done at once. This hint spared a lot of grief, as it was necessary to split t he data file for statistical purposes as well as to recode variables several times. F iles were split to obtain a covariance matrix on each subgroup, for instance. Any fil es that were split off were saved under a different filename; the original data editor produced by data cleaning and all the recoding was never saved.
154 List of References Ahmed, N. U., Smith, G. L., Flores, A. M., Pamies, R. J., Mason, H. R., Woods, K. F., et al. (2005). Racial/ethnic disparity and pred ictors of leisure-time physical activity among U.S. men. Ethnicity & Disease, 15 (1), 40-52. Allen, J., Markovitz, J., Jacobs, D. R., Jr., & Knox, S. S. (2001). Social support and health behavior in hostile black and white men and women in CARDIA. Coronary Artery Risk Development in Young Adu lts. Psychosomatic Medicine, 63 (4), 609-618. Allison, K. R., Dwyer, J. J., Goldenberg, E., Fein, A ., Yoshida, K. K., & Boutilier, M. (2005). Male adolescents' reasons for participating in physical activity, barriers to participation, and suggestions for increasing participation. Adolescence, 40 (157), 155-170. Allison, K. R., Dwyer, J. J., & Makin, S. (1999). Perce ived barriers to physical activity among high school students. Preventive Medicine, 28 (6), 608-615. American College of Sports Medicine. (1995). American C ollege of Sports Medicine position stand. Osteoporosis and exercise. Medicine and Science in Sports and Exercise, 27 (4), i-vii. Anderson, C. (2003). When more is better: number of motives and reasons for quitting as correlates of physical activity in women. Health Education Research, 18 (5), 525-537.
155 Anderson, E. S., Wojcik, J. R., Winett, R. A., & Willi ams, D. M. (2006). Socialcognitive determinants of physical activity: the influe nce of social support, self-efficacy, outcome expectations, and self-regulatio n among participants in a church-based health promotion study. Health Psychology: Official Journal of the Division of Health Psychology, Am erican Psychological Association, 25 (4), 510-520. Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in practice: A review and recommended two-step approach. Psychological Bulletin, 103 (3), 411-423. Bandura, A. (1986). Social foundations of though and action: A social cogniti ve theory Upper Saddle, NJ: Prentice Hall. Bandura, A. (1994). Self-efficacy. In H. Friedman (E d.), Encyclopedia of mental health (Vol. 4, pp. 71-81). San Diego: Academic Press, 1998. Bandura, A. (1997). The exercise of control New York: Freeman. Bandura, A. (2004). Health promotion by social cognit ive means. Health Education and Behavior, 31 (2), 143-164. Bandura, A., Adams, N. E., & Beyer, J. (1977). Cognit ive processes: Mediating behavioral change. Journal of Personality and Social Psychology, 35 (3), 125-139. Bandura, A., Caprara, G. V., Barbaranelli, C., Gerb ino, M., & Pastorelli, C. (2003). Role of affective self-regulatory efficacy in d iverse spheres of psychosocial functioning. Child Development, 74 (3), 769-782.
156 Baron, R. M., & Kenny, D. A. (1986). The moderatormediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51 (6), 11731182. Barrera, M., Jr., Toobert, D. J., Angell, K. L., Glasg ow, R. E., & Mackinnon, D. P. (2006). Social support and social-ecological resources as m ediators of lifestyle intervention effects for type II diabetes. Journal of Health Psychology, 11 (3), 483-495. Beckstead, J. W. (2002a). Confirmatory factor analysis of the Maslach burnout inventory among Florida nurses. International Journal of Nursing Studies, 39 (8), 785-792. Beckstead, J. W. (2002b). Modeling attitudinal antecede nts of nurses' decisions to report impaired colleagues. Western Journal of Nursing Research, 24 (5), 537-551. Beckstead, J. W. (2005). Reporting peer wrongdoing in the healthcare profession: The role of incompetence and substance abuse in formation. International Journal of Nursing Studies, 42 (3), 325-331. Beckstead, J.W., Yang, C.-Y., & Lengacher, C.A. (2006). Assessing crosscultural validity of scales: A methodological review an d illustrative example. International Journal of Nursing Studies (in press). Bentler, P. M., & Chou, C.-P. (1987). Practical issues in structural modeling. Sociological Methods & Research, 16 (1), 78-117.
157 Bigal, M. E., Liberman, J. N., & Lipton, R. B. (2006 ). Obesity and migraine: A population study. Neurology, 66 (4), 545-550. Borer, K. T. (2005). Physical activity in the preventio n and amelioration of osteoporosis in women: Interaction of mechanical, hormona l and dietary factors. Sports Medicine 35 (9), 779-830. Boyd, C. J., Frey, M. A., & Aaronson, L. S. (1988). Structural equation models and nursing research: Part I. Nursing Research, 37 (4), 249-252. Bulwer, B. E. (2004). Sedentary lifestyles, physical act ivity, and cardiovascular disease: From research to practice. Critical Pathways in Cardiology, 3 (4), 184-193. Callaghan, D. M. (2006). Basic conditioning factors' in fluences on adolescents' healthy behaviors, self-efficacy, and self-care. Issues in Comprehensive Pediatric Nursing, 29 191-204. Centers for Disease Control and, P. (2005). Trends in l eisure-time physical inactivity by age, sex, and race/ethnicity--United Sta tes, 1994-2004. MMWR. Morbidity and Mortality Weekly Report, 54 (39), 991-994. Centers for Disease Control and Prevention. (2006). Physical activity for everyone: Recommendations Retrieved October 26, 2006 from http://www.cdc.gov/nccdphp/dnpa/physical/recommendations/i ndex.htm Cerin, E., Taylor, L. M., Leslie, E., & Owen, N. (20 06). Small-scale randomized controlled trials need more powerful methods of mediat ional analysis than the Baron-Kenny method. Journal of Clinical Epidemiology, 59 (5), 457464.
158 Chalder, T., Berelowitz, G., Pawlikowska, T., Watts, L. Wessely, S., Wright, D., et al. (1993). Development of a fatigue scale. Journal of Psychosomatic Research, 37 (2), 147-153. Chatrath, R., Shenoy, R., Serratto, M., & Thoele, D G. (2002). Physical fitness of urban American children. Pediatric Cardiology, 23 (6), 608-612. Chen, M. Y., James, K., & Wang, E. K. (2007). Compar ison of health-promoting behavior between Taiwanese and American adolescents: A crosssectional questionnaire survey. International Journal of Nursing Studies, 44 (1), 59-69. Cobb, S. E. (2005). A qualitative analysis of the bicultural aspects of physical activity in adolescents and young adults. University of South Florida. Cobb, S. E. (2006). Pilot data from the College Exe rcise Project. University of South Florida. Cook, A. J., Brawer, P. A., & Vowles, K. E. (2006). Th e fear-avoidance model of chronic pain: validation and age analysis using structural equation modeling. Pain, 121 (3), 195-206. Correa-de-Araujo, R., Stevens, B., Moy, E., Nilasena, D., Chesley, F., & McDermott, K. (2006). Gender differences across racial an d ethnic groups in the quality of care for acute myocardial infarction and heart failure associated with comorbidities. Women's Health Issues: Official Publication of the Jacobs Institute of Women's Health, 16 (2), 44-55.
