|USFDC Home | USF Electronic Theses and Dissertations||| RSS|
This item is only available as the following downloads:
Cross-Cultural Measurement Invariance Of Work/Family Conflict Scales Across EnglishSpeaking Samples by David Evan Loran Herst A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Psychology College of Arts and Sciences University of South Florida Major Professor: Michael Brannick, Ph.D. Tammy Allen, Ph.D. Walter Borman, Ph.D. Vicky Phares, Ph.D. Judith Becker-Bryant, Ph.D. Date of Approval: September 18, 2003 Keywords: women, work/life balance, work/life fit Copyright 2003 David Evan Loran Herst
Dedication This work is dedicated to a ll of those who had faith in me when I did not have faith in myself. It is especially dedicated to the loving memory of my sister, who departed this world far too soon for me to share my joy with on a mortal basis. Elissa, I know you see me now, watch me live and thrive, and I know you will always be with me.
Acknowledgements This paper would not have been possible wi thout the love and support from my dearest friends Patricia Vandenberg and Tim Carey. It would never have happened without the consistent encouragement of my parents Adolf & Patty Herst, and Barbara & Kevin Kelly. I cant thank my advisor Michael Brannick enough for his calm guidance and in-depth conversations into the frustrati ons of confirmatory factor anal ysis. Special thanks to my committee, including Walter Borman, Judy Bryant, Vi cky Phares, and Tammy Allen. All of this would not have happened without Walter Nords agreement to chair the defense at the last minute! Lastly, the defense and final manus cript submission may have caused a nervous breakdown without the calm and courageous support of Dana Collier.
i Table of Contents List of Tables iii List of Figures vi Abstract vii Introduction 1 Work and Family 2 Work/life balance 2 Work/family conflict 4 Measuring Work/Family Conflict 7 Bohen and Viveros-Long (1981) 9 Kopelman, Greenhaus, and Connolly (1983) 13 Wiley (1987) 15 Netemeyer, Boles, and McMurrian (1996) 17 Stephens and Sommer (1996) 19 Carlson, Kacmar, and Williams (2000) 21 Summary of work/family conflict measurement 26 Testing Work/Family Conflict Measurement Equivalence Across Multiple Populations 26 Measure structure vs. stimulation 27 Cultural similarities between the United States, Australia, and New Zealand 28 Cross-cultural gender issues in work/family conflict 33 Purpose of the Current Study 38 Method 41 Participants 41 Validation sample 41 Comparison sample 42 Measures 44 Work/family conflict 44 Procedure 44 Design and Analysis 45 Identification of the Factor Model 47 Determining Measurement Invariance 48 Determining manifest loading invariance 49 Partial measurement invariance 50 Testing Model Fit 51
ii Results 51 Data Preparation 51 Comparison sets 51 Imputation of missing values 52 Descriptive statistics 52 Determining Measurement Invariance 53 Comparing variance/covariance matrices 53 Hypotheses 1a and 1b: F actor pattern invariance 53 Invariance of manifest va riable factor loadings 54 Statistical identification of the model 55 Partial measurement invariance 57 Exploratory Factor Analysis 61 Summary and explanation of exploratory factor and components analysis 66 Hypothesis 2: Gender Differences in the Australian/New Zealand Sample 67 Comparing variance/covariance matrices 67 Hypothesis 2: Factor pattern invariance 67 Invariance of manifest va riable factor loadings 67 Statistical identification of the model 68 Partial measurement invariance 68 Research Question 1: Differences between Males and Females In the Australian/New Zealand Sample 70 Discussion 70 Testing for Measurement Invariance 71 Hypotheses 1a and 1b 71 Invariance of manifest va riable factor loadings 73 Gender Differences in the Australian/New Zealand Sample 77 Hypothesis 2 and research question 1 77 Issues in Interpreting Study Results 79 Summary of Results 81 Impact on Work/Family Conflic t and Future Implications 83 Conclusions 85 References 86 About the Author End Page
iii List of Tables Table 1 Chronology of Primary Tools Used in Measuring Generalized Work/Family Conflict 97 Table 2 Komarovskys (1977) Six Modes of Role Tension 98 Table 3 Burley (1989) Work/Family Conflict Scale 98 Table 4 Carlson et al. Work/Family Conflict Scale 99 Table 5 Item Intercorrelations for Ma les in the United States Sample 100 Table 6 Means, Standard Deviations, Reliabilities, and Intercorrelation of Scales for Males in the United States 101 Table 7 Item Intercorrelations for Male s in the Australia/New Zealand Sample 101 Table 8 Means, Standard Deviations, Reliabilities, and Intercorrelation of Scales for Males in Australia/New Zealand 102 Table 9 Item Means and Standard Devia tions for Males in the United States and Australia/New Zealand Samples 103 Table 10 Item Intercorrelations for Females in the United States Sample 104 Table 11 Means, Standard Deviations, Reliabilities, and Intercorrelation of Scales for Females in the United States 105 Table 12 Item Intercorrelations for Females in the Australia/New Zealand Sample 105 Table 13 Means, Standard Deviations, Reliabilities, and Intercorrelation of Scales for Females in Australia/New Zealand 106 Table 14 Item Means and Standard Deviatio ns for Females in the United States and Australia/New Zealand Samples 107 Table 15 Item Intercorrelations Between Males in the United States and Australia/New Zealand 108
iv Table 16 Means, Standard Deviations, a nd Intercorrelation of Scales for Males in the United States a nd Australia/New Zealand 109 Table 17 Item Intercorrelations Between Females in the United States and Australia/New Zealand 109 Table 18 Means, Standard Deviations, and Intercorrelation of Scales for Females in the United States and Australia/New Zealand 110 Table 19 Level of Overall Fit Acro ss Multiple Models of Invariance 111 Table 20 Steps in Statistical Identificati on of the Model for Males and Females 111 Table 21 Partial Measurement Invari ance Fit Statistics for Males 112 Table 22 Partial Measurement Invari ance Fit Statistics for Females 113 Table 23 Within-Group Completely Standa rdized Manifest Variable Loadings and Modification Indices for Males 113 Table 24 Within-Group Completely Sta ndardized Phi Solution for Males 114 Table 25 Within-Group Completely Standa rdized Theta-Delta Error Terms for Males 115 Table 26 Within-Group Completely Standa rdized Manifest Variable Loadings and Modification Indices for Females 115 Table 27 Within-Group Completely Standardized Phi Solution for Females 116 Table 28 Within-Group Completely Standa rdized Theta-Delta Error Terms for Females 117 Table 29 Exploratory Analysis Loadings and Patterns in a Forced 6-Factor Solution for Australia/New Zealand Males 118 Table 30 Exploratory Analysis Loadings and Patterns in a Forced 5-Factor Solution between US and Australia/New Zealand Males 119 Table 31 Exploratory Analysis Loadings and Patterns in a Forced 6-Factor Solution between US and Australia/New Zealand Females 120 Table 32 Exploratory Analysis Loadings and Patterns in a Forced 5-Factor Solution between US and Australia/New Zealand Females 121
v Table 33 Principal Components Analysis Lo adings and Patterns in a Forced 6Component Solution between US a nd Australia/New Zealand Males 122 Table 34 Principal Components Analysis Lo adings and Patterns in a Forced 5Component Solution between US a nd Australia/New Zealand Males 123 Table 35 Principal Components Analysis Lo adings and Patterns in a Forced 6Component Solution between US and Australia/New Zealand Females 124 Table 36 Principal Components Analysis Lo adings and Patterns in a Forced 5Component Solution between US and Australia/New Zealand Females 125 Table 37 Comparison of Overall Fit Across Multiple Models for Measurement Invariance between Australian/New Zealand Males and Females 125 Table 38 Partial Measurement Invariance Fit Statistics between ANZ Males and Females 126 Table 39 Within-Group Completely Standa rdized Manifest Variable Loadings and Modification Indices for ANZ Males and Females 127 Table 40 Within-Group Completely Standardized Phi Solution for ANZ Males and Females 127 Table 41 Within-Group Completely Standa rdized Theta-Delta Error Terms for ANZ Males and Females 128 Table 42 Mean Group Comparisons Betw een Australian/New Zealand Males and Females 128
vi List of Figures Figure 1. Carlson et al. (2000) model of work/family conflict 129
vii Cross-Cultural Measurement I nvariance of Work/Family Conflict Scales Across EnglishSpeaking Samples David Evan Loran Herst ABSTRACT The factor structure of the work/fam ily conflict scale developed by Carlson, Kacmar, & Williams (2000) was analyzed for measurement invariance between a US and an Australian/New Zealand (ANZ) sample using a multisample confirmatory factor analysis procedure. Results indicated that factor pattern fit between the female samples on the common model was good-to-mediocre, a nd factor pattern fit between the male samples and the common model was medi ocre-to-poor. Both samples exhibited significant changes in chi square when testi ng for the more restrictive factor loading equivalence. Partial measurement invariance revealed a better fit between the male samples when three of the items were unconstrained, and when eight items were unconstrained between the female samples. Finally, males and females in the ANZ sample exhibited factor pattern invariance, but required two item s to be unconstrained before factor loading invariance was achie ved. Mean differences on the six scales revealed higher levels of time-based wo rk interference with family and family interference with work for males than for females in the ANZ sample.
1 Cross-Cultural Measurement Invariance of Work/Family Conflict Scales across EnglishSpeaking Samples Over the past 20 years a gradual shift in the makeup of the labor pool has resulted in a tremendous influx of women into th e US workforce (Howard, 1995). Yet women have not only entered the workforce but have also rapidly moved into managerial and professional positions that have ever increa sing role requirements, pulling from time and effort previously spent dealing with home a nd family matters. Because this shift has not been accompanied by any noticeable increase in government or private assistance in childcare, healthcare, or eldercare, beyond the Family and Medical Leave Act of 1993, the result has been a situation in which both men and women are left with mounting responsibilities within their work and family roles. Not surprisingly, these mounting responsibilities have led to a grea ter potential for role conflict. In their seminal work on organizationa l stress, Kahn, Wolfe, Quinn, Snoek, and Rosenthal (1964) defined role conflict as th e simultaneous occurre nce of two (or more) sets of pressures such that compliance w ith one would make more difficult compliance with the other (in Greenhaus & Beutell, 1985, p. 77). These sets of pressures can originate both within and between roles, th ereby setting up the possibility of multiple conflict scenarios from multiple sources. Fo r instance, for the domains of work and family, role conflict may origin ate within the family itself (i ntrarole conflict, e.g. parent vs. spouse), within work itself (manager ial obligation to the company vs. to
2 subordinates), or between the domains (interrole conflict, e.g. work role vs. family role conflict). Yet the measurement of these form s of conflict has often been conducted with study-specific scales that were not devel oped using rigorous psychometric procedures. This is particularly true for work/family conflict, where, until recently, the lack of standardized measurement has presented bot h theoretical and practical problems for researchers. The purpose of this study wa s to validate the structure of a current work/family conflict scale across two Englis h-speaking cultures, thereby furthering the standardization of work/fam ily conflict measurement. Work and Family Research on the intersection between wo rk and family roles has focused on the area of conflict, where role demands from each domain are considered to be mutually incompatible (Parasuraman & Greenhaus, 2002). This stems from a scarcity hypothesis, where an individuals resources (such as tim e) are fixed, and participating in multiple roles inevitably leads to conflict. Yet more recent research has investigated the idea that work and family roles may be mutually benefici al. This line of inquiry is better known as work/life balance, or work/life fit. Work/life balance. Work/life balance, also known as work/life fit, has been defined as satisfaction and good functioning at work and at home with a minimum of role conflict (p. 349; Clark, 2001). This st ems from role theory, which indicates that participation in multiple roles can enhance an individuals quality of life (Sieber, 1974; in Parasuraman & Greenhaus, 2002). Current resear ch has found evidence that this type of role interaction does occur between the work and family domains. For instance, using daily repeated measures from diary entrie s, Doumos, Margolin, and John (2003) studied
3 the spillover and crossover effects between health promoting be haviors and marital interaction. Their findings indi cate that marital interactio ns affected both wives and husbands hours worked. The importance of wo rk/life balance is also found in studies of telework. Specifically, when compared to office workers, Hill, Miller, and Colihan (1998) found that teleworking did not enhan ce the employees perception of work/life balance. In fact, qualitative analysis seemed to indicat e an equivocal reaction, where increased mobility was viewed as positive, but location in home was viewed as a breakdown in the boundary between the two domains. Clark (2001) pointed out that much of th e work/life balance literature has focused on individual facets of the construct, such as work outcomes, role outcomes, or role conflict and stress. She also argued that limiting study variables to conflict presupposes that balance is the absence of conflict, and instead recomme nds that research expand to include multiple aspects of the work/life ba lance construct when considering how it may be affected. In her own work, Clark has looked at the effects of workplace culture on five aspects of work/life balance: work sa tisfaction, home satisfaction, work functioning, home functioning, and role conflict. Her fi ndings show the work culture dimension of operational flexibility (ability to alter ones schedule) was most often associated with work/family balance, specifically with work satisfaction and family well-being. In addition, supportive supervision was associated with increased employee citizenship. However, one interesting finding was that temp oral flexibility was unassociated with any of the studys measures of work/life balance. Finally, none of the work culture measures were associated with home activity satis faction or the amount of role conflict.
4 Overall, the emerging work /life balance literature indicates a construct that includes both work/family conflict and the extent to which spillover positively affects both the work and family domains. However, the exact nature of th e construct has yet to be fully understood. Researchers continue to both qualitative ly (Clark, 2001b; Voydanoff, 2002) and quantitatively (D eBord, Canu, & Kerpelman, 2000; Zimmerman, Haddock, Current, & Ziemba, 2003) search for a comprehensive theory that may or may not exist. However, it is not by accident that many cues in this line of inquiry have been taken from the extensive work/family conflict literature. As a resu lt, this study focuses on the work/family conflict aspect of work and family research. Work/family conflict. Greenhaus and Beutell (1985) defined work/family conflict as a form of interrole conflict in which the role pressures from the work and family domains are mutually incompatible in some re spect (p. 77). According to their model, work/family conflict is both bi-directional and multi-dimensional. The bi-directionality comes from the concept that wo rk can interfere with family, and that family can interfere with work. In addition, the multi-dimensional nature of work/family conflict occurs in each direction. Specifically, both family inte rference with work and work interference with family are composed of three dimens ions: time-, strain-, and behavior-based conflict. Time-based work/family conflict arises when the time demands from one role make it physically impossible to meet the requirements of another role, and when preoccupation with one roles requirements occurs, even when physically involved in meeting the requirements of another role (Greenhaus & Beutell, 1985). Work-related sources of conflict include hours worked a nd commuted per week, amount and frequency of overtime, irregularity of shift work, and inflexibility in the work schedule. Family-
5 related sources include number of children, younger children, and family size (including older relatives). Next, strain-based work/family conflict is derived from role-produced strain, when strain from one role inte rferes with fulfilling responsibi lities in another role. Workrelated sources of strain-bas ed conflict include work-role ambiguity, intrarole workconflict, low levels of leader support, a nd high physical and psychological demands. Family-related sources include lack of spous al support, husband-wife dissimilarity in career orientation, husband-wife disagreement about family roles, and husband-wife dissimilarity in attitudes towards a wifes employment status. Lastly, behavior-based work/family conflic t occurs when behavior in one role may be incompatible with expectations for behavior in another role. Work-related antecedents include work ambiguity and work involvement, and family related antecedents include family-role ambiguity, intr a-family role conflict, social support, and family role involvement (Carlson, Kacmar, & Williams, 2000). In recent years, research on conflict between the different role domains of work and family has dramatically increased (Burke & Greenglass, 2001; Fu & Shaffer, 2001). Organizational outcomes such as job satisfac tion, commitment, performance, turnover intentions, etc. have been extensively inve stigated. For instance, higher levels of work/family conflict are rela ted to lower levels of j ob satisfaction, organizational commitment, job performance, higher levels of turnover intention, and burnout (Allen, Herst, Bruck, & Sutton, 2000; Aryee, 1992; Bacharach, Bamberger, & Conley, 1991; Kossek & Ozeki, 1998). Effects are not limited to the realm of work, as higher levels of work/family conflict are related to lower levels of life satis faction, marital satisfaction,
6 family satisfaction, and increased family di stress (Allen et al. 2000; Frone, Yardley, & Markel, 1997; Kinnunen & Mauno, 1998). In addition, work/family conflict has been indirectly linked to child acti ng out and problems at school. Testing a model of fathers work attitudes and interrole status, Stewart and Barling (1996) found th at higher levels of work/family conflict are positiv ely related to negative pare nting practices (punishing behaviors and rejecting behaviors). These ne gative parenting practices were then found to directly relate to child behavior, sp ecifically acting out and school competence. Work/family conflict is also re lated to overall physical heal th, blood pressure, and health complaints (Adams & Jex, 1999; Frone, Russe ll, & Barnes, 1996; Thomas & Ganster, 1995), not to mention general psychological strain, substance abuse, and depression (Beatty, 1996; Burke, 1988; Frone, Russell, & Cooper, 1993). In fact, Frone (2000) found that the probabili ty of having an anxiety, mood, or substance abuse disorder was directly related to levels of both work interf erence with family and family interference with work. In addition, Frone found a Gender by work/family conflic t interaction, where men with high levels of family interference with work had a higher probability of having an anxiety disorder than did women. Researchers have also discovered a multit ude of variables that directly and indirectly affect the previously mentione d outcomes as antecedents to work/family conflict. These include supervisor support/work climate (A llen, 2001; Thompson, Beauvais, & Lyness, 1999; Carlson & Pe rrewe, 1999; Kirchmeyer & Cohen, 1999; Carlson & Perrewe, 1999), dispositional variab les such as Type A behavior and Negative Affectivity (Burke, 1988; Carlson, 1999; Stoeva, Chui, & Greenhaus, 2002), role conflict/overload (Frone, Yardley, & Ma rkel, 1997; Carlson, 1999; Shamir, 1983;
7 Wallace, 1999;), and coping behaviors (Matsu i, Ohsawa, & Onglatco, 1995; Kirchmeyer & Cohen, 1999; Stoeva, Chui, & Greenhaus, 20 02). As can be seen, the impact of work/family conflict on an individuals life is vast. Clearly work/family conflict is deserving of the attention it has garnered. In addition to antecedents, consequences, and covariates, the constructs influence has also been found to be similar across cultures. For example, using a Hong Kong sample, Aryee, Fields, and Luk (1999) replic ated an explanatory model of work/family conflict previously validated by Frone, Russell, and Coope r (1992) on a US sample. Aryee et al. concluded that the reciprocal nature and as ymmetrical boundary permeability of the two forms of work/family conflict we re not culture specific. In another study, Matsui, Ohsawa, and Onglatco (1995) found th at levels of work/family conflict were related to parental demands among Japanese wive s, results that were also found in Aryee, Luk, Leung, and Los (1999) Hong Kong samp le, and Frone, Yardley, and Markels (1997) study using a US sample. Finally, Ca rlson (1999) found that negative affectivity was related to all three types of work/family conflict (time, strain, & behavior-based). Using a Hong Kong sample, Stoeva, Chui and Greenhaus (2002) found, among other things, similar results as Carls on did with her US sample. As a result of these studies, it does appear that work/family conflict is not just a western phe nomenon, but also an important stressor in the lives of people the world over. Measuring Work/Family Conflict Despite evidence of the widespread im portance of work/family conflict, the research to date has some gaps and oversi ghts. As previously stated, a lack of standardized measurement exists, which has resulted in theoretical (i.e., construct
8 clarification) and practical (i.e., comparing results from multiple studies) problems. Even after the delineation of the different dimens ions of work/family conflict by Greenhaus and Beutell (1985), nearly all of the literature has been built on measures that do not adequately sample the construct. For in stance, Frone, Yardley, and Markels (1997) model determined that each direction of wo rk/family conflict was related to the other indirectly by way of domain overload and dom ain distress, as well as being related to performance within the opposite domain. Ye t Frone et al.s measure of work/family conflict consisted of two items for each direction of the construct, making generalizability of the results difficult. In fact the vast majority of measures have either been self-developed (Frone, Russell, & Cooper, 1993 & 1994; Matsui, Ohsawa, & Onglatco, 1995; ODriscoll, Ilgen, & Hildreth, 1992; Rice, Frone, & McFarlin, 1992), developed without using psychometrically rigo rous procedures (Pleck, Stains, & Lang, 1980; Small & Riley, 1990), or failed to disti nguish between either the bi-directionality of the construct (Cooke & Rousseau, 1984; Greenglass, Pantony, & Burke, 1988; Stephens & Sommer, 1996) or its multi-dimensional nature (Netemeyer, Boles, & McMurrian, 1996). To date, only one measure (Carlson, Kacmar, & Williams, 2000) has both addressed the need for rigorous test development and the multidirectional/dimensional nature of work/family conflict. Table 1 presents a list of the primary measurement tools used in generalized work/family conflict research to date, star ting with three initia l studies conducted by Holahan and Gilbert (1979), Burke, Weir, and DuWors (1979), and Pleck, Stains, and Lang (1980). These early attempts at measuri ng the construct were loosely based on role theory, but generally were not written with an understanding of the bi-directional or
9 multi-dimensional nature of work/family conflict. The scales used in these studies had additional difficulties, such as having not been developed using rigorous psychometric practices (Holahan & Gilbert), containing items that were not designed to be combined to measure a distinct construct (Pleck et al.), or mixing multiple constructs (such as satisfaction and role overload) with interrole conflict (Burke et al.). However, the research formed the foundation from which future measures of work/family conflict could be built. For instance, using the 1977 Quality of Employment Survey (Quinn & Staines, 1978), Pleck et al. determined that there were seven diffe rent types of workfamily conflict. The authors further identif ied excessive work time, schedule conflicts, and fatigue/irritability as the three most pr evalent. This information provided a basic understanding of the types of generalized c onflict that can occur between work and family roles. Yet again, due to the lack of theory at the time, the scale was not constructed to measure the severity of occu rrence for each of the types of conflict the authors identified. Issues such as this w ould need to be addressed before a measure could be considered to truly sample the constr uct. What follows is a detailed description of the development, contributions, and defi ciencies of the scales used to measure work/family conflict. Bohen and Viveros-Long (1981). The development of a tool addressing the aforementioned deficiencies was first conduc ted by Bohen and ViverosLong in their seminal book, Balancing Jobs and Family Life: Do Flexible Work Schedules Help? In their work, the authors sought to address the e ffects of flextime on family stress, family work, and equity of spouse family roles. Of pa rticular interest to th e present study is their definition of family stress, which they stated was tension or pressure arising at the points
10 where peoples work and family roles connect or overlap (p. 100). To measure this they constructed two surveys, the Job-Family Role Strain scal e, which centered on general worries about fulfilling both work and family roles, and the Job-Family Management scale, which measured the ease/difficulty of managing family activities. Subsequent work/family conflict research has extensively u tilized the Job-Family Role Strain scale. The Job-Family Role Strain scale was specifically built to tap stress related to internalized values and emotions such as self-doubt, worry, guilt, and pressure but also feelings of contentment, fulfillment, se lf-respect, and balance in regard to job and family obligations (p. 233). The authors ba sed the items content on role strain work done by Komarovsky (1977, in Bohen & Viveros-Long, 1981), who determined that there are six modes in which tension may occur (reproduced in Table 2). Bohen and Viveros-Long utilized five of the six modes as the basis for the content of the items in the Job-Family Role scale, excluding congruity between a persons personality and a particular social role due to limitations in the scope and design of their study. Item generation for the Job-Family Role scale occurred in three stages: first by gathering and classifying statements relating to work/family strain from five independent studies and from interviews with a group of 10 families, followed by panel and expert review of the items generate d by these statements, and fina lly through a pilot study where reliability was determined and feedback on the items solicited. In the first stage, statements from five independent studies were classified into the five Komarovsky modes of role strain used in the study. Next, i ndividual and group conve rsations of varying length (10 minutes to one hour) were held with the fathers, mothers, and children of 10 different families. For fathers and mother s, these conversations centered on the role
11 strain involved in being succe ssful workers and parents, whil e for children they centered on discussing the kinds of stra ins they thought their parents fe lt. Statements from these conversations were then sorted into the fi ve modes and combined with those extracted from the five independent studies. Items were then written based on this database. In the second phase of item developm ent, the previously written items were presented to two groups of six federal employ ees. Each group reviewed the item content and discussed whether these statements refl ected their own feeli ngs and any different feelings that they had (p. 236). They were th en designated as part of the Role Strain or Management scale based on whether they conc erned internalized values, or feelings about the logistics of work/lif e role management. Next, the items were presented to a third group of six federal employees. These employees were considered expert judges and consisted of a psychologist, a sociologist, and three fede ral personnel experts. The panel rated the items according to how well they tapped the content of the scale. Items that adequately passed this phase of content an alysis were subsequently used in the pilot study. In the final phase of item development, a pre-test was conducted by administering the preliminary scale to 50 federal employ ees. An alpha of .71 was computed, and positive relationships were found with whethe r or not a spouse worked (r=.17), having the main responsibility in home chores (r=.13), number of hours spent doing home chores (r=.18), percent of respondent time spent doing home chores (r=.23), having the main responsibility with the children (r=.17), perc ent of time spent with the children (r=.24), perception of work/family inte rference (r=.52), age (r=-.22), and finally number of hours working and commuting (r=.11). Additional comments led to the deletion of a number of
12 items and the splitting of the scale into two distinct pieces, based on whether items were directed generally towards adults, and those it ems directed specifically to parents/dealing with children. The splitting of the items in to two surveys was done to increase the range of populations that qualify to take the scale. After elimin ation of some items due to content validity questions, final alphas of th e adult scale (.72) and the total scale (.71) were determined using a second sample (N=449) Internal consistency for the parenting items was not calculated. Lastly, a fact or analysis using the second sample was conducted on both the adult and parenting versions of the scale, as well as the total scale. From these, a final 16-item scale was constr ucted consisting of ten generalized items geared towards adults, and 6 more sp ecific items geared towards parents. Bohen and Viveros-Longs (1981) Job-Fa mily Role Strain scale was the first attempt to develop a measure of work and fa mily interrole conflict that was not based on frequency counts and whose development em ployed rigorous psychometric techniques. Its influence has been widespread, both in its use as a measure of work/family conflict (Bedeian, Burke, & Moffitt, 1988; Duxbury, Higgins, & Lee, 1994; Higgins, Duxbury, & Irving, 1992), and as a basis from which futu re scales were developed (Carlson, Kacmar, & Williams, 2000). However, the scale is not without shortcomings. To begin, factor analysis by the authors revealed a four-fact or structure that do es not fit the five Komarovsky (1977, in Bohen & Viveros-L ong, 1981) modes used in developing the items, or the bi-modal emphasis of adult vs. pare nt-oriented item content. In addition, the items tapped internalized feelings of guilt and worry (e.g., I worry whether I should work less and spend more time with my children), which may be due to actual negative spillover between the roles, or may result from personality-based issues, making it
13 difficult to distinguish between the differe nt constructs. Finally, the 16-item scale contains statements which tap generalized work/family conflict (I have a good balance between my job and my family time), role overload (I feel I have more to do than I can comfortably handle), family interference with work (I worry that other people at work think my family/friends interfere with my j ob), child interference with work (I worry about how my kids are when Im at work) and work interference with nonwork (My jobs keep me away from my family/friends too much). This combination of various aspects of work/family conflict, as well as aspects of role overload, may result in both attenuation and difficulty in explaining th e meaning of relationships with other constructs. Kopelman, Greenhaus, and Connolly (1983). The development of a work/family interference measure whose content contained a cleaner interpretation of the construct was addressed by Kopelman, Greenhaus, & C onnolly. The author's based their work on Kahn, Wolfe, Quinn, Snoek, and Rosenthal's (1964) dissemination of role theory, and more specifically, interrole conflic t. Kahn et al. indicated that with interrole conflict, "the role pressures associated with membership in one organization are in conflict with pressures stemming from membership in other groups" (p. 20). They also indicated that, "Demands from role senders on the job fo r overtime or take-home work may conflict with pressures from one's wife (or husband) to give individual attention to family affairs during evening hours" (p. 20). According to Kope lman et al., previous studies had used a variety of interrole conflict measures, many of which were open-ended or single-item questions (Beutell & Greenhaus, 1982; Jones & Butler, 1980; in Kopelman et al.). This,
14 they argued, indicated a need for an empi rically valid measure, if only to promote consistency across organizational studies. In this study, the authors developed meas ures of both work and family intrarole conflict, along with a scale for work interference with family. The items for interrole conflict were based on work done by Pleck, Stains, and Lang (1980). As previously stated, Pleck et al. determined that there were seven different types of work-family conflict. The authors further identified ex cessive work time, schedule conflicts, and fatigue/irritability as the three most prevalen t. Kopelman et al. chose to collapse these three categories, classifying interrole conflict as arising from limited time resources (time-based), or excessive strains (strain-based) associated with either of the roles. Factor analysis indicated that items loaded clean ly on a single factor for two independent samples, with loadings ranging from .32 to .69 for sample #1 (alpha = .70), and .38 to .85 for sample #2 (alpha=.89). Although the measure of inte rrole conflict developed by Kopelman et al. (1983) has been used extensively in the work/family conflict literature (Parasuraman, Greenhaus, & Granrose, 1992; Thomas & Ganster, 1995), pr oblems with both its development and its conceptualization exist. To be gin with, the author's admit that sample size (N=91) in the second of the two studies seriously compromise d the generalizability of results. In fact, the author's caution against interpreting the factor structure from the second study. Given that nearly half of the items listed in the fi nal scale were developed from the results of this second study, the final factor structure of the measure could be questioned. In addition, whereas Kopelman et als scale does sample three forms of interrole conflict, i.e., work time, scheduling, and excessive fa tigue, they nevertheless do not sample other
15 forms such as schedule uncertainty, work trav el, and vacation-related problems (Pleck et al., 1980). More importantly, the 8-item s cale contains no facet scores, making it impossible to estimate the relative contribution of each type of interrole conflict (timestrain-, or behavior-based) to work interference with family. Lastly and most importantly, Kopelman et al.'s scale only measures work interference with family, and not family interference with work. Burley (1989) addressed this deficiency by re-wording four of the sc ale items, reversing the direction of their focus, thus producing an ad hoc family inte rference with work scale (Table 3). Gutek, Searle, and Klepa (1991) tested this format using two independent samples. Findings indicated that internal consistency in both in stances was at acceptable levels (alpha levels of .79 and .83) for the family interference with work scale. Factor analyses were also conducted using each sample, revealing a dual factor structure wh ere items for work interference with family and family inte rference with work loaded cleanly on two separate factors. Finally, the two scal es correlated r=.26 and r=.10 in each sample, revealing that although they were related, they also appeared to measure distinct aspects of the work/family conflict construct. Howe ver, although this information is promising, the validity of Burleys (1989) family inte rference with work scale has never been adequately tested and, because the items were worded to mirror those in the Kopelman et al (1983) scale, any deficiencies within that measure would likewi se appear in this one. Wiley (1987). The primary contribution by Wiley to the measurement of work/family conflict was to provide statistical evidence of the bi-directional nature of the construct. Instead of developing an entire ly new scale, Wiley chose 22 of the 50 items used by Burke, Weir, and DuWors (1979). Items were chosen based on whether they
16 tapped the extent to which work interfered with both family and personal roles. A subsequent factor analysis revealed four factors, including role overload, job/person conflict (measuring general wo rk interference with non-work), job/family conflict (measuring work interference with family), and family/job conflict (measuring family interference with work). Intercorrelations between the three forms of conflict were relatively high, the largest be tween job/person and job/fam ily conflict (r=.54), followed by job/person and family/job conflict (r=.48) and finally by job/family and family/job conflict (r=.40). This appears to indicate that the scales ar e measuring distinct forms of work/family conflict, yet there is also a high degree of overlap between them. Subsequent analysis revealed significant relationships between all three forms of conflict and role overload (c orrelations ranging between .45 and .56); life satisfaction (correlations ranging between -.23 and -.28) ; job satisfaction (correlations ranging between -.12 and -.26); job involvement (c orrelations ranging between -.10 and .20); and organizational commitment (correlations rang ing between -.10 and -.24). In addition, when regressing outcome variables onto all f our independent variable s, significant beta weights were found for job satisfaction ( =-.32) and organizational commitment ( =-.34) onto job/person conflict; for re gressing job involvement ( =.35) and organizational commitment ( =.32) onto job/family conflict; an d for regressing job involvement ( =.24) onto family/job conflict. These analyses provide preliminary evid ence of the bi-direct ional nature of work/family conflict, as postulated by Gr eenhaus and Beutell (1985). However, the scales used in this study were not develope d with the benefit of this theory. More importantly, the items taken from Burke et al (1979) were not intended to be separated
17 into different scales, and thus their content most likely did not adequately sample the construct. Finally, internal consistencies of the job/family, job/person, and family/job scales were not provided, and additional construct validity was not pursued in depth. Therefore, although the work done in this st udy provides valuable information regarding the structure of the work/family conflict cons truct, it also highlighted the need for the development of a measure that is both psychometrically sound and that is built under the umbrella of current theory. Netemeyer, Boles, and McMurrian (1996). It would be another 9 years before Netemeyer, Boles, and McMurrian devel oped the first bi-directional measure of work/family conflict using rigorous psyc hometric procedures for both the work interference with family and family interferen ce with work subscales. For their scale, a pool of 110 items was initially generated from previous published sources and reduced to 43 (22 work interference with family, and 21 family interference with work) based on ratings from a panel of four judges from other universities. An important aspect of these items is that they did not reflect behavior-b ased work/family conflict. Instead, the items reflected general demand, time, and strain-b ased, work interference with family and family interference with work. The authors used three independe nt, non-student samples to determine the factor stru cture and criterion-based validity of their scale. These included elementary and high school teacher s and administrators (sample #1, n=182), small business owners (sample #2, n=162), a nd real estate sales people (sample #3, n=186). Within each sample, measure purifi cation was conducted by way of an iterative confirmatory procedure using LISREL VII (Jreskog & Srbom, 1989; in Netemeyer et al.). Using a series of five heuristics (i.e. completely sta ndardized factor loadings of
18 greater-than .50, highly redundant in terms of wording with other items, etc), items in the first sample were reduced from 43 to 24 (13 work interference with family, 11 family interference with work). A modified set of heuristics (i .e., still exhibited correlated measurement errors, had completely standard ized factor loadings greater-than .60, etc) were used to further reduce items from 24 to 13 (seven work interference with family, six family interference with work) in the second sample. Lastly, an additional three items were deleted in the third sample on the basis of author judgment. The final scale consists of five items measuring time and strain-bas ed work interference with family, and five items measuring time and strain-based fa mily interference with work. Internal consistency was high for both sc ales, ranging from .88 for work interference with family, to .89 for family interference with work. In addition to measure purification, Ne temeyer et al. (1996) used covariance structure modeling to compare a two-factor a nd single factor scale structure within each of the three samples. Results indicated th at across all samples a two-factor model of work/family conflict fit the data, and the sing le factor model did not. Item loadings ranged from .60 to .89 on a priori factors, a nd internal consistenc y ranged from .82 to .90 across samples. In addition to the validati on of the scales stru cture, criterion-based validity was determined by way of comparis ons with 17 additional variables. These included job/life/relationship satisfacti on, role conflict/ambiguity, and physical symptomology. Predictions using both direct ions and differences between the construct of work/family confli ct were confirmed. The strength of the Netemeyer et al. (1996) scale lies in the psychometrically rigorous fashion in which it was developed. This was demonstrated in their use of expert
19 judgment to reduce the item pool; employing three independent, nonstudent samples; and covariance structure modeling. However, the weakness associated with the measure include only partial adhere nce to current theory, and reliance on previously used measures for its own item pool. As previous ly mentioned, work/family conflict is both multi-directional and multi-dimensional (G reenhaus & Beutell, 1985). Whereas Netemeyer et al. concentrated on choosing items that reflected the multi-directional aspect of work/family conflict, they largel y ignored the multi-dimensional nature of the construct. This is reflected in their choice to use a two-factor structure in their analysis, as well as the choice to omit behavior-based conflict from the item-pool. In addition, the authors reliance on previous ly published scale items limited them to study-specific measures that even they admitted would often reflect potential outcomes of the constructs rather than their content doma in (p. 400). In summary, Netemeyer et al. demonstrated that their scale was both stru cturally sound and criter ion-valid, yet did not adequately sample the content domain. Stephens and Sommer (1996). The Netemeyer et al. (1996) scale provided critical evidence that work/family conflict is bi-directional. However, it was Stephens and Sommer (1996) who first investigated the mu lti-dimensional aspect of the construct. Using similarly rigorous psychometric procedures as Netemeyer et al., the authors sought to measure the time-, strain-, and behavior-based nature of work interference with family. After developing a pool of 28 items from a revi ew of the literature, a group of 47 subject matter experts (SMEs) were given definitions of the three types of conflict and asked to classify the statements accordingly. A final exploratory pool of 16 items remained after items that did not achieve at least 80% agreement among SMEs were eliminated.