159 Corwin, E. J., Brownstead, J., Barton, N., Heckard, S., & Morin, K. (2005). The impact of fatigue on the development of postpartum de pression. Journal of Obstetric, Gynecologic, and Neonatal Nursing: JOGNN / NA ACOG, 34 (5), 577-586. Council on Sports Medicine and Fitness, & Council on Scho ol Health. (2006). Active healthy living: Prevention of childhood obesity through increased physical activity. Pediatrics, 117 (5), 1834-1842. Courneya, K. S., Plotnikoff, R. C., Hotz, S. B., & B irkett, N. J. (2001). Predicting exercise stage transitions over two consecutive 6-month peri ods: A test of the theory of planned behaviour in a population-base d sample. British Journal Health Psychology, 6 (Pt 2), 135-150. Craig, C. L., Marshall, A. L., Sjostrom, M., Bauman, A. E., Booth, M. L., Ainsworth, B. E., et al. (2003). International physica l activity questionnaire: 12-country reliability and validity. Medicine and Science in Sports and Exercise, 35 (8), 1381-1395. Crane, P. B. (2005). Fatigue and physical activity in older women after myocardial infarction. Heart & Lung: The Journal of Critical Care, 34 (1), 30-38. Curran, P. J., West, S. G., & Finch, J. F. (1996). Th e robustness of test statistics to nonnormality and specification error in confirmator y factor analysis. Psychological Methods, 1 (1), 16-29.
160 De Bourdeaudhuij, I., Lefevre, J., Deforche, B., Wijn daele, K., Matton, L., & Philippaerts, R. (2005). Physical activity and psychosocial correlates in normal weight and overweight 11 to 19 year olds. Obesity Research, 13 (6), 1097-1105. De Bourdeaudhuij, I., Philippaerts, R., Crombez, G., Matton, L., Wijndaele, K., Balduck, A. L., et al. (2005). Stages of change for phy sical activity in a community sample of adolescents. Health Education Research, 20 (3), 357-366. De Bourdeaudhuij, I., & Sallis, J. (2002). Relative co ntribution of psychosocial variables to the explanation of physical activity in th ree population-based adult samples. Preventive Medicine, 34 (2), 279-288. Deforche, B., De Bourdeaudhuij, I., Tanghe, A., Hill s, A. P., & De Bode, P. (2004). Changes in physical activity and psychosocial deter minants of physical activity in children and adolescents treated for obesity. Patient Education and Counseling, 55 (3), 407-415. Diamantopoulos, A., & Siguaw, J. A. (2005). Introducing LISREL Thousand Oaks, California: SAGE Publications. Dillman, D. A. (2007). Mail and Internet surveys: The tailored design method (2nd ed.). New Jersey: John Wiley & Sons. Dishman, R. K., Motl, R. W., Sallis, J. F., Dunn, A. L ., Birnbaum, A. S., Welk, G. J., et al. (2005). Self-management strategies mediate self-efficacy and physical activity. American Journal of Preventive Medicine, 29 (1), 10-18.
161 DuToit, S., DuToit, M., Mels, G., & Cheng, Y. (2005) LISREL for Windows: PRELIS user's guide NA: SSI: Scientific Software International. Elgan, C., & Fridlund, B. (2006). Bone mineral densi ty in relation to body mass index among young women: A prospective cohort study. International Journal of Nursing Studies, 43 (6), 663-672. Enthoven, P., Skargren, E., Carstensen, J., & Oberg, B. (2006). Predictive factors for 1-year and 5-year outcome for disability i n a working population of patients with low back pain treated in primary care. Pain, 122 (1-2), 137144. Fabricatore, A. N. (2007). Behavior therapy and cogni tive-behavioral therapy of obesity: Is there a difference? Journal of the American Dietetic Association, 107 (1), 92-99. Fauchald, S. K., & Smith, D. (2005). Transdisciplinary research partnerships: Making research happen! Nursing Economic$, 23 (3), 131-5, 107. Fuemmeler, B. F., Baffi, C., Masse, L. C., Atienza, A A., & Evans, W. D. (2007). Employer and healthcare policy interventions aimed at adult obesity. American Journal of Preventive Medicine, 32 (1), 44-51. Garber, C. E., & Friedman, J. H. (2003). Effects of fa tigue on physical activity and function in patients with ParkinsonÂ’s disease. Neurology, 60 (7), 11191124. Grubbs, L., & Carter, J. (2002). The relationship of perceived benefits and barriers to reported exercise behaviors in college under graduates. Family & Community Health, 25 (2), 76-84.
162 Gyurcsik, N. C., Bray, S. R., & Brittain, D. R. (2004) Coping with barriers to vigorous physical activity during transition to universit y. Family & Community Health, 27 (2), 130-142. Harris, A. H., Cronkite, R., & Moos, R. (2006). Physical activity, exercise coping, and depression in a 10-year cohort study of depressed pat ients. Journal of Affective Disorders, 93 (1-3), 79-85. Harris, K. M., Gordon-Larsen, P., Chantala, K., & Udr y, J. R. (2006). Longitudinal trends in race/ethnic disparities in leading health indi cators from adolescence to young adulthood. Archives of Pediatrics & Adolescent Medicine, 160 (1), 74-81. Hartshorne, T. S. (1993). Psychometric properties and con firmatory factor analysis of the UCLA loneliness scale. Journal of Personality Assessment, 61 (1), 182-195. Havranek, E. P., Lapuerta, P., Simon, T. A., L'Itali en, G., Block, A. J., & Rouleau, J. L. (2001). A health perception score predicts cardiac ev ents in patients with heart failure: Results from the IMPRESS trial. Journal of Cardiac Failure, 7 (2), 153-157. Hays, R. D., & DiMatteo, M. R. (1987). A short-form m easure of loneliness. Journal of Personality Assessment, 51 (1), 69-81. Holm, P., Sattler, A., & Fregosi, R. F. (2004). Endu rance training of respiratory muscles improves cycling performance in fit young cyclists. BMC Physiology, 4 9.
163 Iannuzzi, A., Licenziati, M. R., Acampora, C., Renis, M., Agrusta, M., Romano, L., et al. (2006). Carotid artery stiffness in obese ch ildren with the metabolic syndrome. The American Journal of Cardiology, 97 (4), 528-531. IPAQ. (2005, November). Guidelines for data processing and analysis of the International Physical Activity Questionnaire (IPAQ). Retrieved June 20, 2006, from http://www.ipaq.ki.se Jaccard, J., & Wan, C. K. (1996). LISREL approaches to interaction effects in multiple regression Thousand Oaks, CA: SAGE Publications. Jia, H., & Lubetkin, E. I. (2005). The impact of obesit y on health-related qualityof-life in the general adult US population. Journal of Public Health, 27 (2), 156-164. Joreskog, K. S., & Sorbom, D. (2005). LISREL 8.72S. Available online at http://www.ssicentral.com Kasa-Vubu, J. Z., Ye, W., Borer, K. T., Rosenthal, A., & Meckmongkol, T. (2006). Twenty-four hour growth hormone and leptin secretion in active post-pubertal adolescent girls: Impact of fitness, fatness and a ge at menarche. The Journal of Clinical Endocrinology and Metabolism 91 (10), 3535-3540. Kear, M. (2000). Concept analysis of self-efficacy. Graduate Research in Nursing [electronic resource]. Retrieved January 26, 2007, fro m http://graduateresearch.com/Kear.htm Kelloway, E. K. (1998). Using LISREL for structural equation modeling: A researcher's guide Thousand Oaks, California: SAGE Publications.