20 The next step was to conduct an explor atory factor analysis on data from 300 employees of a large research hospital. Crit eria were established for identifying factors and retention of items. For determining the nu mber of factors, four criteria were used: Eigenvalues greater than 1, a sc ree plot test, a priori dete rmination, and percentage of variance explained. Using these rules, a three-factor solution emerged. The first factor consisted of 8 items, 4 items each that were hypothesized to measure timeand strainbased conflict. Meanwhile, th e second and third factors cons isted of three items each, all hypothesized to tap behavior-based work interf erence with family. More specifically, the second factor contained positively worded it ems, while the third factor contained negatively worded items. For re taining items, two criteria were used: Item loadings must exceed .45, and item loadings must differ by at least .1 between crossfactor loadings. From this 2 items were eliminated, leaving a scale structure consisting of three factors and 14 items. As a final step, the authors conducted a confirmatory factor analysis using responses from 273 employees gathered from a state government agency (n=145) and a scientific testing firm (n=128). Stephens and Sommer (1996) tested three models: A null model with a single latent va riable, a two-factor solution as found in the exploratory factor analysis (with positively and negatively worded behavior-based items combined), and a three factor solution where time-, strain -, and behavior-based items were influenced by separate latent variables. Both the null model (GFI = .34; AGFI = .24) and the twofactor model (GFI = .92; AGFI = .88; NFI2 = .871) failed to display adequate fit to the data. However, the three-factor solution cont aining the a priori latent variables of time-, strain-, and behavior-based conflict was found to acceptably f it the data (GFI = .95; AGFI
21 = .93; NFI2 = .986). Since chi-square goodness-of -fit tests tend to be overly sensitive to sample size and small departures from multivar iate normality (Bentler, 1990; in Stephens and Sommer, 1996), a chi-square to degrees-offreedom ratio was calculated. According to Hanisch and Hulin (1991, in Stephens and Sommer), a ratio of less than 2.0 indicates a positive model fit. Stephens and Sommer found the ratio for the three-factor solution to be 1.35. Although their exploratory factor analysis did not reveal the a priori factors of time-, strain-, and behavior-based conflict, the confirmatory fa ctor analysis did corroborate the authors three-factor soluti on. This was the first time someone had attempted to confirm the multi-dimensional nature of Greenhaus and Beutells (1985) theory. While not conclusive, the study doe s provide evidence that a multi-dimensional factor structure could exist, something no other study had done. Combined with the more conclusive evidence of bi-directionality in the construct (Netem eyer et al, 1996), the probability that measurement of work/family conflict should be done using a six-factor scale was high. Carlson, Kacmar, and Williams (2000). The act of producing a scale that covered both the bi-directional and multi-dimensional nature of work/family conflict was realized in this recent study. Similar to Netemeye r et al. (1996) and St ephens and Sommer (1996), Carlson et al. first developed an item pool based on previously published measures. A total of 31 non-redundant item s were generated from Bohen and ViverosLong (1981); Burley (1989); Duxbury, Higgins and Mills (1992); Frone, Russell, and Cooper (1992); Gutek, Searle, and Kleppa (1991); Koppelman et al. (1983); Pleck (1978); and Stephens and Somm er (1993). Items from other studies were excluded
22 because they specifically addressed job dema nds or nonwork conflict, which the authors deemed beyond the scope of their study. Us ing content analysis guidelines provided by Schriesheim et al. (1993; in Ca rlson et al.) and a sample of 236 undergraduates enrolled in an upper level business course, the authors selected 20 items from the initial pool of 31. The remaining items were then administered as a survey of work/family conflict to a second sample of 390 employees of a state government agency. Exploratory factor analysis revealed that all items loaded on onl y one of three possible factors, yet did not provide a comprehensive representation of each of the six conflict dimensions. For example, none of the items represented behavi or-based conflict, and only one represented the strain-based work interference with family dimension. To increase the scale content validity, Ca rlson et al. developed an additional 34 items based on review of the literature as we ll as on personal and anecdotal experience (p. 257). All 54 items were then presen ted to a sample of 132 MBA students to categorize and rate. For an item to be retain ed it had to be categor ized in the correct definition at least 70% of the time (using si x different definitions, one for each dimension of work/family conflict), and have a mean score of 3.5 (aga in, 70%) for the rating-content adequacy testing. This process reduced the item count to 33, with an additional three items removed for parsimony. The remaining 30 items represented the five best items for each of the six scales. Two additional studies were then conducted. In the first study, 228 graduates from an Executive MBA program were given the 30-item measure. A confirmatory factor analysis was then run to determine if a six-factor model fit the data, with each factor represented by five items in the scale. An additional 11 items were eliminated due
23 to either factor loadings found to be less th an .50, items being more strongly associated with any factor than the one for which it wa s intended, correlated measurement error, or wording redundancy. The process produced a fi nal scale of 18 items, three items for each of the six dimensions of work/family conflict. In the second and final study, the dimensionality, reliability, discriminant validity, factor structure, gender diffe rences, and differential predic tions of the new scale were investigated. A new, independent sample of 225 individuals who were employed fulltime in various organizations was collected to help conduct these tests. Dimensionality was assessed by testing six, three, two and one factor models using confirmatory factor analysis, with items forced to load on sp ecific factors and factors being allowed to correlate. Results indicated that the six-factor model had the best fit [ 2 (120) = 237.40, p=.00; CFI=.95, RMSEA=.06], and all factor load ings were significant (ranging .69-.91). In addition, the internal consistency of each factor ranged between .78 and .87. Factors were found to correlate between .24 and .83, with only two correlations above .60, indicating discriminant validity among the six different factors. The factor structure of the scale was then tested for invariance across samples. Both the current sample and the sample from th e initial confirmatory factor analysis were used to compare four two-group models. M odels varied according to whether factor loadings, factor correlations, and error varian ces were held constant or allowed to vary freely. Generalizability across samples was confirmed when two different data sets mapped well to the model (Carlson, Kacmar, & Williams, 2000). Specifically, the only instance where the factor structure did not hold across samples was the most constrained model. However, invariant e rror variances are considered th e least important in testing
24 factor structure invariance across groups (Marsh, 1995; Williams, Bozdogan, & AimanSmith, 1996; in Carlson, et al.), and because the fit indices indicated adequate fit across the most constrained model, evidence of measurement invariance across samples was considered to be sound. This further confir med the six-factor structure of the model. Similar procedures were used to determin e if the six-factor model was invariant across males and females. For this analysis, th e most recently collected sample was used. As with the factor structure analysis, four two-group models (males vs. females) were again compared by varying the degree of cons traint on the factor loadings, correlations, and error variances. Unlike the overall factor structure, differences across gender were discovered when the factor correlations and e rror variances were held invariant. Further analysis revealed that the pa ttern of factor correlations was similar, even if the correlations themselves were different. For instance, the average factor correlations for males was .47, and .45 for females. In additi on, two-thirds of the individual differences were less than .20 and the largest difference be tween correlations was .37 (Carlson et al., 2000, p. 266). A set of t-tests were then conduc ted, revealing that females were found to experience significantly more conflict than men for all three forms of family interference with work, as well as for strain-based work interference with family. Carlson, et al. hypothesized that the inconsistent findings in past research on gender differences may be explained by the fact that females are more likely to experience more conflict than men on only some, not all, forms of conflict (p. 267). Finally, Carlson et al. (2000) used both wo rk and family-oriented measures for the antecedents of role conflict, role ambiguity, social support, and role involvement with the outcomes of job satisfaction, family satisfa ction, life satisfacti on, and organizational
25 commitment to test differential predictions of work/family conflict. Two path models were examined: In the first, all three forms of work inte rference with family conflict were compared with the four work related ante cedents and the four outcome variables. In addition, direct paths from the antecedents to the outcomes were also examined. The second model consisted of the three forms of family interf erence with work, which were compared to the four family-related antecedents and four outcomes. Direct paths from the antecedents to the outcomes were also examined in this model. Significant path coefficients were used to determine relationships. Each of the six dimensions was found to differentially relate to both antecedents (role ambiguity, role conflict, social support, and role i nvolvement) and consequences (job/family/life satisfaction, and organizational commitment). For work interference with family, all four antecedents were related to strain-based conflict, work ambiguity and work involvement were related to behavio r-based conflict, and only work involvement was related to time-based conflict. Meanwhile strain and behavio r-based conflict were negatively related to the outcomes of job a nd life satisfaction, and time-based conflict was not related to any of the outcome variable s. For family interference with work, all four antecedents predicted behavior-based conflict, and only family-role conflict and social support predicted time and strain-based conflict. In addition, strain-based conflict predicted job, family, and life satisfaction ( not predicted by any of the other family interference with work scales), whereas behavior-based conf lict significantly predicted organizational commitment. Overall, Carlson et al. (2000) demonstrat ed that six separate dimensions could represent the work/family conflict construct. The scale they developed to measure these
26 shows discriminant and criterion-based validit y, as well as a solid factor structure across two independent samples. In addition, the au thors demonstrated that males and females are affected in different wa ys by experiencing work/family conflict. This finding may provide evidence as to why some authors find differential permeability between work interference with family and fa mily interference with work. Summary of work/family c onflict measurement. Over the past 25 years, the assessment of work/family conflict has evolve d from measures that treated the construct as a singular facet and often contained items that overlapped with ot her constructs (e.g., role ambiguity, job satisfaction) to scales th at are able to differentiate between both the direction and the type of c onflict being measured Although all of these scales tap generalized feelings of conflict (as opposed to role-spe cific perceptions), research has nevertheless indicated they are different from one another. Scales that treat work/family conflict as a singular facet have different predictive capabilities than those that treat it as a multifaceted construct. As a result, con tinued research into both the theoretical underpinnings of work/family conflict, and the predictive capabi lities of current measurement techniques, is warranted. Testing Work/Family Conflict Measurement Equivalence Across Mu ltiple Populations Carlson et al. (2000) provided a major step forward in the measurement of work/family conflict. The authors were able both to sample the content domain appropriately and to provide evidence of a high degree of reliability and validity. They also provided the first measure of Greenhaus and Beutells (1985) model of work/family conflict as a bi-directional, multi-dimensional construct that fit across several independent samples. In addition they f ound that each dimension expresses differential
27 relationships with outcome variables. Given these findings, an appropriate next step in both the evolution of this m easure and the evolution of th e measurement of work/family conflict in general would be to examine its structure across a multitude of populations, particularly those from other nationalities a nd cultures. Recall that individual studies have provided evidence of the cross-cultural nature of work/family conflict (Aryee, Fields & Luk, 1999; Matsui, Ohsawa, & Onglatco, 1995). Although these studies test ed the bi-directional nature of the construct, they did not include measures of its multi-dimensionality. To test whether the full six-factor structure of work/family conf lict will hold outside of the US, a measure of the six subgroups first needed to be built. Carlson et al. provided such a sc ale, and its continued validation will lay the foundation from which the current model of work/family conflict can be tested. To that end, the purpose of th is study was to further validate the Carlson et al. scale by testing for measurement invari ance between a US and a foreign Englishspeaking sample. In this case, the forei gn English-speaking samp le was obtained from Australia and New Zealand. Measure structure vs. stimulation. By using a foreign English-speaking sample, problems with language translation are minimi zed and the structure of the measure can better be tested. This is particularly he lpful because vast differences exist between testing the cross-cultural viability of a meas ure (i.e., measurement invariance, also known as measurement equivalence) and developing cr oss-lingual versions of it. According to Hulin (1987), the goal of equivalence is to provide an equivalent structure to the material to achieve equivalent stimuli rather th an equivalent structure as an end in itself (p. 117). In other words, dire ctly translating a m easure often results in confusion due to
28 differences in how the target language expresse s the concepts elucidated in the original wording of the items. Although this may repli cate the structure of the measure, it does not guarantee a replication of a given scale s ability to stimulate answers that provide information on the construct of interest, which is what Hulin referred to as stimuli. Yet testing for equivalent stimuli before adequa tely testing the cross-cultural structure of a new measure may be putting the proverbial cart before the horse. Thus, you have what appears to be a catch-22: a test for equiva lent structure may be hampered by a lack of equivalent stimulation, and testing for equiva lent stimulation may be hindered by a lack of equivalent structure. Th e only apparent solution is to simultaneously test for stimuli and structure at multiple points in the measure development process, a procedure that dramatically increases time and cost requireme nts. Although this may be a scientifically desirable solution, it is often not practical. A second solution would be to test meas urement equivalence using samples with fewer cultural differences as a first step. By reducing cultural differences, the structure of a measure may be tested in non-US sample s with fewer questions about equivalent stimulation of the construct. This could e ffectively remove the alternative hypothesis that equivalent stimulation was not adequately achie ved, and thus the crosscultural version of the measure is not valid, regardless of struct ural strength. Such a test could provide evidence of structural cross-cu ltural equivalence. Once this is established, equivalent stimulation could become the focus of further research. Cultural similarities between the Unite d States, Australia, and New Zealand. To void the alternative hypothesis that equivalent stimulation was not achieved, an Englishspeaking sample that is cultura lly similar to the US must be used. For the present study,
29 a sample from Australia and New Zealand was used to determine wh ether the Carlson et al. (2000) scales six-fa ctor structure holds in a non-US sample. Similarities between Australia, New Zealand, and the United States have been well documented. For instance, Fallon (1996; in Fallon, 1997) successfully repl icated the Frone et al. (1992) model of work/family conflict in an Australian samp le, in which job/family involvement and job/family stressors influenced work/family conflict, which in turn affected job/family distress and individual leve ls of depression. Other st udies have provided more generalized information. For instance, a Gallup pole (1976; in Ouweneel & Veenhoven, 1991) found that mean self-report levels of happiness were identical for Australian and US groups, as were the standard deviations in the samples. Sagie and Schwartz (1994) used a measur e of dispersion to determine similarity between cultures. In their study of grade school teachers in 36 nations, the authors calculated the level of variance in ratings of value importance to determine value consensus in a population. Speci fically, the participants rate d 56 single values according to their importance as a guiding principle in my life (p. 220). Th e ratings were then combined into ten broad values based on a typology by Schwartz (1992; in Sagie & Schwartz): benevolence, trad ition, performance, security, power, achievement, hedonism, stimulation, self-direction, and universalism. Variances calc ulated on these broad value types were then averaged to determine a nati ons overall level of value consensus, where lower variance meant higher levels of congrue nce. Interestingly, the value consensus ratings for Australia, New Zealand, and the United States fell within .14 of each other. In addition, participants from all three nations rated their country as high in socioeconomic modernization and democratizati on of the political system. These ratings i ndicate that
30 they are similar not only in value consensus, but that they are dissimilar to nations with higher value consensus as well. Specifi cally, higher levels of socioeconomic modernization and political democratizati on are related to lower levels of value congruence. On the other hand, higher leve ls of modernization and lower levels of democratization (e.g., total itarianism) lead to higher levels of value consensus. Thus, not only are Australia, New Zealand, and the United States similar to each other, they are also different than other nations in similar ways. Finally, similarity between these three nations can also be found in the study of cultural dimensions. According to Hofstede (1997), dimensions of culture are aspects that can be measured relative to other cu ltures (p. 14). Draw ing largely from his seminal work involving 50 nations and 3 world regions, Hofstede (1997) found many similarities between Australia, New Zealand, and the United States on his dimensions of power distance, individualism/collectivis m, masculinity/femininity, uncertainty avoidance, and long-term orie ntation variables. For inst ance, Hofstede defined power distance as the extent to which less powerfu l members of institutions and organizations within a country expect and accept that pow er is distributed unequally (p. 28). In countries with a low level of power distance, supervisors have a consultative style of decision-making, where they speak with subordi nates before reaching a decision. On the other hand, in countries where a high level of power distance exists, subordinates have a dependent relationship with th eir supervisors, where they are not consulted and instead are lead in an autocratic way. When compari ng the nations of interest to this study, on a 100-point scale, all three scored in the lower 40th percentage on power distance and were within 18 points of each other.
31 Hofstede (1997) found similar results when comparing the three nations on levels of individualism/collectivism. Individualism pertains to societies in which the ties between individuals are loose: ev eryone is expected to look af ter him or herself and his or her immediate family, and collectivism is c onsidered the opposite of individualism and pertains to societies in which people from birth onwards are in tegrated into strong, cohesive groups, which throughout peoples lifetime continue to protect them in exchange for unquestioning loyalty (p. 51). As with power distance, Australia, New Zealand, and the United States all scored extremely high on the Individualism Index (IDV), within 12 points of each other on the 100-point scale. The similarities among these three nati ons are even more pronounced in the masculinity/femininity, uncertainty-avoidance, and long-range orientation dimensions. Hofstede (1997) defined masculine societies as places in which social gender roles are distinct (men are assertive, tough, and focu sed on the material aspects of living, while women are modest, tender, and concerned with the quality of life), and feminine societies as places in which gender roles overlap, e .g. men and women can be concerned with the quality of life. Meanwhile, Hofstede defi ned uncertainty avoidance as the extent to which the members of a culture feel threat ened by uncertain or unknown situations, and long-range orientation as the extent to which a society promotes the values of persistence, thrift, ordering relationships by status and observing this order, and having a sense of shame. This is opposed to promoting short-term orientation values, which include personal steadiness, protecting your face, re spect for tradition, and reciprocation of greetings, favors, and gifts. For Australia, New Zealand, and the Un ited States, scores on these three dimensions on a 100-point scale were within four, five, and two points,
32 respectively. This, along with the ve ry similar scores on power distance and individualism/collectivism, indi cates that the three nations are essentially the same in their levels of these five dimensions. One final similarity between these three coun tries is in their laws regarding family leave. In the US, the Family and Medical Leave Act of 1993 states that a person may take up to 12 weeks of unpaid leave for a ne wborn or newly adopted child. Individuals may then return to work to eith er their same job or a job of si milar status and salary level. In addition, many companies offer paid leave of absences during this timeframe, with the number of businesses offering such benefits growing from year to year. On the other hand, Australian and New Zealand law indicates th at a person may take parental leave for up to 52 weeks, also unpaid. However, unlike in the US, few if any private companies offer paid leave during this timeframe (S Garvey, personal co mmunication, January 22nd, 2003). Also unlike the US, Australian state governments often offer some kind of paid leave to their workers, and do provide a onetime maternity payment at the time of birth. Therefore, although the Austra lia/New Zealand time for unpaid leave is longer, its benefits are mitigated by a lack of compen sation. For this reason it is likely that individuals on parental leave in Australia/Ne w Zealand return to work within a similar time period as those on leav e in the United States. In summary, research indicates that the cultures of the United States, Australia, and New Zealand appear to promote similar va lue systems. This includes similar levels of value congruence, as well as levels of cultural dimensions, as defined by Hofstede (1997). In addition, between Australia and th e United States, a model for explaining the antecedents and consequences of work/family conflict was found to replicate in both
33 nations, a particularly relevant finding to the current study. Finally, the parental leave laws for all three nations are sufficiently similar to warrant the assumption that individuals return to work after approximatel y the same amount of time on leave. More to the point, the United States, Australia, a nd New Zealand all offer unpaid leave, and all have ever-increasing numbers of women in the workforce (Fallon, 1996), leading to greater numbers of dual-career couples, and the need for dualearners in a given family. Because of these results, it appears that struct ural measurement invariance can be tested in the Carlson et al. (2000) measure of work /family conflict with a significantly reduced threat of non-equivalent stimula tion as an altern ative hypothesis. C ross-cultural gender issues in work/family conflict. Although measurement invariance across cultures of the Carlson et al. (2000) scale would provide support for the usefulness of the tool as a measure of work /family conflict, its examination would be incomplete without some test of its ability to detect differences within cultures. One particularly salient test w ould be to determine whether men and women have similar reactions to work/family conflict. This is relevant since ge nder differences in work/family conflict have been the fo cus of a multitude of studies. Gutek, Searle, and Klepa (1991) addresse d the issue of gender by testing two separate models of work/family conflict. Th e authors looked at the rational model, where conflict is linearly related to the total amount of time spent in family and work, and the gender-role model, where gender role expecta tions are believed to affect reported levels of conflict. In other words, gender is hypot hesized to either aff ect conflict directly, and/or to moderate the relations between the number of hours spent in a specific domain (work vs. family) and conflict. Therefore the hours an individual spends in their gender-
34 congruent domain (e.g., more hours for men at their job and for women doing familyrelated work) will have less of an impact on conflict than hours spent in a non-congruent domain (e.g., men doing more family-related work). Results indicated that both the rational and the gender role models explained the connection between hours in a domain and repo rted conflict, with more support for the rational perspective. For inst ance, in both studies there wa s a fairly large relationship between hours spent in a speci fic domain, and conflict arisin g from that same area (e.g., the more hours worked, the larger the reported levels of work interference with family). Nevertheless, women did report significantly stronger levels of work interference with family than men did, despite the fact that bo th groups, in both studies, reported similar amounts of time spent in paid labor. This is reflected in the significant variance explained in the gender-by-hours interaction term entered into the regression equation to predict work/family conflict. In other wo rds, although the rational model received full support in their study, some of the variance was better explained by domain-specific gender expectations. Further evidence of gender differences in reported work/family conflict can be found in Borovsky and Stepanskis (1999) me ta-analytic review. The authors first gathered 15 studies containing da ta on six variables, including wo rk role stressors, family role stressors, work/family conflict, j ob satisfaction, family satisfaction, and life satisfaction. These correlations were then meta-analytically combined producing two sets of 15 meta-analytic correlations; one set ea ch for males and females. A series of Ztests across gender revealed si gnificant differences between all of the 15 correlations. Next, using a model presented by Bedeia n, Burke, and Moffett (1988), the authors
35 analyzed gender differences by comparing m odel fit between men and women on the six variables. This was done by creating a ma trix out of each set of meta-analytic correlations, and subsequently analyzing them with structured equation modeling (SEM). More precisely, the two matrices were firs t compared with an unconstrained model, where male and female path coefficients were estimated separately. The matrices were then compared in a constrained model, where path coefficients were forced to equality for both genders. A significant ch i-square difference between th e two models indicated that the unconstrained model was a better fit to th e data. In other wo rds, it appears that gender does affect the relationships in the model. The results found by Borovsky and Stepan ski (1999) were not conclusive. Although a significant difference was found between the chi-squares, th ey also reported that the overall fit of the model to the enti re dataset (that is, in cluding both males and females) was moderate at best. This supports the findings of Gutek et al. (1991), in that gender as a moderator may explai n additional variance, but not in all circumstances. In addition, Borovsky and Stepanski did not ha ve enough studies to use a bi-directional measure of work/family conflict, let alone a measure that also tapped the multidimensional nature of the construct. This l eaves the possibility that the findings were entirely due to one dimension or directi on, as was found with Gutek et al. (1991). However, additional support for gender differences does exist. Burley (1995) found that women exhibited greater levels of work interference with family than men. Meanwhile, Duxbury, Higgins, and Thomas (1996) found a significant main effect with gender on a generalized conflict s cale, but failed to find any in teractive effects with their primary variable: computer-support s upplementary work at home.
36 Finally, this difficulty in determining the nature of gender differences in work/family conflict is also present in the resu lts found by Duxbury and Higgins (1991). Using an expanded model originally concep tualized by Kopelman et al. (1983), the authors hypothesized that work /family expectations, work c onflict, family conflict, and work/family involvement would all act as ant ecedents to work/family conflict. In turn, conflict would directly affect quality of work/family life, and through these, indirectly affect life satisfaction. From this model they hypothesized numerous gender differences, including: the relationship between work i nvolvement and conflict will be larger for women than men; the relationship between fa mily involvement and conflict will be higher for men than women; the relationship between work-expectations and conflict will be stronger for men than women; the rela tionship between family-expectations and conflict will be stronger for women than men; the relationship betw een work-conflict and work/family conflict will be stronger for men than wome n; the relationship between family-conflict and work/family conflict will be stronger for women than men. The authors collected data on 240 individua ls who were either married or had children living at home. They then used a partial least-squares analysis to determine gender differences between generalized work/family conflict and a number of different variables. Results indicated that higher levels of involve ment in a non-gender sanctioned role (e.g., work involvement fo r women, family involvement fo r men) resulted in greater levels of work/family conflict. In addition, quality of a non-gender sanctioned role had a negative relationship with work /family conflict, such that men had lower quality of family life, and women lower quality of work life, when conflict was high. These results build on the gender model tested by Gutek et al. (1991). Ye t as with that study, other
37 findings in this study paint a more complicated picture of the relationship between gender and work/family conflict. For instance, a stronger relationship between work expectations and conflict was found for men than women, yet gender differences could not be tested on family expectations and conflict because no significant relationship between the variables was found for women. In addition, women had a stronger relationship between family conflict and work/family conflict than did men, but no gender differences were found when compari ng the relationship between work conflict and work/family conflict. These results appear to conflict with thos e regarding work and family involvement, making it difficult to determ ine a clear picture of gender similarities and differences when inter-role conflict is measured. Other studies also bear out the compli cated picture of gender and work/family conflict. For instance, Frone, Russell, and Cooper (1992) developed a model of antecedents and consequences to work/family conflict that included variables similar to those studied by Borovsky and Stepanski (1999), including job/family distress, job/family involvement, work/family conflict, and depressi on (used as a measur e of life satisfaction by Borovsky and Stepanski). Yet when co mparing males and females in a multi-group analyses utilizing SEM, no gender differences we re found, regardless of the fact that a bidirectional measure of conf lict was included. Eagle, Miles, & Icenogle (1997) specifically tested for gender differences usi ng three different bi-d irectional scales of work/family conflict, and also found no gender differences. In summary, many studies have found gender differences (Borovsky & Step anski, 1999; Burley, 1995; Gutek et al., 1991) and many others have found no differences at all (Frone et al; Eagle et al). This
38 leaves open the question of when, and under what circumstances, gender differences exist. A clearer picture of gender differences in work/family conflict may become apparent with proper sampling of the cons truct, namely by measuring both its bidirectional and multi-dimensional nature. As previously stated, Carlson et al. (2000) provided a test of gender differences in thei r scale development. The authors found that females experienced greater levels of tim e-, strain-, and behavior-based family interference with work, as well as greater leve ls of strain-based work interference with family. In addition, the aut hors found that, although the f actor loadings were similar between men and women, the factor corre lations and error variances differed significantly. However, these findings do not ad dress the possibility that there are crosscultural dissimilarities in the relationship between gender and conflict, even if the different cultures are similar in levels of masculinity/feminin ity (Hofstede, 1997). In summary, gender differences may al so occur in non-US, English-speaking samples, yet this does not necessarily mean that the differences will be the same as those found in English-speaking samples in the Unite d States. However, given that there are cultural similarities between the United States and the Australian/New Zealand sample to be used in this study (Hofsted e, 1997), it is reasonable to predict that males and females have differential relationships in time-, strain -, and behavior-based family interference with work, and work interference with family. Purpose of the Current Study As previously stated, the purpose of this study was to validate further the Carlson et al. (2000) scale by testi ng for measurement invariance between a US and a foreign
39 English-speaking sample. In this case, the foreign sample consisted of a group of fulltime employees in the Australian and New Z ealand branches of an S&P 500 company. As noted earlier, the similar ity between these English-speak ing countries and the US is extensive enough to mitigate the alternative hypo thesis of appropriate stimulation when testing for measurement invariance in attit udinal scales. However, in this instance directly testing one sample agai nst the other may be presumptive. Carlson et al.s results indicate that gender differences in factor correlations and error variances did occur in the same US sample to be used in this study. As a result, should any differences be found between the Australian/New Zealand group and its US counterpart, it would be impossible to determine whether they originated from dissimilarities in culture, or dissimilarities between men and women. Ther efore, to determine whether measurement invariance exists between the US and Austra lian/New Zealand groups on the Carlson et al. work/family conflict scale, cross-cultura l intra-gender comparisons were conducted. Specifically: Hypothesis 1a: An equivalent six-factor stru cture of Carlson et al.s scale will exist for males in the English-speaking Australian/New Zealand sa mple and males in the English-speaking sample from the United States. Hypotheseis 1b: An equivalent six-factor struct ure of Carlson et al.s scale will exist for females in the English-speaking Australian/New Zealand sample and females in the English-speaking sample from the United States.
40 Again, given the similarities between th e cultures, it was also expected that dissimilarities between men and women woul d exist in the Australian/New Zealand sample itself. Specifically: Hypotheseis 2: Gender differences in work/family conflict will exst in the English-speaking Australian/New Zealand sample. Although differences likely exist in how men and women experience work/family conflict in Australia and New Z ealand, what those differences ar e is difficult to predict. In other words, while similarities with the United States across cultural dimensions may allow us to predict that gender differences exist at a macro le vel, the specific nature of those differences is not so easily determin ed without direct study. Because research using this population is not available on this topic, an exploratory investigation was addressed. Specifically: Research Question 1: If gender differences do exist in the Eng lish-speaking Australian/New Zealand sample, are those differences the same as the ones f ound by Carlson et al. in their sample from the United States? As previously stated, the current stud y aimed to investigate these research questions using archival data gathered from the United States and Australia/New Zealand. Data gathered in the United Stat es (the validation sample) consists of one sample used by Carlson et al. (2000) in their original valida tion study. Data gathered in Australia/New Zealand (the comparison sample ) was part of a larger data set acquired during a local validation study by an S&P 500 company.
41 Method Participants Participants for this study consisted of two independent samples collected at an earlier date. One of the data sets was us ed by Carlson et al. ( 2000) in the original validation study. The second sample is a comparison sample consisting of archival data gathered during a local validation study in which employees within the Australian and New Zealand division of an S&P 500 company were invited to participate. Data were collected between July and August of 2001. The study used several measures, including the full Carlson et al. (2 000) measure of WFC. Validation sample. The second set of participan ts from the third study in the original Carlson et al. (2000) validation rese arch was employed as the validation sample in this paper. Because the responses of these participants were not used for item analysis/elimination, or final scale structure de velopment, they are essentially unbiased in representing the structure of the work/family conflict scale. The sample was gathered using a snowball strategy, where a group of i ndividuals in a night class for working undergraduates were given the surveys and th en asked to distribute them to 5 other working individuals. Instructions for taking th e survey were as follows, Please indicate the degree of your agreement with each of the following questions regarding your work and family life by circling the appropriate response, where one is strongly disagree and five is strong agree. Finally, participan ts were employed in nu merous organizations across a wide range of jobs. The sample consists of 222 individuals, 83 males and 139 females. On average the males were 37.25 years old, had an averag e job tenure of 6.44 years, and on average
42 had 1.3 children and .90 living with them. Roughly 3.6% were Asian, 12.0% African American, 1.2% Hispanic, 81.9% White, and 1.2% other. In addition, 16.9% were single, 72.3% were married, and 10.8% were divorced. For females, the average age was 35.42 years old, average job tenur e 6.35 years, and on average 1.34 children and 1.05 living with them. Roughly 2.2% were Asian, 8.6% African American, 0% Hispanic, 87.8% White, and 1.4% other. In addition, 24.5% were single, 58.3 % were married, 13.7% were divorced, and 2.9% were widowed. Comparison sample. A total of 2,394 employees within the Australian and New Zealand division of an S&P 500 company were i nvited to participate in a local validation study. The company was engaged in recruitment and placement services, as well as consulting and marketing. Respondents include d managers, secretaries, recruiters, and other individuals in white-collar jobs. Instructions for taking the survey were as follows, The following questions will ask you about your attitudes and behavior at work and in your life in general. They are for research purposes only, so please answer them as honestly as possible. All answers will be kept anonymous, and your responses will not be made available to individuals in the company outside of the research group. Of the 2,394, 579 took the Carlson et al. (2000) work/family conflict scale, resulting in a 24% response rate. An addi tional 187 responses were excluded due to incomplete answers, resulting in a total samp le of 392 full-time employees. Within the usable sample, 287 (73.2%) were from Australia, and 249 (63.5%) were female. For Australian males (n=105), participants were on average 35.35 years old, had held at least 4 full-time positions in their lifetimes, held th eir current positions and been a part of their current organization for at leas t one year but less than thr ee years, and 82% claimed that
43 their current position was representative of their professional fiel d. For New Zealand males (n=34), participants were on average 38.38 years old, had held close to 5 full-time positions in their lifetimes, held their current positions and been a part of their current organization for at least one year but less th an three years, and 89% claimed that their current position was representative of their professional field. For Australian females (n=173), particip ants were on average 29.50 years old, had held almost 4 full-time positions in their lif etimes, held their current positions and been a part of their current organizati on for at least one year but le ss than three years, and 72% claimed that their current position was repres entative of their professional field. For New Zealand females (n=34), participants were on average 29.92 years old, had held over 4 full-time positions in their lifetimes, held thei r current positions and been a part of their current organization for at leas t one year but less than thr ee years, and 86% claimed that their current position was representa tive of their professional field. Additional information about Australia n demographics can be found at the Australian Bureau of Statistics (2001). Specifically, on average females give birth to 1.75 children, 41.2% of all families have chil dren under the age of 15, median age of marriage for males is 28.5, and the median age of first marriage for females is 26.7. Of all couple-families, 78.3% of them have children under the age of 15, and 56.7% are dual-earner families. Similarly, Statisti cs New Zealand (http://www.stats.govt.nz/) reports that in 2001 women gave birth to 1.97 children on average. In addition, 36.1% of all families in the country have children under the age of 15, the median age at marriage for bachelors is 29.30 years, and the medi an age marriage for spinsters is 27.49.