164 Kerns, R. D., Turk, D. C., & Rudy, T. E. (1985). The West Haven-Yale Multidimensional Pain Inventory (WHYMPI). Pain, 23 (4), 345-356. Killeen, C. (1998). Loneliness: An epidemic in modern society. Journal of Advanced Nursing, 28 (4), 762-770. Kim, K.S. (2005). The relation among fit indexes, po wer, and sample size in structural equation modeling. Structural Equation Modeling, 12 (3), 368390. Kim, Y. H. (2006). Application of the transtheoretical model to identify psychological constructs influencing exercise behavior: A qu estionnaire survey. International Journal of Nursing Studies (article in press, available online at http://dx.doi.org/10.1016/j.ijnurstu.2006.03.008 ) Krumholz, H. M., Peterson, E. D., Ayanian, J. Z., Ch in, M. H., DeBusk, R. F., Goldman, L., et al. (2005). Report of the National Heart, Lung, and Blood Institute Working Group on outcomes research in cardiova scular disease. Circulation, 111 (23), 3158-3166. LaMonte, M. J., Barlow, C. E., Jurca, R., Kampert, J. B., Church, T. S., & Blair, S. N. (2005). Cardiorespiratory fitness is inversely associate d with the incidence of metabolic syndrome: A prospective study of m en and women. Circulation, 112 (4), 505-512. Landers, D. M. (2006). The influence of exercise on me ntal health. The President's Council on Physical Fitness and Sport's Research Digest, Series 2 (12). Available online at http://www.fitness.gov/mentalhealth/mentalhealth.htm
165 Larson, C. O. (2002). Use of the SF-12 instrument for measuring the health of homeless persons. Health Services Research, 37 (3), 733-750. Lee, Y. S., & Laffrey, S. C. (2006). Predictors of ph ysical activity in older adults with borderline hypertension. Nursing Research, 55 (2), 110-120. Leece, P., Bhandari, M., Sprague, S., Swiontkowski, M F., Schemitsch, E. H., & Tornetta, P. (2006). Does flattery work? A comparison o f 2 different cover letters for an international survey of orthopedic surge ons. Canadian Journal of Surgery, 49 (2), 90-95. Leino-Arjas, P., Solovieva, S., Riihimaki, H., Kirjon en, J., & Telama, R. (2004). Leisure time physical activity and strenuousness of work as predictors of physical functioning: a 28 year follow up of a cohort o f industrial employees. Occupational and Environmental Medicine, 61 (12), 10321038. Lenz, E. R., Pugh, L. C., Milligan, R. A., Gift, A. & Suppe, F. (1997). The middlerange theory of unpleasant symptoms: An update. ANS. Advances in Nursing Science, 19 (3), 14-27. Lenz, E. R., Suppe, F., Gift, A. G., Pugh, L. C., & Milligan, R. A. (1995). Collaborative development of middle-range nursing th eories: Toward a theory of unpleasant symptoms. ANS. Advances in Nursing Science, 17 (3), 1-13. Lethem, J., Slade, P. D., Troup, J. D., & Bentley, G. (1983). Outline of a FearAvoidance model of exaggerated pain perception--I. Behaviour research and therapy, 21 (4), 401-408.
166 Liu, H. E. (2006). Fatigue and associated factors in hem odialysis patients in Taiwan. Research in Nursing & Health, 29 (1), 40-50. MacCallum, R. C., Browne, M. W., & Sugawara, H. M. ( 1996). Power analysis and determination of sample size for covariance structur e modeling. Psychological Methods, 1 (2), 130-149. MacCallum, R. C., Roznowski, M., & Necowitz, L. B. (199 2). Model modifications in covariance structure analysis: the problem of capitaliz ation on chance. Psychological Bulletin, 111 (3), 490-504. MacCallum, R. C., Widaman, K. F., Zhang, S., & Hong, S. (1999). Sample size in factor analysis. Psychological Bulletin, 4 (1), 84-99. MacKinnon, D. P., Krull, J. L., & Lockwood, C. M. (2000 ). Equivalence of the mediation, confounding and suppression effect. Prevention Science, 1 (4), 173-181. Mahon, N. E., Yarcheski, A., & Yarcheski, T. J. (1998). S ocial support and positive health practices in young adults. Loneliness as a mediating variable. Clinical Nursing Research, 7 (3), 292-308. Mahon, N. E., Yarcheski, A., & Yarcheski, T. J. (2001). M ental health variables and positive health practices in early adolescents. Psychological Reports, 88 (3 Pt 2), 1023-1030. Mahon, N. E., Yarcheski, A., & Yarcheski, T. J. (2004). S ocial support and positive health practices in early adolescents: a test o f mediating variables. Clinical Nursing Research, 13 (3), 216-236.
167 Mahon, N. E., Yarcheski, T. J., & Yarcheski, A. (1995). V alidation of the revised UCLA Loneliness Scale for adolescents. Research in Nursing and Health, 18 (3), 263-270. Marquez, D. X., & McAuley, E. (2006). Social cognitive correlates of leisure time physical activity among Latinos. Journal of Behavioral Medicine 29(3), 281-289. Marsh, H. W., Balla, J. R., & McDonald, R. P. (1988). Goodness of fit indexes in confirmatory factor analysis: The effect of sample size. Psychological Bulletin, 103 (3), 391-410. McCann, K., & Boore, J. R. (2000). Fatigue in persons with renal failure who require maintenance haemodialysis. Journal of Advanced Nursing, 32 (5), 1132-1142. McDevitt, J., Snyder, M., Miller, A., & Wilbur, J. (20 06). Perceptions of barriers and benefits to physical activity among outpatients in p sychiatric rehabilitation. Journal of Nursing Scholarship: An official publication of Sigma Theta Tau International Honor Society of Nursin g / Sigma Theta Tau, 38 (1), 50-55. McGavock, J. M., Anderson, T. J., & Lewanczuk, R. Z. (2006) Sedentary lifestyle and antecedents of cardiovascular disease in young adul ts. American Journal of Hypertension: Journal of the American Society o f Hypertension, 19 (7), 701-707.
168 McNeill, L. H., Wyrwich, K. W., Brownson, R. C., Clark, E. M., & Kreuter, M. W. (2006). Individual, social environmental and physical environmental influences on physical activity among black and white adu lts: A structural equation analysis. Annals of Behavioral Medicine, 31 (1), 36-44. Mears, C. J., Taylor, R. R., Jordan, K. M., Binns, H. J. & Pediatric Practice Research, G. (2004). Sociodemographic and symptom correla tes of fatigue in an adolescent primary care sample. The Journal of Adolescent Health: Official publication of the Society for Adolescen t Medicine, 35 (6), 528e.521-526. Melzack, R. (2001). Pain and the neuromatrix in the b rain. Journal Of Dental Education, 65 (12), 1378-1382. Merriam-Webster. (2007). The Merriam-Webster dictionary. Retrieved January 8, 2007, from http://www.m-w.com/dictionary/ Miech, R. A., Kumanyika, S. K., Stettler, N., Link, B. G., Phelan, J. C., & Chang, V. W. (2006). Trends in the association of poverty wit h overweight among US adolescents, 1971-2004. JAMA: The Journal of the American Medical Association, 295 (20), 2385-2393. Mokdad, A. H., Marks, J. S., Stroup, D. F., & Gerberdin g, J. L. (2004). Actual causes of death in the United States, 2000. JAMA: The Journal of the American Medical Association, 291 (10), 1238-1245. Moritz, S. E., Feltz, D. L., Fahrbach, K. R., & Mack, D. E. (2000). The relation of self-efficacy measures to sport performance: A meta-analyt ic review. Research Quarterly Exercise and Sport, 71 (3), 280-294.
169 Morriss, R. K., Wearden, A. J., & Mullis, R. (1998). Ex ploring the validity of the Chalder Fatigue scale in chronic fatigue syndrome. Journal of Psychosomatic Research, 45 (5), 411-417. Motl, R. W., Dishman, R. K., Saunders, R. P., Dowda, M., & Pate, R. R. (2007). Perceptions of physical and social environment variables and self-efficacy as correlates of self-reported physical activity among ado lescent girls. Journal of Pediatric Psychology, 32 (1), 6-12. Motl, R. W., Dishman, R. K., Ward, D. S., Saunders, R P., Dowda, M., Felton, G., et al. (2002). Examining social-cognitive determin ants of intention and physical activity among black and white adolescent girls usi ng structural equation modeling. Health Psychology, 21 (5), 459-467. National Heart Lung and Blood Institute working group (2006). NHLBI working group report: Preventing weight gain in young adults. Retrieved July 17, 2006, from http://www.nhlbi.nih.gov/meetings/workshops/wgt-gain.ht m National Institutes Health. (1998). NIH policy and guidelines on the inclusion of children as participants in research involving human subje cts. Retrieved June 19, 2006, from http://grants.nih.gov/grants/guide/notice-files/not98024.html Nelson, M. C., & Gordon-Larsen, P. (2006). Physical acti vity and sedentary behavior patterns are associated with selected adolescent health risk behaviors. Pediatrics, 117 (4), 1281-1290.