44 Finally, of all couple-families, 91.10% of th em have children under the age of 15, and 86.6% are full-time dual-earner couples. Measures Work/family conflict. As previously stated, work /family conflict was measured using Carlson et al.s (2000) 18 item scale (Table 4). The sc ale consists of six subscales (3 items each), measuring time-, strain-, a nd behavior-based conflict for both work interference with family and family interferen ce with work. An example of an item from the time-based family interference with work scale is, I have to miss work activities due to the amount of time I must spend on family responsibilities. An example item from the time-based work interference with fam ily scale is, My work keeps me from my family activities more than I would like. Alpha reliability estimates for the validation study ranged from .68 (behavior-based work interference with family) to .83 (strainbased family interference with work), and from .78 (behavior-based work interference with family) to .92 (time-based work inte rference with family) for the comparison sample. For the validation sample, a 5-point Likert scale was used ranging from strongly disagree (1) to strongly ag ree (5), whereas the comparison sample was given a Likert scale ranging between strongly disa gree (1) and strongly agree (6). Procedure Archival data for the validation sample were supplied by from Dr. Carlson, who provided an American sample originally used for item purification. Next, data from a local validation study conducted by an S& P 500 company on their Australian/New Zealand division were acquired. These data incl uded responses to the full Carlson et al. (2000) work/family conflict scale, as well as several demographic items. Data collection
45 occurred over a ten-week period between late June and early August of 2001, starting with an invitation to participate by the direct or of the Australia/New Zealand offices to some 2,394 full-time employees. Employees we re given periodic reminders (every 2-3 weeks) of the option to participate and were asked to complete the study by a specific date. As previously stated, both samples were measured on a strongly agree/strongly disagree Likert scale, with the validation sample scale ranging from 1 to 5, and the comparison sample scale ranging from 1 to 6. Due to the discrepancies between the answer format, all scores were analyzed using a completely standardized solution. According to Jreskog and Srbom (1996), a co mpletely standardized solution is one in which both the latent and observed variables ar e standardized for analysis. In addition, the comparison scale contained a Not appli cable choice. Individuals who chose this option for any of the questions were removed from the current studys sample due to the difficulty in determining the reasons for thei r choice. In other words, without a full understanding of the reasons for the choice, imputation of substitute scores would dramatically increase the probability of corrupting the comparison sample data set. Because power (sample size) did not appear to be an issue, exclusion was thought to be the most prudent decision. The final sample sizes are noted in the participants section. Design and Analysis Multi-sample confirmatory factor analys is (also known as a simultaneous factor analysis) was used to determine measurem ent invariance of the work/family conflict scale (Carlson et al., 2000) acros s the two independent samples. A diagram of the overall factor model is presented in Figure 1. M easurement invariance (also called measurement
46 equivalence) was evaluated through a series of comparisons between the groups, and was based upon the fundamental covariance equati on for confirmatory factor analysis, namely: Where is the variance/covariance matrix of a ll the observed variables in the dataset, is the matrix of item factor loadings, is the variance/covariance matrix of the latent variables (factors), and is the diagonal matrix of uni que variances. This equation is the model describing observed item covari ances as a function of common and unique factors (Vandenberg & Lance, 2000). The steps for testing measurement equivalence come from comparing the different matrices and their components delineated in the equation, increasing in specificity as need ed. For this study, steps comparing the covariance matrices, factor structure, factor loadings, and error vari ances are planned to test for measurement equivalence between the US and Australia/New Zealand populations. The sequence consis ted of two separate tests: (1) between males in the US and males in the Australia/New Zealand sample, and (2) females in the US and females in the Australia/New Zealand sample. Next, similar tests were conducted between males and females in the Australian/New Zealand group to examine Hypothesis 2. Finally, given structural similarities between gende r groups in the Australian/New Zealand sample, t-tests were used to determine wh ether group differences existed in a pattern similar to that reported by Carlson et al. (2000).
47 Identification of the Factor Model Identification of the statistical (in this case, factor) model in structural equation modeling involves determining whether there is a unique set of parameters associated with the data (Byrne, 1998). The first requ irement for identification of the statistical model is to determine a basic model for test ing. Carlson et al. (2000) provided this, fulfilling the initial step. A second requirement (linked to the first) is the need to determine the scale of every latent variable within the model. B ecause latent variables are unobserved, the scales may be estimated usin g the measurement model. This is often accomplished by fixing one manifest variables factor loading parameter to a non-zero number, usually 1.0. This indi cator variable then becomes a reference variable for the estimation of the associated latent variables scale. Hence, in the Carlson et al. scale, each of the six latent variables would have one of their three item-loadings fixed to 1.0. Although this is an accepted practice fo r single-sample confirmatory factor analysis, it provides unique obstacles in determining measurement invariance across multiple groups. By confining a specific f actor loading to 1.0, the loadings of those variables are essentially set to be equivalent, even though they have yet to be tested for equivalence. In addition, th e smaller the number of manifest variables for each latent variable, the larger the impact of this practice in determini ng invariance between groups. For instance, in the current study, using a reference variab le for model identification would result in the assumption that 33% of the variables (six out of the 18 manifest variables, or one from each of the six latent variables) have invariant loadings across groups. Because they would be required fo r model identification, these six loadings could not be statistically test ed for actual measurement invariance. Therefore, using a
48 different approach that provi des fewer restrictions on te sting parameter measurement invariance would be beneficial. A second approach to determining model id entification is to di rectly standardize the latent variables. This procedure eliminat es the need for reference variables, allowing all factor loadings to be freely estima ted (Joreskog & Sorbom, 1993a, 1993b). Direct standardization of latent vari ables using single-sample confir matory factor analysis is often accomplished by fixing the latent variab le variances to some non-zero constant, usually 1.0 (Reise, Widaman, & Pugh, 1993), but preferably population variance values. Yet once more this provides challenges when testing for measurement invariance. For instance, if latent variable variances in bot h samples are fixed at 1.0, but at least one is significantly larger or smaller than 1.0, then discrepancies ma y arise. Therefore, to use standardization of latent va riables in model identificati on, latent variable variances should be estimated. The best practice for estimating latent variables in multi-sample analysis is to equate them across groups and allow LISREL to determine the values accordingly (Jreskog & Srbom, 1996). Thr ough estimation using equivalence across groups, LISREL is given a value, or guide post from which it may estimate latent variable variances. This is useful wh en population values are not available. Steps taken for statistical identification of the model are presented in Table 20. Determining Measurement Invariance According to Cheung and Rensvold (1999), steps for determining measurement invariance include: 1. 2 1 Testing invariance begins with a comparison of sample covariance matrices (Srbom, 1976; cited in Jreskog & Srbom, 1996). If the null hypothesis is
49 not rejected, then invariance is said to exist. If the covariance matrices are equal, it follows that all of the matrices (factor loading, factor covariance, and uniqueness) must be equal across groups, and no further te sts are needed. However, this is a very rigorous test (Cheung & Rensvold), and re jection of the null usually follows, prompting additional analysis. 2. 2 1 form form Next, the fit of the theoretically derived model presented in Figure 1 is tested for invariance across groups. In this instance, factor loadings are not constrained to be equal across groups, only th e pattern of manifest (item) and latent variable loadings. If inva riance is confirmed, then the 2statistic used in this unconstrained baseline model is used in subsequent tests (2 uncon). If not, then further tests of invariance are not warranted si nce a common factor structure cannot be supported by the data. In the wake of su ch a finding, the only reasonable courses of action are single-sample confirmatory factor analysis and exploratory factor analysis. However, because the factor structures are known to differ (because of the results of step 2), no further tests of equivalence are warranted. Determining manifest loading invariance. Manifest (item) loading invariance involves both full and partial invariance te sting. Given factor pattern invariance, procedures for these and subsequent tests would include the following: 3. 2 1 Manifest (item) loadings are held invariant and the difference between the 2 uncon and the subsequent 2statistic is calculated. In the case of a significant difference, subsequent tests would not be wa rranted. However, a series of tests would be conducted to determine partial measuremen t invariance within the lambda matrix.
50 A nonsignificant difference between the unc onstrained and fully constrained model would indicate measurement invariance. Subsequent tests would include: 4. 2 1 A test of factor variance/covarian ce invariance to pr ovide the basis for testing equality of f actor intercorrelations. 5. 2 1 Assuming factor variance/covariance similarity, invariance between unique variance matrices would indicate comparab le reliability between the two samples. Partial measurement invariance. If the same pattern of loadings across groups has been established, and the equality of the set loadings across groups has been rejected, partial measurement invariance can be inves tigated. Partial measurement invariance refers to examining which item loadings in a measure are invariant across multiple samples, and which are not. Although there is no generally accepted procedure for such an analysis, suggestions exist within the literature (C heung & Rensvold, 1999; Vandenberg, 2000). For the present study, the following procedure was chosen: 1. One item at a time was held equivalent. All other items in the measure were not constrained to be equal across groups. 2. As with prior tests, differences between the 2uncon and the present 2were used to determine whether an item loading was equivalent across groups. a. For directly standardized latent variables, all three items were tested in this manner. b. For indirectly standardized latent variables using a re ference variable, each item was alternately used both as a reference variable and a constrained variable, with the third va riable set to vary freely. Results
51 determined which items were invariant, and which item was best used as the reference variable (Cheung & Rensvold, 1999). Testing Model Fit Because the 2 Likelihood Ratio Test is highly se nsitive to both sample size and departures from multivariate normality (Byrne, 1998), three additional indices were also used to analyze model fit. These were th e standardized root mean square residual (SRMR), the non-normed fit index (NNFI), and the root mean square error of approximation (RMSEA). The three were c hosen for their ability to identify models with underparameterized factor covariances an d factor loadings (e.g., when one or more parameters are fixed to zero when their popu lation values are not zero; Hu & Bentler, 1998). In addition, the SRMR was chosen for its robust results when sample sizes are small (e.g., N<250). For SRMR and RMSEA, a value less than .05 is considered a good fit. In addition for RMSEA, a value between .05 and .08 is considered fair fit, and .08 to .10 moderate to poor fit. Finally, for NNFI, a value at or above .95 is considered good fit (Hu & Bentler). Results Data Preparation Comparison sets. To test whether the ANZ and US samples had similar six-factor structures, the samples were split between, th en compared within, gender: Males from the ANZ sample were compared to males from the US sample, and females from the ANZ sample were compared to females from the US sample. As previously stated, comparing the samples in this way ruled out gender as a reason for any differences found across cultures. Tests for measurement invariance we re then duplicated ac ross both sets within
52 the ANZ sample. Specifically, to test Hypot hesis 1a, tests for full and (when necessary) partial measurement invariance were applied to the US and ANZ male samples. To answer Hypothesis 1b, the same tests were then applied to the US and ANZ female samples. In addition, to address Hypothes is 2, tests for full and partial measurement invariance were also applied to the ANZ sa mple split by gender (males vs. females). Imputation of missing values. For males in the US sample, 15 separate items were found to have missing data, spread across 13 different people. Missing data were subsequently imputed by predicting responses based on completed answers to same-scale items. First, regression weights and error te rms were calculated us ing individuals with complete data sets. Since no scale containe d more than one missing value for any given individual, each equation consisted of two pred ictors. Missing values were then replaced by finding predicted values for each missing value using the appropriate regression equation. Descriptive statistics. Correlations and descriptive statistics for items and scales are presented in Tables 5 through 18. The largest number of significant inter-item correlations was found between males in the US (132 out of 162) followed by females in the US (112 out of 162), males in ANZ (112 out of 162), and females in ANZ, respectively (98 out of 162). However, item intercorrelations, means, and standard deviations appeared roughly the same across datasets. For scale scores, the number of significant intercorrelations was largest for the US females (15 out of 15), followed by ANZ males (13 out of 15), US males (12 out of 15), and ANZ females (10 out of 15). As with the items, the size of the intercorrelati ons, means, and standard deviations for the scales appeared similar across samples.
53 Determining Measurement Invariance Comparing variance/covariance matrices. Significant 2 between the sets of males (2(171)=282.03, p<.01) and females (2(171)=427.29, p<.01), as well as mediocre-to-poor fit results (males: RM SEA=.062; SRMR=.10; NNFI=.92. Females: RMSEA=.08; SRMR=.12; NNFI= .90), indicate that the groups do not share common variance/covariance matrices (see Table 19). Nevertheless, as stat ed in the previous section, the comparison of covariance matrices is often considered an extremely rigorous test, because it is equivalent to holding factor pattern, factor loading, and uniqueness matrices constant across groups (Cheung & Rensvold, 1999). Additional analysis is therefore warranted, because less comp lete invariance is still possible. Hypotheses 1a and 1b: Factor pattern invariance. Hypotheses 1a and 1b stated that an equivalent six-fact or pattern would be found betw een both males and females in the US and ANZ samples. As an initial ste p, the first item in each factor was designated as a reference variable and its loading constrained to 1.0. According to Jreskog and Srbom (1996), a reference variable for a late nt variable is an observed variable that represents the latent variable in the sense of being a valid and reli able measure of it. There can be only one reference variable for each latent variable (p. 18). Because the scale of latent variables must be estimated from the observed variables, designating a reference variable by constrai ning its loading to a non-zero number provides the basis from which this may occur. Conventionall y, researchers have constrained reference variables to 1.0 (Reise, Widaman, & Pugh, 1993), a practice that was followed in this study. Using such a procedure, statistical identification of the model was achieved for both males and females. Overall, model fit was good for females (2(240)=464.76,
54 p<.01; RMSEA=.066; SRMR=.049; NNFI=.94), and moderate-to-poor for males (2(240)=445.30, p<.01; RMSEA=.0 79; SRMR=.070; NNFI=.90). Results are therefore similar enough between the samples for both males and females to warrant further investigation of measuremen t invariance, supporting both hypotheses 1a and 1b. As previously stated, all subse quent tests of equivalence were conducted by examining the difference between the unc onstrained factor pattern 2 (2uncon), and the more constrained models 2 (2 con). Invariance of manifest variable factor loadings. Manifest variable factor loadings were tested for invariance by equating each samples lambda matrix, thus holding the factor pattern and manifest variable loadings equal across groups. As with testing for factor pattern invariance, results indicated significant 2for both males (2(257)=475.67, p<.01) and females (2(255)=509.91, p<.01), with mediocre -to-poor model fit for males (RMSEA=.081; SRMR=.14; NNFI=.84), and goodto-fair fit for females (RMSEA=.068; SRMR=.079; NNFI=.93). In addi tion, significant changes in 2between the unconstrained and constrained model for both males (2(17)=30.37, p<.05) and females (2(15)=45.15, p<.05) indicated non-invariance of manifest variable factor loadings. The remaining fit indices also indicated a decrease in overall model fit, with increases in RMSEA (males: RMSEA= +.002; females: RMSEA= +.002) and SRMR (males: SRMR= +.07; females: SRMR= +.03), and decreases in NNFI (males: NNFI= -.06; females: NNFI= -.01), for both males and females. These results indicate that further invariance tests such as factor variance/ covariance comparisons and unique variance comparisons, or scale mean comparisons across samples, would not be meaningful.
55 Statistical identification of the model. Because full measurement invariance was not found, results were analyzed for partial invariance across groups. However, before this could be undertaken, an alternative method for estimating the scale of latent variables, and subsequently ach ieving statistical identificatio n of the model, was sought. Previous identification was achieved by fixing reference variable item loadings within each scale to 1.0. However, using this method precluded testing the e quality of reference variable loadings across samples. To free th ese parameters for subs equent testing, latent variable scaling needed to be achieved by a lternative means. One such way is to fix latent variable variances to some nonzero value. Jreskog and Srbom (1996) recommend using values that represent popul ation variances, and co mmon practice is to equate the variances to 1.0 (Reise, Widaman, & Pugh, 1993). However, because population information is not available, and actual factor variances can not be known, estimation of factor variances was considered th e best possible practice in this situation. However, to free the references variables for parameter testing, estimate the factor variances, and identify the statistical model, so me reference point is still needed for latent variable scaling. Fortunately, with multi-samp le analysis, latent-variable scaling can be achieved by equating factor vari ances across samples. This avoids the difficulty of determining actual latent vari able variances required when designating fixed values. In summary, given the lack of population inform ation on the work/family conflict scales, equating factor variances across samples for la tent scale estimation in lieu of using fixed values was considered more appropriate. For males, using the unconstrained model where only factor patterns are designated as equal resulted in model identifi cation and identical fit statistics as when
56 reference variables were used. However, co mparisons between the female sets resulted in non-positive definite covariance matrix er rors in the estimated common model, and no results were provided by LISREL. This indicates that at leas t one of the la tent variable scales could not be estimated by equating factor variance across groups. To determine which factor or factors were causing the non-pos itive definite covariance matrix error in the LISREL program, identification was reverted to using reference variables for each of the factors. Thereafter, an ite rative approach to determini ng which factor variances could not be estimated by equating factor variances was undertaken. Descriptions of the steps taken during successive iterations are pres ented in Table 20 and elucidated below. Using the original model that utilized reference variables to achieve statistical identification, changes were made one factor at a time. For each iteration, a single latent variables variance was equated across groups and the corresponding reference variable freed. Given model identification after such a modification, subsequent iterations would include the previous modificat ion, and test of the next f actor in like manner. For example, the original program was modified so that the reference variable for time-based work interference with family was freed, and th e latent variable variances in each sample equated within the LISREL program. All other factor scales continued to be estimated using reference variables. When the program was executed, statistical identification of the model was again achieved, yet now the lo adings for all three items for time-based work interference with family may be test ed for measurement equivalence. For the second iteration of this process, the results of the first iteration are kept in place, and the procedure of freeing the reference variable a nd equating the latent variable variances for the next factor is repeated, and so forth unt il all relevant factors have been tested.
57 Through this process it was found that strain -based family interference with work required the use of a reference variable to de termine latent variable scaling between the female samples, while all other factors could be scaled by equating the factor variances. Therefore, the variance for this factor was allo wed to vary freely, and a reference variable re-assigned to be used in determining latent variable scaling. The remaining five factors were scaled by equating their respective factor variances. As a result, no errors were encountered, statistical identif ication of the model was achieved, and testing for partial measurement invariance resu med for both comparison sets. Partial measurement invariance. Partial measurement invariance was determined on an item-by-item basis. In turn, each mani fest variable loading was constrained to be equal across samples, while simultaneously the loadings of all other items in the measure were allowed to vary freely. For males, the resulting change in 2 and fit statistics were significant for items three (2=5.42, p<.05; RMSEA= +.001, SRMR= +.012, NNFI= 0.00; I have to miss family activitie s due to the amount of time I must spend on work responsibilities), seven (2=6.04, p<.05; RMSEA= +.002, SRMR= +.007, NNFI= 0.00 ; when I get home from work I am often too frazzled to participate in family activities/responsib ilities), and eleven (2=5.06, p<.05; RMSEA= +.001, SRMR= +.021, NNFI= 0.00 ; because I am of ten stressed from family responsibilities, I have a hard time con centrating on my work). Although the NNIF statistics indicated no change, the results we re still poor. Combined with the significant 2, large negative changes in SRMR, and sm all negative changes in RMSEA, it was
58 determined that the model would fit the data better if these items were allowed to vary freely. As a result, the final model consisted of equating factor patterns between the samples, as well as all manifest variable loadings with the exception of items three, seven, and eleven. Overall change in 2 for males was not significantly different from the unconstrained model (2(15)=15.14, p>.05), yet fit wa s again mediocre-to-poor (RMSEA=.078; SRMR=.11; NNFI=.91). Howe ver, fit indices do show a small improvement using partial measurement inva riance when compared to full measurement invariance (RMSEA= -.001, SRMR=0.00, NNFI= +.01). Fit statistics for each of the manifest variable loading tests between the male samples are presented in Table 21, with final lambda, phi, and error matrices, along with estimated modification indices for males, in Tables 23 through 25. Modification indices provide the estimated change in 2 should the specific item invariance between samples be set to vary freely. For males, LISREL estimated that removing the equality constraints between samples for items 10, 12, and 15 would reduce the 2 by 10.42 (2.57, 4.87, and 2.98, respectively). All other modifications would result in 2 reductions of less than two. However, this reduction would not result in a nonsignificant statistic. Furthermore, the change in 2between the unconditional and conditional model was already non-significant. Therefore, removal of any equality constraints wound neither improve measurement i nvariance nor model fit. As a result, all remaining equality constraints between th e two male samples were left intact.