170 Noakes, T. D. (2000). Physiological models to understand exercise fatigue and the adaptations that predict or enhance athletic perfo rmance. Scandinavian Journal of Medicine & Science in Sports, 10 (3), 123-145. O'Brien, B., Evans, M., & White-McDonald, E. (2002). Isolation from "being alive": Coping with severe nausea and vomiting of pregnancy. Nursing Research, 51 (5), 302-308. Olsson, U. H., Foss, T., Troye, S. V., & Howell, R. D. (2000). The performance of ML, GLS, and WLS estimation in structural equation mo deling under conditions of misspecification and nonnormality. Structural Equation Modeling, 7 (4), 557-595. Page, R. M., & Tucker, L. A. (1994). Psychosocial discomf ort and exercise frequency: An epidemiological study of adolescents. Adolescence, 29 (113), 183-191. Parks, S. E., Housemann, R. A., & Brownson, R. C. (2003 ). Differential correlates of physical activity in urban and rural adults of vario us socioeconomic backgrounds in the United States. Journal of Epidemiology and Community Health, 57 (1), 29-35. Poulton, R., Trevena, J., Reeder, A. I., & Richard, R. (2002). Physical health correlates of overprediction of physical discomfort during exercise. Behaviour research and therapy; Behaviour Research and The rapy, 40 (4), 401-414. Prezza Technologies. Ultimate Survey. Available onli ne at http://www.prezzatech.com/biz/company
171 Pronk, N. P., Goodman, M. J., O'Connor, P. J., & Marti nson, B. C. (1999). Relationship between modifiable health risks and shortterm health care charges. JAMA: the Journal of the American Medical Association, 282 (23), 2235-2239. Ratner, P. A., Bottorff, J. L., & Johnson, J. L. (1998) Structural equation modeling. In P. J. Brink & M. J. Wood (Eds.), Advanced design in nursing research (2nd ed., pp. 179-209). Thousand Oaks, CA: Sage. Ream, E., & Richardson, A. (1996). Fatigue: A concept analysis. International Journal of Nursing Studies, 33 (5), 519-529. Redeker, N. S., Lev, E. L., & Ruggiero, J. (2000). In somnia, fatigue, anxiety, depression, and quality of life of cancer patients unde rgoing chemotherapy. Scholarly Inquiry For Nursing Practice, 14 (4), 275-90; discussion 291-8. Reis, R. S., & Sallis, J. (2005). Reliability and vali dity of the Brazilian version of the social support for exercise scale for adolescents. R. Bras. Ci e Mov, 13 (2), 7-14. Reishtein, J. L. (2005). Relationship between symptoms and functional performance in COPD. Research in Nursing & Health, 28 (1), 39-47. Resnick, B. (2001). A prediction model of aerobic exercise in older adults living in a continuing-care retirement community. Journal of Aging and Health, 13 (2), 287-310. Resnick, B. (2005). Reliability and validity of the Ou tcome Expectations for Exercise Scale-2. Journal of Aging & Physical Activity, 13 (4), 382-394.
172 Resnick, B., & Nigg, C. (2003). Testing a theoretical mo del of exercise behavior for older adults. Nursing Research, 52 (2), 80-88. Rhee, H. (2005). Racial/ethnic differences in adolescents' physical symptoms. Journal of Pediatric Nursing, 20 (3), 153-162. Rhee, H., Miles, M. S., Halpern, C. T., & Holditch-Da vis, D. (2005). Prevalence of recurrent physical symptoms in U.S. adolescents. Pediatric Nursing, 31 (4), 314-319, 350. Rimes, K. A., Goodman, R., Hotopf, M., Wessely, S., M eltzer, H., & Chalder, T. (2007). Incidence, prognosis, and risk factors for fatigue and chronic fatigue syndrome in adolescents: A prospective community st udy. Pediatrics, 119 (3), e603-9. Rovniak, L. S., Anderson, E. S., Winett, R. A., & St ephens, R. S. (2002). Social cognitive determinants of physical activity in young adu lts: a prospective structural equation analysis. Annals of Behavioral Medicine: A Publication of the Society of Behavioral Medicine, 24 (2), 149-156. Russell, D., Peplau, L. A., & Ferguson, M. L. (1978). Developing a measure of loneliness. Journal of Personality Assessment, 42 (3), 290-294. Sallis, J. F., Grossman, R. M., Pinski, R. B., Patterson, T. L., & Nader, P. R. (1987). The development of scales to measure social suppor t for diet and exercise behaviors. Preventive Medicine, 16 (6), 825-836. Shin, Y., Jang, H., & Pender, N. J. (2001). Psychomet ric evaluation of the exercise self-efficacy scale among Korean adults with chron ic diseases. Research in Nursing & Health, 24 (1), 68-76.
173 Slade, P. D., Troup, J. D., Lethem, J., & Bentley, G. (1983). The Fear-Avoidance Model of exaggerated pain perception--II. Behaviour Research and Therapy, 21 (4), 409-416. Smith, K. W., Avis, N. E., & Assmann, S. F. (1999). Di stinguishing between quality of life and health status in quality of life research: a meta-analysis. Quality Of Life Research: An International Journal of Quality of Life Aspects of Treatment, Care and Rehabilitation, 8 (5), 447-459. SPSS. (2002). Statistical Package for Social Sciences (Ver sion 11.4). Storch, E. A., Milsom, V. A., Debraganza, N., Lewin, A. B., Geffken, G. R., & Silverstein, J. H. (2007). Peer victimization, psychosocia l adjustment, and physical activity in overweight and at-risk-for-overweig ht youth. Journal of Pediatric Psychology 32 (1), 80-9. Tabachnick, B., & Fidell, L. S. (2001). Using multivariate statistics (4th ed.). Boston: Allyn and Bacon. Taylor, E. D., Theim, K. R., Mirch, M. C., Ghorbani, S., Tanofsky-Kraff, M., AdlerWailes, D. C., et al. (2006). Orthopedic complications of overweight in children and adolescents. Pediatrics, 117 (6), 2167-2174. ter Wolbeek, M., van Doornen, L. J., Kavelaars, A., & Heijnen, C. J. (2006). Severe fatigue in adolescents: A common phenomenon? Pediatrics, 117 (6), e1078-1086.
174 U.S. Department of Health and Human Services. (2002) Healthy People 2010: Understanding and improving health (2nd ed.). Retrieved December 17, 2003, from http://www.healthypeople.gov/Document/tableofcontent s. htm#partb University of South Florida. (2006a). Student directo ry public information. University of South Florida. (2006b). Summer 2006 final enrollment profile. Retrieved July 8, 2006, from http://usfweb2.usf.edu/infomart/eprofiles/ currentprofile/eProfilesUSFSystem.htm Vicente-Rodriguez, G. (2006). How does exercise affect b one development during growth? Sports Medicine, 36 (7), 561-568. Von Ah, D., Ebert, S., Ngamvitroj, A., Park, N., & K ang, D. H. (2004). Predictors of health behaviours in college students. Journal of Advanced Nursing, 48 (5), 463-474. Wang, G., & Dietz, W. H. (2002). Economic burden of o besity in youths aged 6 to 17 years: 1979-1999. Pediatrics, 109 (5), E81-81. Warburton, D. E., Nicol, C. W., & Bredin, S. S. (200 6). Health benefits of physical activity: the evidence. CMAJ: Canadian Medical Association journal = Journal de l'Association Medicale Canadienne, 174 (6), 801-809. Ward, D. S., Dowda, M., Trost, S. G., Felton, G. M., Dishman, R. K., & Pate, R. R. (2006). Physical activity correlates in adolescent girl s who differ by weight status. Obesity, 14 (1), 97-105.