59 For females, the resulting change in 2 and fit statistics were significant for items one (2=5.35, p<.05; RMSEA= 0.001, SRMR= +.015, NNFI= 0.00; my work keeps me from my family activitie s more than I would like), two (2=6.84, p<.05; RMSEA= 0.001, SRMR= +.018, NNFI= -.01; the time I must devote to my job keeps me from participating equally in house hold responsibilities a nd activities), four (2=4.49, p<.05; RMSEA= 0.00, SRMR= +.010, NNFI= 0.00; the time I spend on family responsibilities often interferes with my work responsibilities), seven (2=8.30, p<.05; RMSEA= +.001, SRMR= +.0031, NNFI= -.01; when I get home from work I am often too frazzled to participate in family activities/responsibilities ), eight (2=5.77, p<.05; RMSEA= 0.001, SRMR= +.023, NNFI= -.01; I am often so emotionally drained when I get home from work that it prevents me from contributing to my family), 10 (2=3.95, p<.05; RMSEA= 0.00, SRMR= +.003, NNFI= 0.00; due to stress at home, I am often preo ccupied with family matters at work), 15 (2=5.88, p<.05; RMSEA= 0.00, SRMR= +.017, NNFI= -.01; the behaviors I perform that make me effective at work do not help me to be a better parent or spouse), and 18 (2=4.99, p<.05; RMSEA= 0.00, SRMR= +.011, NNFI= 0.00; The problem-solving behaviors that work for me at home do not seem to be as useful at my work). Although changes in SRMR were large, changes in RMSEA and NNFI were negligible. Combined with the significant 2, it was determined that the model would better fit the data if these items were allowed to vary freely. Howeve r, since eight out of 18 of the items displayed differences in factor loadings, use of the measure for comparisons between cultures is not supported.
60 In addition, it was found that designating item eleven (because I am often stressed from family responsibilities, I have a hard time concentrating on my work) as a reference variable for strain-based family interference with work resulted in better parameter invariance than using other items. This was determined using Cheung and Rensvolds (1999) triangle heuris tic. According to the authors, certain items may work better as reference variables when testing for invariance in multiple groups. They proposed a system by which items in a factor were alternately used as the reference variable, set free to vary, or constrained to equality across groups. The pattern that provides the greatest amount of equality across gr oups is then used in identification of the model. Specifically, for strain-based family interference with work, using item 12 as the reference variable resulted in a significant 2 for item 11, but not vice-versa, while item 10 had a significant 2 regardless of which was used to scale the latent variable. By adjusting for equality based on these findings, overall change in 2 was not significantly different from the unconstrained model (2(9)=13.96, p>.05), and a goodto-mediocre fit with the model was realized (RMSEA=.065; SRMR=.064; NNFI=.94). As with the males, fit indices for partial measurement invariance showed an improvement over the test for full measurement invariance (2(10)=43.03, p<.05; RMSEA= -.003, SRMR= -.056, NNFI= +.01). However, since only 10 of the 18 items could be used in the adjusted scale, the viab ility of the measure as an indicator across cultures of work/family conflict is questiona ble. Fit statistics for each of the manifest variable loading tests are presented in Table 22, with final lambda, phi, and error matrices, along with estimated modification indices for females, in Tables 26 through 28.
61 Modification indices provide the estimated change in 2 should the specific item invariance between samples be set to vary freely. For females, LISREL estimated that removing the equality constraints between samples for item 13 would reduce the 2 by 4.11. All other modifications would result in 2 reductions of less than two. However, as with the males, this reduction would not result in a non-significant statistic. Furthermore, the change in 2between the unconditional and modifi ed conditional model was already non-significant. Therefore, removal of a ny equality constraints wound neither improve measurement invariance nor model fit. As a result, all equality constraints between the two female samples remained intact. Exploratory factor analysis Exploratory factor analysis was conducted to determ ine why model fit between US and ANZ males was medioc re-to-poor, but fit between the female samples was goodto-mediocre. Initially, princi pal axis factoring with a Prom ax rotation and a forced sixfactor solution was used to determine if the sample factor patterns matched those espoused in theory. Factor patterns and lo adings are presented in Tables 29 to 30 for males and 31 to 32 for females. Results indi cated that a six-fact or solution showing simple structure and conforming to theory coul d not be extracted from data from males in the US. For males in ANZ, all but three items loaded exclusively on their a priori factors. Specifically, items 13 and 15 double-loaded on their a priori behavior-based work interference with family factor, and a f actor containing behavior-based family interference with work, while item 14 only loaded on the factor containing behaviorbased family interference with work.
62 Results from the females in the ANZ sample also indicated that items from the behavior-based work/family conflict scales lo aded on the same factor. However, unlike the males, only item 17 loaded on a separate a priori factor, and item 13 double-loaded on the factor containing items from the strain-bas ed family interference with work scale. For the US female data, a forced six-factor solution also provides evidence that the original theory may be viable. However, unl ike the ANZ males, items for the strain and time-based family interference with work scal es loaded on the same factor. Yet items 4 and 5 of the time-based family interference wi th work scale also loaded on a separate factor. All other items loaded on their a pr iori factors, with item 15 double-loading on the behavior-based family inte rference with work factor. Given the difficulties in fitting the data to a six-factor model, a forced five-factor solution was also investigated. For ANZ ma les, a clear five-factor solution emerged, with no complex loadings, and items from th e two behavior-based scales loading on a single factor. For US males, the results were not as clean. Again, the items from both behavior-based scales loaded on the same fact or, as did item nine from the strain-based work interference with family scale. Item eight loaded on the strain-based family interference with work scale, and items four and 14 doubleloaded on this same factor. Finally, a garbage factor appeared containing item se ven, and double-loadings for items 13 and 16. For the females, a five-factor solution fit the data well, albeit for different reasons. Like their male counterparts, the ANZ female s displayed a clean structure, with items from the two behavior-based scales loadi ng on the same factor. The one exception was item 13, which loaded on both the behavior-bas ed factor and the strain-based family
63 interference with work factor. For US females, time and strain-based family interference with work scale items again loaded on the same factor. In addition, items 13 and 15 double-loaded on their a priori behavior-based work interference with family factor as well as a more general behavior-based factor with items from the behavior-based family interference with work scale. When only a small number of items load on each dimension, and the communalities of the items are relatively low, results from exploratory factor and principal components analysis can differ wi dely (Floyd & Widamen, 1995). Therefore, principle components analysis with a prom ax rotation was also computed to help determine the extent to which the data sets do not fit the a priori work/family conflict theory. Component patterns and loadings are presented in Tables 33 to 34 for males and Tables 35 to 36 for females. Using a forced six-component solution with the males, both the order of component extracti on and pattern of loadings are different for each sample. In the US group, a combined behavior-based dimension that included both work interference with family (items 13, 14, & 15) and family interference with work (items 16, 17, & 18) emerged. A similar finding occurred with males in the ANZ sample, with the exception of item 13, which loaded on a separate dimension. Also in the ANZ sample, the other two behavior-based work interference with family items (items 14 & 15) double-loaded on the dimension containing it em 13. As with the exploratory factor analysis with the ANZ male group, the double-lo ading of these items presents two clear components for behavior-based work interference with family in the ANZ sample. It also provides evidence that the scales may be m easuring similar, yet distinct, factors of work/family conflict. Lastly, it is a clear departure from the component structure of
64 males in the US sample, at least in part ac counting for the level of fit between the two groups. Other differences between the US and ANZ male sample component structures are seen in the time-based family interferen ce with work dimension (items 4, 5, & 6). For the US, item 6 of this component load s on its own dimension, while for the ANZ males, all three items load on the same component. In addition to this, several items in the US sample have double-loadings on unrel ated components. For instance, items 8 (strain-based work interference with family ) and 14 (behavior-based work interference with family) load on their designated dime nsions and on the strain-based family interference with work dimension. In a ddition, item 13 loads on both its designated behavior-based work interference with fam ily dimension, and the strain-based work interference with family dimension. Finall y, using a forced five-component solution did not significantly change these findings. In summary, exploratory com ponent analysis revealed that data from males in the US sample appear well represented by a 5-component solution, and show a complex loading pattern with some items. In contrast, data from males in the ANZ sample were at least partially repres ented by the theorized 6-component solution, with some complex loadings between the highly correlated be havior-based work/family conflict scales. These dissimilarities appear large enough to af fect the level of measurement invariance between the two samples. Regardless of the dissimilarities between th e two male samples, the patterns are still similar enough to provide evidence of fit with a common model. Specifically, 16 of the 18 items loaded on their theoretical a priori components. In addition, the majority of
65 the complex items displayed lower or nega tive loadings on non-a priori dimensions, indicating that the majority of their variance was accounted for by the correct a priori components. For these reasons, the simultaneous confirmatory factor analysis provided mediocre levels of fit on at least some of the indicators. Given the small sample size of males in the US (N=83), and the previously mentioned imputation of some of the US samples data, the level of fit found between the US and ANZ group may be viewed as appropriate. For females, the differences in pattern loadings are not as complex as those with the males. To begin with, the US sample is of a considerably larger size (N=139), that allows for more diversity in answers. As a result, exploratory components analysis revealed a 6-factor structure much closer to the theoretical model than what was found in the US males. All items lo aded on their a priori dimensions, and only two had doubleloadings: item 15 loaded on both behavior-bas ed factors, and item six loaded on both the behavior based and time-based family interference with work factors. The structure of the ANZ female data wa s not as consistent with that of the original theory as the US female sample. To begin, like their male counterparts, items from both behavior-based scales loaded on the same component, with the exception of item 13, which loaded on an independent di mension. However, unlike the ANZ males, items 14 & 15 of the behavior-based work in terference with family scale did not doubleload on the same component as item 13. In ot her words, the female sample essentially contained only one component for behavior-based work interference with family. All other dimension patterns for the ANZ female sample were identical to those of the US sample. The only exception to this was item 4, which double-loaded on both time-based
66 dimensions. In short, the similarities between the fema le samples are large enough to warrant a good-to-mediocre fit in measurem ent invariance, but the dissimilarities between them on the behavior-based component prohibits a better fit between the groups. Summary and explanation of exploratory factor and components analysis. Put simply, sample characteristics may be attenua ting fit between the groups in simultaneous confirmatory factor analysis. Most obvious of these characteristics is the small number of data points in the US male sample. It is lik ely that diversity of scores in larger samples would provide a better represen tation of the population and th erefore a better fit with the theoretical model. More specifically, as sa mple size increases, not only is the population factor structure better reveal ed, but also smaller differences between the samples become significant. For all groups, high correlations between the behavior-based factors resulted in exploratory findings where a five-f actor solution fit the data ( US males)=.89; ANZ males)=.84; US females)=.79; ANZ females)=.98). This indicates a high degree of similarity between the two behavior-based latent variables, and in how these scales are being answered. In addition, for US females, hi gh correlations between the time and strainbased family interference with work factors (85) reveal why items from those two scales loaded on the same dimensions. Neve rtheless, there is enough similarity between the two sets of samples to wa rrant a study of measurement inva riance. This is evidenced by the non-significant 2 and positive changes in other fit statistics when testing for factor pattern invariance.
67 Hypothesis 2: Gender Differences in the Australian/New Zealand Sample. Hypothesis 2 stated that di fferences would exist between males and females in the Australian/New Zealand sample on Carlsons work/family conflict scale. To determine the extent to which measurement noninvariance o ccurred, steps of the prior analyses were repeated on Australian/New Zealand males and females. Comparing variance/covariance matrices. Significant 2 between males and females (2(171)=322.74, p<.01) in Australia/New Z ealand indicate that the groups may not share common variance/c ovariance matrices (see Table 37). However, good-tomediocre fit results (RMSEA=.062; SRMR= .081; NNFI=.94) indicate that measurement invariance may still exist. Since the comparison of covariances is considered an extremely rigorous test, holding pattern and factor loadings constant across groups (Cheung & Rensvold, 1999), additiona l analysis is warranted. Hypothesis 2: Factor Pattern Invariance. As with prior analyses, an initial step was to use the first item in each factor as a reference variable, constraining its loading to 1.0. Using this procedure, statistical iden tification of the mode l was achieved, with a significant 2 ((240)=497.98, p<.01), and mediocre fit statistics (RMSEA=.070; SRMR=.060; NNFI=.93). However, results ar e similar enough between the samples to warrant further investigation of measurement i nvariance. As with previous analyses, all future tests of equivalence will be conducte d by examining the difference between this unconstrained factor pattern 2 (2uncon), and the more constrained models 2 (2con). Invariance of manifest variable factor loadings. Manifest variable factor loadings were tested for invariance by equating each samples lambda matrix, thus holding the
68 factor pattern and manifest variable loadings equivalent across groups As with testing for factor pattern invariance, results indicated significant 2between males and females (2(258)=529.04, p<.01), with mediocre-to-fa ir model fit (RMSEA=.071; SRMR=.083; NNFI=.93). In addition, significant changes in 2between the unconstrained and constrained model (2(18)=31.06, p<.05) indicates non-inva riance of manifest variable factor loadings, supporting Hypothesis 2. Both RMSEA and SRMR also indicate a decrease in overall model fit (RMSEA= +.001; SRMR= +.023), with no change in NNFI. These results indicate that furt her invariance tests such as factor variance/covariance comparisons and unique variance comparisons, or scale mean comparisons across samples, would not be meaningful without further modification. Statistical identification of the model. Since full measuremen t invariance was not found, results were analyzed for partial inva riance across groups. As with the prior analyses, estimating the scale of latent variab les by equating factor va riances was used in favor of reference variables. Subsequentl y, each respective refere nce variable for the factors was freed for parameter testing. Th e new pattern of constraints resulted in identification of the model (Table 37) and testing for partial measurement invariance resumed. Partial measurement invariance. Partial measurement invariance was determined on an item-by-item basis. For each manifest variable, loadings were constrained to be equal across samples while all other items in the measure were allowed to vary freely. Resulting change in 2 and fit statistics were significant for items four (2=4.28, p<.05; RMSEA= +.001, SRMR= +.008, NNFI= 0.00; the time I spend on family
69 responsibilities often interf eres with my work responsibilities), and eight (2=4.69, p<.05; RMSEA= +.001, SRMR= +.009, NNFI= 0.00 ; I am often so emotionally drained when I get home from work that it prev ents me from contributing to my family). Although the NNFI statistics indicated no change, the combined significant 2, large changes in SRMR, and small changes in RMSEA, it was determined that the model would fit the data better if these item s were allowed to vary freely. As a result, the final model consisted of equating factor patterns between the samples, as well as all manifest variable lo adings with the excep tion of items four and eight. Using these changes, overall 2 between ANZ males and females was not significantly different from the unconstrained model (2(16)=22.23, p>.05), yet fit was again mediocre (RMSEA=.070; SRMR=.075; NNFI= .93). Fit statistics for each of the manifest variable loading tests between the male and female ANZ samples are presented in Table 38, with final lambda, phi, error matr ices, and estimated modi fication indices, in Tables 39 through 41. Modification indices provide the estimated change in 2 should the specific item invariance between samples be set to vary freely. It was estimated that removing the equality constraints between males and fe males for items 7,10, 11, and 12 would reduce the 2 by 18.23 (2.49, 10.03, 3.14, and 2.57, respectively) All other modifications would result in 2 reductions of less than two. However, this reduction would not result in a non-significant 2 statistic. Furthermore, the change in 2between the unconditional and conditional model was already non-significant. Therefore, removal of any equality constraints wound neither improve measurement i nvariance nor model fit. As a result, all
70 remaining equality constraints between the Australian/New Zealand male and female samples remained intact. Research Question 1: Differences between Males and Females in the Australian/New Zealand Sample Since non-invariance exists in items f our and eight between Australian/New Zealand males and females, results of mean comparisons for the time-based family interference with work scale, and the strain-based work inte rference with family scale, are subject to debate. However, the other f our scales indicate m easurement invariance and therefore can be compared across groups without question. Resu lts indicate that significant gender differences only exist for ti me-based work interference with family and family interference with work, where male s reported higher levels of conflict than females. No differences across gender for st rain or behavior-based family interference with work or work interference with family were found (Table 42). This is markedly different than the Carlson et al. (2000) findings, in which fe males indicated significantly higher mean scores on all three forms of fam ily interference with work (time, strain, behavior), as well as strain-based work in terference with family. In short, gender differences did exist between Australian/ New Zealand males and females, supporting hypothesis 2. Yet these differe nces are not the same as t hose found in the United States by Carlson et al., and are dissimilar e nough to warrant further investigation. Discussion The purpose of this study was to examine the level of measurement invariance in a scale of work/family conflict between samples from the United States and Australia/New Zealand. Samples from both cultures were also split into males and
71 females to rule out gender as a possible altern ative explanation to findings. As a result, a total of four datasets were analyzed: US males vs. ANZ ma les, and US females vs. ANZ females. In addition, gender differences were examined within the Australia/New Zealand sample to determine if males and females reported similar patterns of work/family conflict compared to their US counterparts. Measurement invariance was conducted on both sets of analyses through th e use of simultaneous confirmatory factor analysis using the LISREL 8.3 statistical p ackage. In addition, exploratory factor analysis was conducted on the four data sets to determine if discre pancies were the result of underlying differences in factor patterns th at are not revealed in confirmatory factor analysis. Finally, mean differences between males and females within the ANZ sample were compared on scales where measurement invariance was determined. Testing for Measurement Invariance Hypotheses 1a and 1b. Hypotheses 1a and 1b stated th at an equivalent six-factor structure of Carlson et al.s (2000) work/f amily conflict scale w ould be found between males in the English-speaking Australian/New Zealand sample and males in the Englishspeaking sample from the United States, as well as between females in the US sample and females in the ANZ sample. One pro cedure for testing these hypotheses was to compare sample covariance matrices. When th e matrices prove to be similar due to a good fit with the common model, then measuremen t invariance is said to exist and further tests are unwarranted. The findi ngs indicated that a poor fit existed for both the male and female sets. Nevertheless, this test is cons idered very rigorous, and a poor fit with the common model does not necessari ly mean that full measurem ent non-invariance exits.