175 Ware, J.E, Jr., Kosinski, M., & Keller, S. D. (1996). A 12-Item Short-Form Health Survey: construction of scales and preliminary tests of reli ability and validity. Medical Care, 34 (3), 220-233. Ware, J. E., Jr., & Sherbourne, C. D. (1992). The MO S 36-item short-form health survey (SF-36). I. Conceptual framework and item select ion. Medical Care, 30 (6), 473-483. Weiss, J. P., Froelicher, V. F., Myers, J. N., & Heidenre ich, P. A. (2004). Healthcare costs and exercise capacity. Chest, 126 (2), 608-613. Weiss, R., Dziura, J., Burgert, T. S., Tamborlane, W. V., Taksali, S. E., Yeckel, C. W., et al. (2004). Obesity and the metabolic syndro me in children and adolescents. The New England Journal of Medicine, 350 (23), 2362-2374. Wessel, T. R., Arant, C. B., Olson, M. B., Johnson, B. D., Reis, S. E., Sharaf, B. L., et al. (2004). Relationship of physical fitness vs bo dy mass index with coronary artery disease and cardiovascular events in wome n. JAMA: the Journal of the American Medical Association, 292 (10), 1179-1187. Whitlock, E. P., Williams, S. B., Gold, R., Smith, P. R., & Shipman, S. A. (2005). Screening and interventions for childhood overweight: a summary of evidence for the US Preventive Services Task Force. Pediatrics, 116 (1), e125-144. Wisen, A. G., Farazdaghi, R. G., & Wohlfart, B. (20 02). A novel rating scale to predict maximal exercise capacity. European Journal of Applied Physiology, 87 (4-5), 350-357.
176 Yang, C. M., & Wu, C. H. (2005). The situational fat igue scale: A different approach to measuring fatigue. Quality Of Life Research: An International Journal of Quality of Life Aspects of Treatment, Care an d Rehabilitation, 14 (5), 1357-1362. Yang, X., Telama, R., Leskinen, E., Mansikkaniemi, K., Viikari, J., & Raitakari, O. T. (2006). Testing a model of physical activity and obesi ty tracking from youth to adulthood: the cardiovascular risk in young Fin ns study. International Journal of Obesity, 31 (3), 521-527. Yarcheski, A., Mahon, N. E., Yarcheski, T. J., & Cannella, B. L. (2004). A metaanalysis of predictors of positive health practices. Journal of Nursing Scholarship: An Official Publication of Sigma Theta Tau International Honor Society of Nursing / Sigma Theta Tau, 36 (2), 102-108. Zimet, G. D., Dahlem, N. W., Zimet, S. G., & Farley G. K. (1988). The Multidimensional Scale of Perceived Social Support. Journal of Personality Assessment, 52 (1), 30-41. Zimmerman, B. J., & Cleary, T. J. (2005). Adolescents' d evelopment of personal agency. In Self-efficacy beliefs of adolescents (pp. 45-69). NA: Information Age Publishing.
178 Appendix A: Exercise Self-efficacy Scale How confident are you that you can exercise regularly (3 or more times per week) under the following circumstances? Rate your confid ence on a scale of 0% (cannot do it) to 100% (certainly can do it). 50% = mo derately certain can do it. 1. When I am feeling tired 2. When I am feeling pressure from work 3. During bad weather 4. After recovering from an injury that caused me to stop exercising 5. During or after experiencing personal problems 6. When I am feeling depressed 7. When I am feeling anxious 8. After recovering from an illness that caused me to stop exercising 9. When I feel physical discomfort with I exercise 10. After a vacation 11.When I have too much work to do 12. When visitors are present 13. When there are other interesting things to do 14. If I donÂ’t reach my exercise goals 15. Without support from my family or friends 16. During a vacation 17.When I have other time commitments 18.After experiencing family problems Note. Compute mean/SD for each subscale as well as for t otal. Items 11, 12, 13, 16, 17, 18 = Factor 1 (situational/i nterpersonal) Items 4, 8,10,14,15 = factor 2 (competing demands) Items 1, 2, 3, 5, 6, 7, 9 = Factor 3 (internal feelin gs) (Shin et al., 2001)
179 Appendix B: Outcome Expectations for Exercise Scale Read the following statements and rate your level of agreement or disagreement. The scale goes from Strongly Agree to Strongly Disagree Exercise: 1. Makes me feel better physically SA A Neither A nor D D SD 2. Makes my mood better in general SA A Neither A nor D D SD 3. Helps me feel less tired SA A Neither A nor D D SD 4. Makes my muscles stronger SA A Neither A nor D D SD 5. Is an activity that I enjoy doing SA A Neither A nor D D SD 6. Gives me a sense of personal accomplishment SA A Neither A nor D D SD 7. Makes me more alert mentally SA A Neither A nor D D SD 8. Improves my endurance in performing my daily activities SA A Neither A nor D D SD 9. Helps to strengthen my bones SA A Neither A nor D D SD Note. SA = 1, A = 2, Neither A nor D = 3, D = 4, SD = 5 Note: Compute mean/SD total. This is the positive subscal e (items 1-9) of the revised version OOE-2 which also includes a 4-item negati ve subscale (not included here). (Resnick et al., 2001)
180 Appendix C: Exercise Goals Scale The following questions refer to how you set exercise go als. Please indicate the extent to which each of the statements below describes yo u: The scale ranges from 1(Does not describe) to 5(Describes completely); 3 = Describes moderately. 1. I often set exercise goals 1 2 3 4 5 2. I usually have more than one major exercise goal 1 2 3 4 5 3. I usually set dates for achieving my exercise goals 1 2 3 4 5 4. My exercise goals help to increase my motivation for doing exercise 1 2 3 4 5 5. I tend to break more difficult exercise goals down into a series of smaller goals 1 2 3 4 5 6. I usually keep track of my progress in meeting my goals 1 2 3 4 5 7. I have developed a series of steps for reaching my exercise goals 1 2 3 4 5 8. I usually achieve the exercise goals I set for myself 1 2 3 4 5 9. If I do not reach an exercise goal, I analyze what went wrong 1 2 3 4 5 10. I make my exercise goals public by telling other people about them 1 2 3 4 5 Score by scale mean of all items. (Rovniak et al., 2002)
181 Appendix D: UCLA Loneliness Scale (ULS-8) Never Most of the time 2. I lack companionship. 1 2 3 4 3. There is no one I can turn to. 1 2 3 4 9. I am an outgoing person. 1 2 3 4 11. I feel left out. 1 2 3 4 14. I feel isolated from others. 1 2 3 4 15. I can find companionship when I want it. 1 2 3 4 17. I am unhappy and withdrawn. 1 2 3 4 18. People are around me but not with me. 1 2 3 4 Note. Reverse score item 9 and 15. Compute sum for scale as well as mean. Mean score > 2 indicates loneliness; May transform to 0 Â– 100 scale (Revised by Hays & DiMatteo, 1987)
182 Appendix E: Social Support for Exercise Scale How much support do you receive from participating in r egular physical activity from the people closest to you? 1 none at all 2 3 4 5 very much Rate the frequency with which the people closest to you have done or said the following in the past month: 1. Exercise with you? 1 none 2 3 4 5 very often 2. Offered to exercise with you? 1 none 2 3 4 5 very often 3. Gave you helpful reminders to exercise? 1 none 2 3 4 5 very often 4. Gave you encouragement to stick with your exercise program? 1 none 2 3 4 5 very often 5. Changed their schedule so you could exercise together? 1 none 2 3 4 5 very often 6. Discussed exercise with you? 1 none 2 3 4 5 very often 7. Complained about the time you spend exercising? 1 none 2 3 4 5 very often 8. Criticized you or made fun of you? 1 none 2 3 4 5 very often 9. Gave me rewards for exercising? 1 none 2 3 4 5 very often 10. Planned for exercise on recreational outings? 1 none 2 3 4 5 very often 11. Helped plan activities around my exercise? 1 none 2 3 4 5 very often 12. Asked me for ideas on how they can get me more exercise? 1 none 2 3 4 5 very often 13. Talked about how much they like to exercise? 1 none 2 3 4 5 very often Note. Compute sum of each subscale (items 1-6 and 13 = f actor 2; items 7-12 = factor 1). (Reis & Sallis, 2005)
183 Appendix F: Multidimensional Scale of Perceived Social Support Rate the following statements about your level of disa greement or agreement. The scale ranges from 1 (very strongly disagree) to 7 (v ery strongly agree). Very SD SD D Neither D nor A A SA Very SA 1. There is a special person around when I am in need. 1 2 3 4 5 6 7 2. There is a special person with whom I can share my joys and sorrows. 1 2 3 4 5 6 7 5. I have a special person who is a real source of comfort to me. 1 2 3 4 5 6 7 10. There is a special person in my life who cares about my feelings. 1 2 3 4 5 6 7 Note. Compute mean/SD for this one subscale. These que stions represent the Â‘significant otherÂ’ subscale (the other 8 questions are i dentical, except for one = family and another = friends). (Zimet et al., 1988)
184 Appendix G: Rating of Perceived Capacity Please select the MOST strenuous exercise capacity level th at you can sustain for 30 minutes without stopping. Are you able, for half an hour or more, to: 1. Sit 2. 3. Walk slowly 4. 5. Walk at normal pace / cycle slowly 6. 7. 8. Jog / cycle 9. 10. Run 11. 12. Run fast / Cycle fast 13. 14. 15. Run very fast (more than 15 km/h) 16. 17. 18. Perform severely difficult elite aerobic training (wo men) 19. 20. Perform severely difficult elite aerobic training (me n) Note. This is a single-item score. (Wizen, Farazdaghi, & Wohlfart, 2002)
185 Appendix H: SF-12 v2 Question 1 Excellent ... Very Good ... Good ... Fair ... In general, would you say your health is excellent, very good, good, fair, or poor? Poor ... Question 2 Limited a lot ... Limited a little ... The following items are about activities you might do during a typical day. Does your health now limit you in these activities? If so, how much? First, moderate activities such as moving a table, pushing a vacuum cleaner, bowling or playing golf. Does your health now limit you a lot, limit you a little, or not limit you at all. Not limited at all ... Question 3 Limited a lot ... Limited a little ... Climbing several flights of stairs. Does your health now limit you a lot, limit you a little, or not limit you at all? Not limited at all ... Question 4 No ... During the past four weeks, have you accomplished less than you would like as a result of your physical health? Yes ...
186 Appendix H (Continued) Question 5 No ... During the past four weeks, were you limited in the kind of work or other regular activities you do as a result of your physical health? Yes ... Question 6 No ... During the past four weeks, have you accomplished less than you would like to as a result of any emotional problems, such as feeling depressed or anxious? Yes ... Question 7 No ... During the past four weeks, did you not do work or other regular activities as carefully as usual as a result of any emotional problems such as feeling depressed or anxious? Yes ... Question 8 Not at all ... Slightly ... Moderately ... Quite a bit ... During the past four weeks, how much did pain interfere with your normal work, including both work outside the home and housework? Did it interfere not at all, slightly, moderately, quite a bit, or extremely? Extremely ... Question 9 All of the time ... Most of the time ... A good bit of the time ... Some of the time ... A little of the time ... These questions are about how you feel and how things have been with you during the past 4 weeks. For each question, please give the one answer that comes closest to the way you have been feeling. How much time during the past 4 weeks have you felt calm and peaceful? All of the time, most of the time, a good bit of the time, some of the time, a little of the time, or none of the time? None of the time ...
187 Appendix H (Continued) Question 10 All of the time ... Most of the time ... A good bit of the time ... Some of the time ... A little of the time ... How much of the time during the past 4 weeks did you have a lot of energy? All of the time, most of the time, a good bit of the time, some of the time, a little of the time, or none of the time? None of the time ... Question 11 All of the time most of the time a good bit of the time some of the time, a little of the time How much time during the past 4 weeks have you felt down? All of the time, most of the time, a good bit of the time, some of the time, a little of the time, or none of the time? none of the time Question 12 All of the time ... Most of the time ... Some of the time ... A little of the time ... During the past 4 weeks, how much of the time has your physical health or emotional problems interfered with your social activities like visiting with friends, relatives etc? All of the time, most of the time, some of the time, a little of the time, or none of the time? None of the time ... (Ware et al., 1996)
188 Appendix I: Situational Fatigue Scale According to your general feelings for the past month, please rate the level of fatigue that you might experience after engaging in the following activities. 0 Â– no fatigue at all; 5 = extreme fatigue 1. Playing a ballgame for 30 minutes 0 1 2 3 4 5 2. Jogging for 20 minutes 3. Taking a walk for an hour 4. Cleaning house for 30 minutes 5. Reading magazines/paper for 1 hour 6. Watching TV for 2 hours 7. Chatting for 1 hour 8. Shopping for 1 hour 9. Driving for 1 hour 10. Hosting a social event for 30 minutes 11. Doing paperwork for 1 hour (e.g. typing, writing, accounting, making plans) 12. Meeting for 2 hours 13. Attending a social activity for 1 hour Note. Items 1-4 = Factor 1 (physical fatigue subscale). Items 5-13 = Factor 2 (Mental fatigue subscale). (Yang & Wu, 2005)
189 Appendix J: Chalder Fatigue Scale No or better than Usual No more than Usual Worse than Usual Much Worse than Usual 1. Do you have problems with tiredness? 1 2 3 4 2. Do you need to rest more? 1 2 3 4 3. Do you feel sleepy or drowsy? 1 2 3 4 4. Do you have problems starting things? 1 2 3 4 5. Do you start things without difficulty but get weak as you go on? 1 2 3 4 6. Are you lacking in energy? 1 2 3 4 7. Do you have less strength in your muscles? 1 2 3 4 8. Do you feel weak? 1 2 3 4 9. Do you have difficulty concentrating? 1 2 3 4 10. Do you have problems thinking clearly? 1 2 3 4 11. Do you make slips of the tongue when speaking? 1 2 3 4 12 Do you find it more difficult to find the correct word? 1 2 3 4 13. How is your memory? 1 2 3 4 14. Have you lost interest in the things you used to do? 1 2 3 4 Note. Factor 1 = physical fatigue (items 1-8) Factor 2 a= Mental fatigue (items 9-14). Shorter version may be used for an 11 item scale (elim inate 5, 12, 14). An even shorter version has been done using items 1-9. Score by total sum or by summing the two factors separate ly. (Chalder et al., 1993)
190 Appendix K: West Haven-Yale Multidimensional Pain Scal e Have you in the last month experienced ache, pain, discomfort, or throbbing due to headache, cramps, muscles, joints, or other non-infectious conditions? Yes/no 7. On the average, how severe has your pain been during the past week? 0 = None 6 = extremely severe 0 1 2 3 4 5 6 12 How much suffering do experience because of your pain? 0 = no suffering 6 = extreme suffering 0 1 2 3 4 5 6 1. Rate the level of your pain at the present moment. 0 = no pain 6 = very intense pain 0 1 2 3 4 5 6 18. During the past week, how irritable have you been? 0= Not at all irritable 6 =extremely irritable 0 1 2 3 4 5 6 20. During the past week, how tense or anxious have you been? 0 = Not at all tense or anxious 6 = extremely tense or anxious 0 1 2 3 4 5 6 6 Rate your overall mood during the past week. 0 = Extremely low mood 6 = extremely high mood 0 1 2 3 4 5 6 Note. Score by computing mean/SD. (Kerns, Turk & Rudy, 1985)
191 Appendix L: International Physical Activity Questionnai re The following questions are about exercise frequency and vigorous intensity. Vigorous intensity is when your heart rate i ncreases or you canÂ’t talk during exercise, or your talking is broken up b y large breaths. Think only about those physical activities that you did for at least 10 minutes at a time. 1. During the last 7 days, on how many days did you do vigorous physical activities? _____ days 2. How much time did you usually spend doing vigorous physical activities on one of those days? ____ ____ Hours per day ___ ____ ____ Minutes per day The following questions are about exercise frequency and moderate intensity. Moderate intensity is when your heart beats faster than normal. You can talk but canÂ’t sing. Think only about those physical activities that you did for at least 10 mi nutes at a time. 3. During the last 7 days, on how many days did you do moderate physical activities? _____ days 4. How much time did you usually spend doing moderate physical activities on one of those days? [____ ____ Hours per day ___ ____ ____ Minutes per day The following questions are about exercise frequency and light intensity. Light intensity is walking at a normal pace. You can talk and sing. Think only about those physical activities that you did for at least 10 minutes at a time.