72 Therefore, additional testing was done to determine if the findings of the initial invariance test were due to minor or major differences between the samples. The next step was to constrain only the pa ttern of factor load ings to equality across groups, not the equality of each of the loadings. When constraining the pattern of factor loadings to be equa l across groups, the common m odel fit between male groups was mediocre-to-poor, and the common model fit between female groups was good. This test was essential to determine if the samples could be compared further. Without at least a moderate fit with the hypot hesized factor model, there would be no justification to continue comparisons using multiple samples. This is because the factor structure would be so different across groups that the numbers resulting fr om the scales would not be comparable. Determining what is enough fit between each sample and the common model is not cut-and-dry. For instance, using only the 2 Likelihood Ratio Test nearly always results in a significant statistic, indicating that the model does not fit the data. Yet it is widely accepted that the 2 test is not robust against devi ations in multivariate normality and deviations due to small sample sizes (n <250; Hu & Bentler, 1998). The three other fit indices used in this study were picked to complement the 2 for their abilities to identify models with underparameterized covariances and factor loadings (RMSEA, NNFI), as well as their robust results when sample sizes are small (SRMR). However these are imperfect in that they are descri ptive statis tics, providing good information only when a model either fits well or poorly, and only when they are similar in their assessments. Such was the case with the fema le data, where all three descriptive indices indicated good model fit, providi ng support for hypothesis 1b.
73 Decidedly different results were found in the male set, where the RMSEA and SRMR indicated mediocre fit, a nd the NNFI indicated poor fit (the 2, as always, was significant). This provided at least part ial support for hypothesis 1a, and is a good example of the gray area in interpreting mode l fit, which occurs when some fit indices indicate a good fit and others indicate a poor fit. Determ ining whether the fit was good enough to continue comparisons between the male samples was based entirely on the relative distance between the findings and specified cutoff scores, as well as common practices in the literature. Using these gui delines, it was determined that the common models fit with both male samples was good enough to warrant further investigation, even though the hypothesis was only considered to be partially supported. Because this is the model to which al l additional (constrained) models were compared, it is referred to as the unconstrained model. Invariance of manifest variable factor loadings. Although hypotheses 1a and 1b were essentially supported, the findings do not complete the study s primary purpose of determining measurement invariance in the work/family conflict scales across both sets of samples. To do so requires additional testing, beginning with the manifest variable factor loadings. While continuing to hold the pattern of factor load ings invari ant across groups, the values of item loadings were also held constant across groups and common model fit assessed. However, unlike the factor pattern test, results indicated that fit with the descriptive indices decreased when manife st variable loadings were held constant across groups, both for the males and fe males. In addition, the change in 2 between the unconstrained model and the pattern-and-lo ading constrained m odel was significant,
74 providing further evidence that measurement noninvariance exists acr oss culture for both males and females. The next step was to determine to what extent measurement non-invariance existed between the samples in each set, also known as partial measurement invariance. Additional analyses revealed th at specific manifest variable factor loadings between the male samples (items 3, 7, & 11) were cau sing the overall compar ison of item factor loadings to fit the data poorly. By allowing these loadings to vary freely for each sample, instead of constraining them to be equal, de scriptive fit statistics improved considerably, and the change in 2 was no longer significant. The s ituation was considerably more complicated between the female samples, where eight items (numbers 1, 2, 4, 7, 8, 10, 15, 18) displayed non-invariant factor loadi ngs when constrained to be equal across samples. Again, when these are set to va ry freely, descriptiv e fit indices improved considerably, and the change in 2 was no longer significant. Nevertheless, the high level of non-invariance between the males and fema le samples would make mean comparisons across cultures difficult to interpret, and ther efore such comparisons were not computed. To determine why measurement non-inva riance occurred and why the male samples had a poorer fit with the common model than the fe male samples, exploratory factor and component analyses were computed on all four samples. Results revealed differences in underlying patterns within each sample. For instance, data from US males contained a considerable number of doubl e-loadings and did not conform to the theoretical model (and the one proposed by Carlson et al., 20 00), yet data from the ANZ males more readily adhered to a six-factor solution. However, the complex item loadings in the male ANZ sample make a five-factor so lution plausible as well. This may explain
75 the mixed findings of the fit indices, where some indicate mediocre fit, but others point to a poor fit between the two data sets. In addition, partial m easurement invariance revealed only three items that could not be equated acro ss groups. These similarities indicate that the dissimilarities between the samples may be due to sample-specific issues as opposed to problems within the theoretical factor struct ure of the measure. For instance, the size of the US male sample (n=83) may not be large enough to provide an adequate representation of how males in the US react to this measure. The cleaner solution for the US males found in the component analysis may indicate that a larger sample representing a diverse set of job-types woul d yield a factor pattern with greater similarity to the one displayed by males in the ANZ sample. In contrast, US females appear ed closer to the a priori 6-factor solution than their male counterparts, yet items from time and st rain-based family interference with work double-loaded on a single fact or. Meanwhile, the ANZ fema les data appeared best represented by a 5-factor solution. Differences between the two female samples results were also found with the behavior-based scal e, where the work and family-based scale items loaded on separate components for US females, but only on one component for ANZ females. However, with the ANZ female sample, a single item from the behaviorbased work interference with family scale loaded on its own component, indicating at least partial similarity in the behavior-based scales factor structur e between the two data sets. Nevertheless, it does appear that con ceptually there are 6 f actors, yet only five appear to be defensible empirically. In addition, unlike the male samples, the females largely displayed clean item loadings in each analysis. This, coupled with the doubleloadings of time and strain-based family interference with work in the US sample,
76 provides evidence for the good-to-mediocre mo del fit found between the two samples in the simultaneous confirmatory factor analyses In addition, the differences in factor structures also provide evidence for why ei ght out of the 18 items were found to have non-invariant loadings across the samples. However, both samples were large enough (US=139; ANZ=249) to provide at least adequa te representation of the population scores. This leads to the question of whether some other sample-related issue is causing such a marked departure from the general work/family conflict theory, as discussed below. Overall, these issues may also be maski ng error in the scales ability to measure the theoretical model. This question arises when looking at the factor analysis results from the two Australia/New Zealand samples, where size was not an issue. Both males and females had cleaner five-fact or structures than six-fact or structures. Although these structures have greater similarities than t hose found between males and females in the US sample, they are nevertheless a departure from the theoretical model. More importantly, problems arise across all the samples with the measurement of behavior-based work/family conflict. Specifically, although be havior-based conflict appears to exist, whether it can be parsed into work and fam ily orientations are unclear. For instance, certain job-types such as litigator, poli ce officer, and military personnel may require changes in behavior that are drastically different between th e work and family domains. If individuals have trouble adjusting when moving between domains, then behaviorbased conflict would be more prevalent and more readily measured by the Carlson et al. scale. Nevertheless, it is interesting that both the US and ANZ samples show conflicting results when I attempted to fit a bi-directio nal solution to this form of work/family conflict. This may indicat e that the measure is not thoroughly sampling the six
77 work/family conflict components, and therefore may need to be revised by increasing the specificity of the item wording. An alternative sample-related explanati on may be in the composition of the data sets. Whereas the US female sample is prim arily made up of indivi duals in their late 30s, the Australian/New Zealand female sample consists of individuals in their early 30s. It may be that younger women have le ss experience dealing with work/family role overlap, be more likely to have younger children at home, and are more likely to be in an earlier career-building phase than older women. These issues may moderate the extent to which younger women are able to leave work-r elated behaviors at the office. In contrast, older women will be more likely to have older, more self-sufficient children, not to mention be more experienced in dealing wi th work/family role overlap and be further along/secure in their career development. These differences may be important in a womans ability to separate work-related vers us family-related behavior. In short, older women may be better equipped both experien tially and environmentally to separate work-related role behaviors from family-relate d role behaviors, hence the differences in the exploratory component analyses result s from those found in exploratory factor analysis. Gender Differences in the Au stralian/New Zealand Sample Hypothesis 2 and research question 1. Hypothesis 2 postulated that gender differences in work/family conflict would ex ist in the Australian/New Zealand sample, and research question 1 was an exploratory analysis into the nature of those differences. However, before mean comparisons could commence, measurement invariance between males and females in the ANZ sample needed to be confirmed. As with hypotheses 1a
78 and 1b, a covariance matrix comparison resulted in poor fit statistic s and a significant 2. Additional testing revealed a good fit with the common model when the factor pattern was held invariant. However, a poor fit with the model occurred when the factor loadings were held invariant as well. Pa rtial measurement invariance testing was then conducted, revealing significant changes in 2 and poorer model fit wh en the loadings for items four and eight were equated across ge nders. Subsequently, improved model fit and a non-significant change in 2 from the unconstrained model resulted when the loadings for these items were allowed to vary freely. The high degree of measurement invariance between males and females in Australia/New Zealand (e.g., only two items ha d non-invariant factor loadings) allowed for mean comparisons between the two groups Results indicated that differences do exist between males and females, providi ng support for hypothesis 2. However, these differences are not the same as the ones reported by Carlson et al. (2000) in their comparisons between males and females in a US sample. Specifically, significant gender differences only existed for time-based work interference with family and family interference with work, where males reported high er levels of conflict than females. No differences across gender for st rain or behavior-based family interference with work or work interference with family were found. This is markedly different than the Carlson et al. (2000) findings, in which females indica ted significantly higher mean scores on all three forms of family interference with work (time, strain, behavio r), as well as strainbased work interference with family. The qualitative conclusion is that males and females in Australia/New Zealand view work /family conflict differently than their counterparts in the United States.
79 The reasons for this may again be root ed in the age differential between the samples. Yet the possibility that cultural di fferences exist cannot be ignored. It may be that men share a greater level of household re sponsibility in Austra lia/New Zealand than in the U.S. In addition, Australian law a llows for up to one year of unpaid maternity leave, and may reflect a greater tolerance fo r pregnancy in working women than in the US. In short, men and women in Australia /New Zealand appear to be experiencing similar levels of work/family conflict across the majority of the scales in Carlson et al.s (2000) measure. Further study is required to determine the exact nature and origin of these differences. However, differences may also be due to biases within the samples, which may cause errors in recovering the factor structure from the data. Issues in Interpreting Study Results Questions about the interpretation of study re sults arise from thre e distinct issues. First, the number of indivi duals in the US dataset may be too small to represent population parameters accurately, particularly in the case of the US males. As previously stated, factor analysis of samples these size s are more easily affected by sample-specific characteristics and biases. Second, the low number of items per dimension increases the likelihood that errors will resu lt when attempting to recover the theoretical model factor structure. For instance, Stephens (1996) i ndicates that any factor defined by only a few items is close to being variable specific. That is, it may be better to interpret it as individual variables than items grouped into a single dimens ion. Lastly, the possibility that cultural differences exist cannot be ru led out, and is conf ounded with the issues surrounding sample size and measure characteristics.
80 The twin issues of sample size and m easure characteristics may be causing unstable factor pattern matrices and decreasing the likelihood of factor structure recovery, respectively. Of the four data sets, only US females and ANZ males produced reasonable six-factor solutions. Yet even in these samples one of the factors contained items with double-loadings, the magnitudes of which were smaller on their a priori factor than on their non-theoretical dimension. In other word s, none of the four samples in this study showed the clean a priori structure on which the Carlson et al. (2000) scale was developed. As previously stated the problems with f actor structure recovery may also be due to cultural differences. For instance, Aust ralian federal government policy provides some benefits that are not availabl e in the US, such as a one-time maternity allowance. In addition, child care assistance and fee subs idies are available through the federal government, even though the benefit has been reduced over the past decade. Finally, all public sector employees enjoy varying degrees of paid maternity leave as mandated by each state, and extra tax rebates are availa ble to single-income families (Buchanan & Thornthwaite, 2001). Although individually thes e benefits do not necessarily apply to this studys sample, collectivel y they indicate a potentially different national attitude towards childcare and working women. This attitude may be behind the differential responses to Carlson et al.s ( 2000) work/family conflict scale. By increasing the number of items per s cale beyond the accepted minimum, better sampling of the construct occurs and a greater amount of the variance is accounted for. Only additional research with a variety of US samples will determine whether the measure corresponds to its hypothesized factor structure in the US.
81 Summary of Results It was thought that the factor structur e of the Carlson et al. (2000) work/family conflict scale would be similar across U.S. and ANZ samples. Clearly, there was some similarity in results across cultures. Ho wever, the support was only marginal between males in the US and ANZ samples due to a medi ocre-to-poor fit with th e a priori 6-factor model. Further tests for measurement invari ance revealed that neither the comparison between the two male samples nor the co mparison between the two female samples resulted in full invariance. Specifically, bot h sets of comparisons showed poor fit when manifest variable loadings were held to equa lity across groups. S ubsequent investigation revealed that the majority of item factor load ings in both sets could be held to equality, yet some could not. The non-invariant item loadings were thereafter allowed to load freely across the sets of comp arisons, and fit with the common model in both the male and female comparisons were significantly improved. The resulting models for both groups are indicative of part ial measurement invariance. Exploratory factor analysis revealed that the difficulties encountered in determining measurement invariance might be due to problems with the underlying factor structure of the scale itself. Specificall y, only males in Australia/New Zealand and females in the US came close to matching the underlying a priori 6-fa ctor structure that Carlson et al.s (2000) scale is based upon. Instead, all of the samples more closely matched a 5-factor structure. Equally importa nt was the fact that sample characteristics may have played a part in exploratory factor analysis results, as well as subsequent invariance testing. To that e nd, the male US sample may have been underpowered due to its small size, and may not have adequately re presented the diversity of scores for males
82 in the United States. In addition, aver age age differences may have introduced moderators such as age of children at home, placement in career development cycle, and general experience in deali ng with work/family role overlap. In short, sample characteristics may have played a part in determining the level of measurement invariance experienced between the two groups in each comparison set. A final alternative explanation for the level of measuremen t invariance found may reside in the differences in culture between Australian/New Zealand and the United States. It is possible that men take a larger role in family responsibilities in Australia/New Zealand than men in the Un ited States. Also, level of acceptance of pregnancy leave and its relative impact on a womans career may be less favorable in the United States than in Australia/New Zeal and. While these questions are beyond the scope of this study, they nevertheless are plausi ble explanations as to why differences in measurement invariance were found. Finally, hypothesis 2 was supported by th e significant mean differences found across work/family conflict scales between men and women in Australia/New Zealand. However, the nature of those differences is different from the gender differences reported in the United States. Specifically, Carlson et al. (2000) reported th at females experienced higher levels on all scales associated with family interference with work, as well as strain-based work interference with famil y. In contrast, this study found that males experienced higher levels of work/family c onflict, specifically within the time-based work interference with family and family inte rference with work scal es. All other scale mean comparisons were not significant. These discrepancies in gender comparisons
83 again point to the possibility that cultural differences between the United States and Australia/New Zealand exist in how the work and family role overlap is perceived. Impact on Work/Family Conflict and Future Implications As stated in the introducti on, the field of work/family conflict has been hampered by a lack of viable measurement tools. Th e Carlson et al. (2000) scale was the first instrument that covered both th e directionality and dimensiona lity of work/family conflict as theorized by Greenhaus and Beutel (1985) The current study sought to expand the construct validity of this s cale by comparing the factor structure across two Englishspeaking cultures. However, results are mixed. The findings of this study indicate that further testing of the scale may be warrante d. Specifically, larger samples than those employed here would help to rule out sample -specific biases as the cause for any crosscultural differences uncovered. The issue can be readily addressed through the collection of additional data and may aid in the recovery of the theoretical models factor structure. However, problems in the recovery of the theo retical models factor structure may also be the result of too few items associated with each dimension. Since a factor composed of only a few items can be considered variab le specific (Stephens, 1996), any additional items would add considerably to the amount of variance accounted for, as well as expand the ability of each scale to adequately sample the construct. Therefore, once samplespecific biases have been ruled out as the cau se of factor structure difficulties, item generation and testing could then commence. Should the addition of variables fail to aid in factor pattern recovery the US, the theoretical model itself may also need to be questioned. However, should recovery succeed in the US and fail cross-culturally, the
84 results could be due to cultural differences in how work/family conflict is perceived. At that point, construction of a culture-specific measure would be called for. Current results also suggest that problems may exist in how individuals view and express behavior-based work/family conflic t. Individuals may not have the selfawareness to determine whether their actions at work are any different than their actions at home. The manager who deals strictly bu t fairly with her subordinates may not fully recognize that this same beha vior is seen as lacking compassion and empathy when applied to her children. Similarly, the succe ssful salesman may find himself working to win arguments with his family when compromi se would be the better tact. Simply put, the dynamics of family relations may require a modification of behaviors that apply well in the workplace. Awareness that one is not modifying these behaviors once one returns home may not be a generally understood phenome non. This also works in reverse, where applying behaviors that work with ones family may be neither appreciated nor appropriate in the workplace. Alternativel y, specific occupations (such as being a litigator or police officer) requi re behavior that it is marked ly different from what is called for in the home environment. Future research may try increasing the participants awareness of crossovers between work a nd family oriented behavior, as well as investigate the moderating properties of specifi c occupations on behavior-based conflict. Additional testing for behavioral awareness betw een roles may reveal that the structure of work/family conflict changes with experience and environment (level of advancement in career, age of children at home, etc). Othe rwise the construct itself may need to be modified, and subsequent measures modified or developed anew, before an adequate study of how these roles coll ide can be undertaken.