192 Appendix L (Continued) 5. During the last 7 days, on how many days did you do light physical activities such as walking ? _____ days 6. How much time did you usually spend doing light physical activities such as walking on one of those days? ____ ____ Hours per day ___ ____ ____ Minutes per day Now think about the time you spent sitting on week days during the last 7 days. Include time spent at work, at home, whil e doing course work, and during leisure time. This may include time s pent sitting at a desk, visiting friends, reading or sitting or lying d own to watch television. Think only about those physical activities that you did for at least 10 minutes at a time. 7. During the last 7 days, on how many days did you spend sitting (that also includes lying down while awake)? _____ days 8. How much time did you usually spend doing sitting activities on one of those days? ____ ____ Hours per day ___ ____ ____ Minutes per day Note. Compute minutes spent in each activity level; mult iply that by the met min per week for each activity level and then sum the met m in level per week total. ( Craig et al, 2002; 2005)
193 Appendix M: Demographic Form 1. How old are you? Please select from the followin g: 1. 18 2. 19 3. 20 4. 21 5. 22 6. 23 7. 24 8. 25 9. Other 2. With which ethnic / cultural group do you most c losely identify? 1. Hispanic 2. Non-Hispanic 3. Other 3. Which 1 or more would you say is your race 1. Caucasian/white 2. American Black or Caribbean Black 3. Native or Alaskan Indian 4. Hawaii or Pacific Islander 5. Asian 6. Other: Please fill in the empty box as needed.
194 Appendix N: Elements of Informed Consent Information for People Who Take Part in Research St udies Researchers at the University of South Florida (USF) stu dy many topics. We want to learn more about the factors affecting studentsÂ’ decisions to exercise. To do this, we need the help of people who agree to take part in a research study. Title of research study: The College Exercise Project Person in charge of study: Sarah Elizabeth Cobb RN MS Study staff who can act on behalf of the person in charge: Mary Evans PHD Where the study will be done: Online using Ultimate Survey Internet data collection tools. Should you take part in this study? This form tells you about this research study. You can decide if you want to take part in it. Y ou do not have to take part. Reading this form can help you decide. You can ask questions: You may call the primary investigator Sarah Elizabeth Cobb RN MS at 813-905-4251 or may email her at email@example.com Why is this research being done? The purpose of this study is to find out how psychosocial factors affect exercise in young women. Why are you being asked to take part? We are asking you to take part in this study because you are a young female between the ages of 18 and 25; we want to learn about age differences between adolescent women (under age 21) and other young women (age 21-25) How long will you be asked to stay in the study? You will be asked to spend about 20 minutes taking the o nline survey. There are no study visits. What other choices do you have if you decide not to take part? If you decide not to take part in this study, that is okay How do you get started? If you decide to take part in this study, you will nee d to access the study using the URL link that is provided to you in this email for you. What will happen during this study? You will be asked questions pertaining to exercise, and will rate how much those items affect your decisions to exercise. Will you be paid for taking part in this study? We will not pay you to take the survey. However I will have a lottery for two prizes of $ 100 each. It will not cost you anything to take part in the study What are the potential benefits if you take part in this study? We donÂ’t know if you will get any benefits by taking part in this study other than knowing you have helped advance the knowledge about decisions to exercise What are the risks if you take part in this study? There are no known risks to those who take part in this study.
195 Appendix N (Continued) What will we do to keep your study records private? Federal law requires us to keep your study records privat e. The data from the Internet survey will be transferred to a dedicated com puter that is kept in a locked cabinet in a locked room. However, certain people may n eed to see your study records. By law, anyone who looks at your records must kee p them confidential. The only people who will be allowed to see these reco rds are: The study staff. People who make sure that we are doing the study in the right way. They also make sure that we protect your rights and safety: o The USF Institutional Review Board (IRB) and its staf f, and any other individuals acting on behalf of USF. o The United States Department of Health and Human Se rvices (DHHS) We may publish what we find out from this study. If we do, we will not use your name or anything else that would let people know who you are. If you decide not to take part: You wonÂ’t be in trouble or lose any rights you normally have. What if you join the study and then later decide yo u want to stop? If you decide you want to stop taking part in the stud y, simply log off from the Internet survey. If you have any questions about this study or in the ev ent of research related harm, call Sarah Elizabeth Cobb at 813-905-4251. If you have questions about your rights as a person who is taking part in a study, call USF Research Integrity and Compliance at (813) 974 -5638. Consent to Take Part in this Research Study ItÂ’s up to you. You can decide if you want to take par t in this study. I understand that this is research. I have receive d a copy of this consent form via this cover letter. I understand that my pa rticipation in the research is voluntary, and that participation indicates my c onsent.
196 Appendix O: Covariances and Variances for Actual Data (N =463) Variable SE1 SE2 SE3 EXP1 EXP2 EXP3 G1 SE1 5.16 SE2 4.00 5.61 SE3 4.18 4.43 5.99 EXP1 1.47 1.38 1.65 5.59 EXP2 1.88 1.82 2.15 4.32 5.72 EXP3 0.98 0.84 1.08 3.14 2.97 2.97 G1 3.23 2.93 3.20 1.86 2.81 1.38 12.00 G2 2.74 2.86 2.84 1.87 2.61 1.42 8.58 G3 3.76 3.61 3.95 2.35 3.39 1.85 9.93 L1 0.73 1.06 1.27 1.15 1.04 1.11 1.95 L2 1.24 1.44 2.19 1.98 1.73 1.16 1.80 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
197 Appendix O (Continued) Variable SE1 SE2 SE3 EXP1 EXP2 EXP3 G1 SS1 0.66 0.20 0.30 0.28 0.49 1.22 1.79 SS2 1.39 0.92 1.18 0.67 0.92 1.33 2.77 SS3 0.43 0.26 0.32 0.31 0.46 0.41 0.67 AGE 0.39 0.40 0.35 0.49 0.51 0.07 0.38 SFTOT 10.07 10.58 10.78 5.42 7.28 3.55 7.00 RPC 2.47 2.62 2.81 1.18 2.16 0.96 2.77 US11 2.45 2.26 2.69 1.13 1.30 0.92 2.31 US12 0.39 0.35 0.40 0.20 0.14 0.17 0.27 US21 0.62 0.14 0.34 0.40 0.50 0.12 0.33 US22 1.65 1.53 1.73 0.66 0.81 0.47 0.73 PA 2.18 2.33 2.13 0.79 1.70 0.91 2.65 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
198 Appendix O (Continued) Variable G2 G3 L1 L2 SS1 SS2 SS3 G2 10.81 G3 10.19 15.73 L1 1.50 2.59 23.92 L2 1.56 3.29 18.61 29.60 SS1 2.04 2.18 3.87 3.69 14.29 SS2 2.99 3.90 3.19 3.29 10.14 11.29 SS3 0.77 1.18 1.00 1.39 3.26 3.17 2.10 AGE 0.01 0.21 0.58 0.62 -0.36 -0.02 -0.06 SFTOT 8.04 10.14 21.81 37.05 0.93 2.62 1.24 RPC 2.23 3.16 1.16 1.61 0.94 1.59 0.53 US11 1.84 3.33 6.49 9.57 1.47 2.06 0.30 US12 0.27 0.37 1.15 1.81 0.26 0.31 0.02 US21 -0.29 -0.19 2.15 3.26 -0.76 -1.20 -0.40 US22 0.54 1.90 5.65 9.07 0.69 0.67 0.17 PA 2.87 3.80 0.48 2.04 0.89 1.64 0.28 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
199 Appendix O (Continued) Variable Age SFTOT RPC US11 US12 US21 US22 PA AGE 4.10 SFTOT 1.53 249.42 RPC -0.30 13.44 9.26 US11 0.42 35.84 2.23 18.26 US12 0.02 5.48 0.33 2.24 0.72 US21 0.25 29.45 0.84 4.72 0.70 14.22 US22 -0.11 25.17 1.04 6.58 1.13 4.65 13.53 PA -1.22 5.80 2.59 2.13 0.33 -1.48 1.37 14.48 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
200 Appendix P: Covariances and Variances for Implied Data (N = 463) Variable US11 US12 US21 US22 PA SE1 US11 18.264 US12 2.236 0.719 US21 4.719 0.704 14.221 US22 6.581 1.127 4.654 13.525 PA 2.129 0.329 -1.479 1.365 14.478 SE1 2.450 0.391 0.616 1.649 2.175 5.162 SE2 2.258 0.346 0.140 1.531 2.329 3.996 SE3 2.691 0.397 0.342 1.731 2.131 4.183 EXP1 1.128 0.197 0.402 0.663 0.791 1.467 EXP2 1.301 0.140 0.501 0.808 1.704 1.883 EXP3 0.921 0.173 0.120 0.469 0.909 0.981 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
201 Appendix P (Continued) Variable US11 US12 US21 US22 PA SE1 G1 2.313 0.267 0.328 0.732 2.650 3.232 G2 1.838 0.267 -0.291 0.542 2.865 2.742 G3 3.331 0.373 -0.186 1.897 3.798 3.757 L1 6.485 1.147 2.150 5.647 0.476 0.725 L2 9.565 1.806 3.255 9.065 2.039 1.237 SS1 1.467 0.256 -0.756 0.687 0.893 0.656 SS2 2.060 0.307 -1.197 0.668 1.640 1.393 SS3 0.295 0.024 -0.399 0.172 0.277 0.431 AGE 0.420 0.024 0.249 -0.109 -1.220 0.389 SF 35.841 5.484 29.446 25.17 5.804 10.071 RPC 2.233 0.333 0.839 1.037 2.591 2.469 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
202 Appendix P (Continued) Variable SE2 SE3 EXP1 EXP2 EXP3 G1 SE2 5.605 SE3 4.431 5.989 EXP1 1.380 1.653 5.590 EXP2 1.817 2.146 4.323 5.722 EXP3 0.842 1.084 3.137 2.967 2.972 G1 2.930 3.202 1.859 2.806 1.384 11.999 G2 2.856 2.840 1.867 2.608 1.423 8.578 G3 3.610 3.954 2.348 3.387 1.848 9.930 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
203 Appendix P (Continued) G2 G3 L1 L2 SS1 SS2 G2 10.809 G3 10.194 15.727 L1 1.502 2.591 23.924 L2 1.558 3.285 18.614 29.602 SS1 2.040 2.175 3.867 3.691 14.285 SS2 2.988 3.897 3.194 3.289 10.135 11.29 SS3 0.770 1.179 0.999 1.391 3.256 3.171 AGE 0.012 0.206 0.581 0.622 -0.363 0.016 SF 8.035 10.144 21.805 37.052 0.932 2.618 RPC 2.225 3.155 1.158 1.613 0.935 1.590 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
204 Appendix P (Continued) SS3 AGE SF RPC SS3 2.101 AGE -0.063 4.103 SF 1.240 1.528 249.42 RPC 0.532 -0.299 13.435 9.260 Note SE = Self-efficacy (SE rescaled by 10 -1 ); EXP = Expectations; G = Goals; L = Loneliness (L rescaled by 10 -1 ); SS = Social Support; SFTOT = SF12-V2; US = Unpleasant Symptom; PA = Physical Activity. (PA r escaled by 1000 -1 ).
205 Appendix Q : Syntax Used for Post-hoc Power Analysis in SPSS title 'power estimation for sem'. compute alpha = 0.05. compute rmsea0 = 0.05. compute rmseaa = 0.08. compute df = 159. compute n = 463. compute ncp0 = (n-1)*df*rmsea0**2. compute ncpa = (n-1)*df*rmseaa**2. do if (rmsea0 rmseaa). compute cval= idf.chisq(alpha,df). compute power = ncdf.chisq(cval,df,ncpa). end if. execute. list alpha df n power. List ALPHA DF N POWER .05 159.00 463.00 1.00 Number of cases read: 1 Number of cases listed: 1
206 Appendix R: Syntax Used to Calculate Delta and Needed Sample Size comment compute noncentrality parameter delta. comment create variables in the data editor. comment df and power first. set mxloop = 1000. compute #alpha = 0.05. compute #df = df. compute #power = power. compute #crit = idf.chisq(1-#alpha, #df). compute delta = rnd(#crit #df). compute #times = 1. compute #direc = 1. compute #amount = 10. loop. + compute delta = delta + #direc*#amount. + compute #pow = 1 ncdf.chisq(#crit,#df,delta). + do if (#direc*(#power #pow) < 0). + compute #times = #times + 1. + compute #direc = -1*#direc. + compute #amount = #amount/10. + end if. end loop if (#times = 8). execute. *********************************note. compute chi= idf.chisq(1-alpha, df). EXECUTE. compute powera = 1ncdf.chisq(chi,df,delta). EXECUTE. Format delta powera (F8.3). EXECUTE. list alpha delta powera. List ALPHA DELTA POWERA .05 49.759 .800 Number of cases read: 1 Number of cases listed: 1 compute rmsea = 0.05. compute n_needed = ((delta power)/(((rmsea*rmsea)*df )) + 1). execute. list alpha delta powera n_needed. List ALPHA DELTA POWERA N_NEEDED .05 49.759 .800 124.17 Number of cases read: 1 Number of cases listed: 1
About the Author Sarah Elizabeth Cobb RN received her Bachelor of Scie nce degree Nursing (BSN) from East Tennessee State University in 1976. She worked in the pediatric/ school health/ community health field for th e intervening years between graduating with the BSN and starting graduate school at the University of South Florida in 2002. She obtained her Master of Science in nursing (MS) from the University of South Florida in 2005, and completed h er Ph.D. in 2007 with an emphasis on pediatrics, childrenÂ’s mental health, and qua ntitative methodology.