85 Conclusions Work/family research is a diverse field that includes such topics as work/family conflict and work/life balance. The ever-increasing incursion of work-related activities on what traditionally was time for the family, the need for dual-earner couples to support a household, and the increasing number of task s required of a single employee, all point to the growing need to balance work and fam ily activities. This ba lance is often difficult to achieve, resulting in conflicts between thes e two roles to an extent that has not been seen (in the US) since the height of Wo rld War II (Goodwin, 1994). The importance of discerning consequences of these difficulties on worker output, organizational adherence, organizational commitment, and employee heal th and welfare cannot be overly stressed. However, the best possible intentions and e xperiments in determining the root of these issues are only as good as the measuremen t tools on which they depend. The present study is an attempt to further the developm ent of Carlson et al s (2000) work/family conflict measure. Results indicate that addi tional testing of both th e theoretical structure of work/family conflict, as we ll as the measure used in this study, are necessary before firm conclusions can be drawn from cro ss-cultural comparisons. Without valid and reliable measurement there can be no valid and reliable answers. Developing this and other tools in conjunction with a more compre hensive theory of work and family will lay the foundation for future investigation, and hope fully help to clear the way for a better work/family environment.
86 References Adams, G.A. and Jex, S.M. (1999). Relatio nships between time management, control, work-family conflict, and strain. Journal of Occupational Health Psychology, 4(1), p. 72-77. Allen, T.D. (2001). Family-supportive work en vironments: The role of organizational perceptions. Journal of Vocational Behavior, 58, p. 414-435. Allen, T.D., Herst, D.E.H., Bruck, C.S., & Sutton, M. (2000). Consequences associated with work to family conflict: A review and agenda for future research. Journal of Occupational Health and Psychology, vol. 5, p. 278-308. Aryee, S. (1992). Antecedents and outcom es of work-family conflict among married professional women: Evid ence from Singapore. Human Relations, vol. 45, p. 813-837. Aryee, S., Fields, D., & Luk, V. (1999). A cr oss-cultural test of a model of the workfamily interface. Journal of Management, 25(4), p. 491-511. Aryee, S., Luk, V., Leung, A., & Lo, S. (1999). Role stressors, interrole conflict, and well-being: The moderating influence of spousal support and coping behaviors among employed parents in Hong Kong. Journal of Vocational Behavior, 54, p. 259-278. Australian Bureau of Statistics (1999). Retrieved November 10th, 2003 from http://www.abs.gov.au/. Bacharach, S.B., Bamberger, P., & Conol y, S. (1991). Work-home conflict among nurses and engineers: Mediating the impact of role stress on burnout and satisfaction at work. Journal of Organizatio nal Behavior, 112(1), p. 39-53.
87 Beatty, C.A. (1996). The stress of managerial and professional wome n: Is the price too high? Journal of Organizational Behavior, 17(3), p. 233-251. Bedeian, A.G., Burke, B.G., & Moffett, R.G. (1988). Outcomes of work-family conflict among married male and female professionals. Journal of Management, 14(3), p. 475-491. Beutell, N.J. and Greenhaus, J.H. (1982). Interrole conflict am ong married women: The influence of husband and wife characteri stics on conflict and coping behavior. Journal of Vocational Behavior, 21(1), p. 99-110. Bohen, H.H. and Viveros-Long, A. (1981). Balancing Jobs and Family Life: Do Flexible Schedules Help? Philadelphia; Temple University Press. Borovsky, D.M., & Stepanski, K. (1999). Me ta-analyzing to models of work-family conflict. Poster presented at the 14th annual conference of the Society for Industrial and Organizationa l Psychology, Atlanta, Ga. Buchanan, J. and Thornthwaite, L. (2001). Paid work and parent ing: Charting a new course for Australian families. Report prepared for the Chifley Research Foundation: Sydney, Australia. Burke, R. J. (1988). Some antecedents a nd consequences of work-family conflict. Journal of Social Behavior and Personality, 3, p. 287-302. Burke, R.J. and Greenglass, E.R. (2001). Ho spital restructuring stressors, work and family concerns and psychologica l well-being among nursing staff. Community, Work & Family, 4(1), p. 49-62.
88 Burke, R.J., Weir, T., and DuWors, R.E., Jr. (1979). Type A behavior of administrators and wives reports of marita l satisfaction and well-being. Journal of Applied Psychology, 64(1), p. 57-65. Burley, K. A. (1989). Work-family conflict and marital adjustme nt in dual career couples: A comparison of three time m odels. Dissertation, Claremont Graduate School, CA. Burley, K.A. (1995). Family variables as mediators of the relationship between workfamily conflict and marital adjustment among duel-career men and women. Journal of Social Psychology, 135(4), p. 483-497. Byrne, Barbara M. (1998). A Primer for LISREL: Basic Applications and Programming for Confirmatory Factor Analytic Models. New York: Springer-Verlag. Carlson, D.S. (1999). Personality and role vari ables as predictors of three forms of workfamily conflict. Journal of Vocational Behavior, 55, p. 236-253. Carlson, D.S., Kacmar, K.M., & Williams, L.J. (2000). Construction and Initial Validation of a Mulitdimensional Measure of Work-Family Conflict. Journal of Vocational Behavior, 56(2), p. 249-276. Carlson, D.S. & Perrewe, P.L. (1999). The role of social support in the stressor-strain relationship: An examination of work-family conflict. Journal of Management, 25(4), p. 515-540. Cheung, G.W. and Rensvold, R.B. (1999). Te sting factorial inva riance across groups: A reconceptualization and proposed new methods. Journal of Management, 25(1), p. 1-27.
89 Clark, Sue Cambell (2001). Work cu ltures and work/family balance. Journal of Vocational Behavior, 58, p. 348-365. Clark, Sue Cambell (2001b). Work/family borde r balance: A new theory of work/family balance. Human Relations, 53(6), p. 747-770. Cooke, R. A. & Rousseau, D. M. (1984). Stress and strain from family roles and workrole expectations. Journal of Applied Psychology, 69, p. 252-260. DeBord, K., Fanning Canu, R., & Kerpelman, J. (2000). Understanding a work-family fit for single parents moving from welfare to work. Social Work, 45(4), p. 313-324. Doumas, D.M., Margolin, G., & John, R.S. (2003). The relationship between daily marital interaction, work, and health-promo ting behaviors in dua l-earner couples. Journal of Family Issues, 24(1), p. 3-20. Duxburry, L. and Higgins, C. (1991). Gender differences in work-family conflict. Journal of Applied Psychology, 76(1), p. 60-74. Duxburry, L., Higgins, C., & Lee, C. (1994). Work-family conflict: A comparison by gender, family type, and perceived control. Journal of Family Issues, 15(3), p. 449-466. Duxbury, L.E., Higgins, C.A., and Mills, S. (1992). After-hours telecommuting and work-family conflict: A comparative analysis. Information Systems Research, 3, p. 173-190. Duxbury, L.E., Higgins, C.A., and Thomas, D.R. (1996). Work and family environments and the adoption of computer-supported supplemental work-at-home. Journal of Vocational Behavior, 49, p. 1-23.
90 Eagle, B.W., Miles, E.W., and Icenogle, M.L. (1997). Interrol e conflicts and the permeability of work-family domains: Are there gender differences? Journal of Vocational Behavior, 50, p. 168-184. Fallon, B. J. (1997). The balance between pa id work and home res ponsibilities: Personal problem or corporate concern? Australian Psychologist, 32(1), p. 1-9. Floyd, F.J. and Widaman, K.F. (1995). Fact or analysis in the development and refinement of clinical assessment instruments. Psychological Assessment, 7(3), p. 286-299. Frone, M.R. (2000). Work-family conflict and employee psychiatric disorders: The national comorbidity study. Journal of Applied Psychology, 85(6), p. 888-895. Frone, M.R., Russell, M., & Cooper, M.L. ( 1992). Antecedents and outcomes of workfamily conflict: Testing a model of the work-family interface. Journal of Applied Psychology, 77, p. 65-78. Frone, M.R., Russell, M., & Cooper, M.L. (1993) Relationship of work-family conflict, gender, and alcohol expectanci es to alcohol use/abuse. Journal of Organizational Behavior, 14(6), p. 545-558. Frone, M.R., Russell, M., & Barnes, G.M. ( 1996). Work-family conflict, gender, and health-related outcomes: A study of employe d parents in two community samples. Journal of Occupational Health Psycology, 1(1), p. 57-69. Frone, M.R., Russell, M., & Cooper, M.L. ( 1994). Relationship between job and family satisfaction: Causal or noncausal covariation? Journal of Management, 20(3), p. 565-579.
91 Frone, M.R., Yardley, J.K., & Markel, K.S. (1997). Developing and testing an integrative model of the work-family interface. Journal of Vocational Behavior, 50, 145-167. Fu, C.K. and Shaffer, M.A. (2001). The tug of work and family: Direct and indirect domain-specific determinants of work-family conflict. Personnel Review, 30(5), p. 502-522. Goodwin, Doris Kearns (1994). No Ordinary Time: Franklin and Eleanor Roosevelt : The Home Front in World War II. New York: Simon & Schuster. Greenhaus, J.H. and Beutell, N.J. (1985). S ources of conflict between work and family roles. Academy of Manageme nt Review, 10(1), p. 76-88. Greenglass, E. R., Pantony, K. L., & Burke, R. J. (1988). A gender-role perspective on role ocnflict, work stre ss and social support. Journal of Social Behavior and Personality, 3, p. 317-328. Gutek, B. A., Searle, S., & Klep a, L. (1991). Rational versus gender role explanations for work-family conflict. Journal of Applied Psychology, 76(4), p. 560-568. Higgins, C. A. & Duxbury, L. E., & Irving, R. H. (1992). Work-family conflict in the dual-career family. Organizational Behavior and Human decision Processes, 51, p. 51-75. Hill, E.J., Miller, B.C., Weiner, S.P., & Colih an, J. (1998). Influences of the virtual office on aspects of work and work/life balance. Personnel Psychology, 53(3), p. 667-683. Hofstede, Geert (1997). Cultures and organizations: Software of the mind. New York, NY: McGraw-Hill, Inc.
92 Holahan, C.K. and Gilbert, L.A. (1979). In terrole confict for working women: Careers versus jobs. Journal of Applied Psychology, 64(1), p. 86-90. Howard, A. E. (Eds.). (1995). The Changing Nature of Work. San Francisco, CA, USA: Jossey-Bass Inc, Publishers. Hu, L., and Bentler, P.M. (1998). Fit in dices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological Methods, 3(4), p. 424-453. Hulin, C.L. (1987). A psychometric theory of evaluations of item and scale translations: Fidelity across languages. Journal of Cross-Cultural Psychology, 18(2), p. 115142. Jreskog, K. and Srbom, D. (1996). Lisrel 8: Users Reference Guide. Scientific Software International, Chicago, Il. Kahn, R.L., Wolfe, D.M., Quinn, R., Snoe k, J.D., & Rosenthal, R.A. (1964). Organizational Stress. New York: Wiley. Kinnunen, U. and Mauno, S. (1998). Antecedents and consequences of work-family conflict among employed women and men in Finland. Human Relations, 51(2), p. 157-177. Kirchmeyer, K. and Cohen, A. (1999). Differe nt strategies for managing the work/nonwork interface: A test of unique pathways to work outcomes. Work & Stress, 15(1), p. 59-73. Kopelman, R. E., Greenhaus, J. H., & Connolly T. F. (1983). A model of work, family, and interrole conflict: A construct validation study. Organizational Behavior and Human Performance, 32, p. 198-215.
93 Kossek, E. E. & Ozeski, C. (1998). Work-fam ily conflict, policies, and the job-life satisfaction relationship: A review and directions for organizational behaviorhuman resources research. Journal of Applied Psychology, 83(2), p. 139-149. Matsui, T., Ohsawa, T., & Onglatco, M. ( 1995). Work-family c onflict and the stress buffering effects of husband support a nd coping behavior among Japanese married working women. Journal of Vocational Behavior, 48, p. 179-192. Netemeyer, R.G., Boles, J.S., & McMurrian, R. (1996). Development and validation of work-family conflicts and work-family conflict scales. Journal of Applied Psychology, 81, p. 400-410. ODriscoll, M.P., Ilgen, D.R., & Hildreth, K. (1992). Time devoted to job and off-job activities, interrole conflict, and affective experiences. Journal of Applied Psychology, 77(3), p. 272-279. Ouweneel, P. & Veenhoven, R. (1991). Cross-na tional differences in happiness: Cultural bias or societal quality? In N. Bleichrodt & P.J.D. Drenth (Eds.), Contemporary issues in cross-cultural psychology (pp. 168-184). Lisse: Swets & Zeitlinger Publishers. Parasuraman, S. and Greenhaus, J.H. (2002). Toward reducing some critical gaps in work-family research. Human Resource Management Review, 12(3), p. 299-312. Parasuraman, S., Greenhaus, J. H., & Granrose C. S. (1992). Role stressors, social support, and well-being among two-career couples. Journal of Organizational Behavior, 13(4), p. 339-356. Pleck, J.H. (1978). Work-family conflict: A national assessment. Presented at the Annual Meeting of the Soci ety for the Study of Social Problems, Boston, MA.
94 Pleck, J.H., Stains, G.L., & Lang, L. (1980). Conflicts between work and family life. Monthly Labor Review, 103, p. 29-32. Quinn, R.P., and Staines, G.L. (1978). The 1977 Quality of Employment Survey Ann Arbor, Mich.: Survey Research Center. Reise, S.P., Widaman, K.F., & Pugh, R.H. (1993) Confirmatory factor analysis and item response theory: Two approaches for exploring measurement invariance. Psychological Bulletin, 114(3), p. 552-566. Rice, R.W., Frone, M.R., & McFarlin, D.B. (1992). Work-nonwork conflict and the perceived quality of life. Journal of Organizational Behavior, 13(2), p. 155-168. Sagie, G. & Schwartz, S.H. (1994). National di fferences in value consensus. In H. Grad, A. Blanco, & J. Georgas (Eds.), Key issues in crosscultural psychology (pp. 217226). Lisse: Swets & Zeitlinger Publishers. Shamir, B. (1983). Some antecedents of work-nonwork conflict. Journal of Vocational Behavior, 23(1), p. 98-111. Small, S.A. and Riley, D. (1990). Toward a multidimensional assessment of work spillover into family life. Journal of Marriage & the Family, 52(1), p. 51-61. Stephens, G.K. and Sommer, S.M. (1993). Work-family conflict, job attitudes, and workplace social support: Investigations of measurement and moderation. Presented at the Annual Conference fo r Academy of Management, Atlanta, Georgia. Stephens, G.K. & Sommer, S.M. (1996). The measurement of work to family conflict. Educational and Psychologica l Measurement, v.56(3), p. 475-486.
95 Stewart, W. and Barling, J. (1996). Fathers work experiences affect childrens behaviors via job-related affect a nd parenting behaviors. Journal of Organizational Behavior, 17(3), p. 221-232. Stoeva, A.Z., Chui, R.K., & Greenhaus, J.H. ( 2002). Negative affectivity, role stress, and work-family conflict. Journal of Vocational Behavior, 60, p. 1-16. Thomas, L.T. and Ganster, D.C. (1995). Impact of family-suppor tive work variables on work-family conflict and strain: A control perspective. Journal of Applied Psychology, 80, p. 6-15. Thompson, C.A., Beauvais, L.L., & Lyness, K.S. (1999). When work-family benefits are not enough: The influence of work-f amily culture on benefit utilization, organizational attachment, and work-family conflict. Journal of Vocational Behavior, 54, 392-415. Vandenberg, R.J. and Lance, C.E. (2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3(1), p. 4-70. Voydanoff, Patricia (2002). Linkages be tween the work-family interface and work, family, and individual outcomes: An integrative model. Journal of Family Issues, 23(1), p. 138-164. Wallace, J.E. (1999). Work-to-non-work conflict among married male and female lawyers. Journal of Organizational Behavior, 20, p. 797-816. Wiley, D. L. (1987). The relationship betw een work/nonwork role conflict and jobrelated outcomes: Some unanticipated findlings. Journal of Management, 13, p. 467-472.
96 Zimmerman, T.S., Haddock, S.A., Current, L.R., & Ziemba, S. (2003). Intimate partnership: Foundation to the successf ul balance of work and family. The American Journal of Family Therapy, 31, p. 107-124.
97 Table 1 Chronology of Primary Tools Used in Me asuring Generalized Work/Family Conflict Citation Description Holahan and Gilbert (1979) Six dichotomous m easures of specific interrole conflict: a) Worker vs. Spouse (3 items), b) Worker vs. Parent (4 items), c) Worker vs. Self (4 items), d) Spouse vs. Parent (3 items), e) Spouse vs. Self (4 items), and f) Parent vs. Self (3 items). Burke, Weir, & DuWors (1979) Wives indicated degree to which husbands jobs negatively impacted home and family life. Fifty items assessed 10 areas including requirement to relocate, personal relationship between husband and wife, etc. Pleck, Staines, & Lang (1980) Used Quality of Employment Survey (Quinn & Staines, 1977) data to determine frequency and type of work/family conflict encount ered by various family structures (i.e., with child ren, employed wives, etc). Bohen & Viveros-Long (1981) Developed the J ob-Family Role Strain scale, a 16-item measure that taps job/parent, job spouse, and generalized interrole conflict. Kopelman, Greenhaus, & Connolly (1983) Developed an 8-item measure of work interference with family interrole conflict based on types of work/family conflict reported by Pl eck et al. (1980). Wiley (1987) Factor analyzed 22 item s from the 50-item scale used by Burke et al. (1979). Found four distinct factors emerged: job/person-, job/family-, and family/job conflict, as well as role overload. Burley (1989; presented in Gutek, Searle, & Klepa, 1991) Reconfigured four of the Kopelman et al (1983) items to tap family interference with work, demonstrating that work/family conflict is bi-directional and the two directions are in fact distinct. Stephens & Sommer (1996) Developed the first work interfer ence with family measure tapping generalized st rain-, time-, and behaviorbased conflict as delineated by Greenhaus & Beutel (1985). Netemeyer, Boles, & McMurrian (1996) Developed the first bi-directional generalized work/family conflict scale utilizing prior measurement items. Each scale consists of five items, a nd were shown to be distinct. Carlson, Kacmar, & Williams (2000) Developed first bi-direction, multi-dimensional scale of work/family conflict. The measure consists of six threeitem subscales tapping both the directions (work interference with family a nd family interference with work) and the dimensions (time-, strain-, behavior-based) of conflict delineated by Gr eenhaus & Beutel (1985).
98 Table 2 Komarovskys (1977) Six Modes of Role Tension. Mode Job-Family Example Ambiguity about which norms regulate a certain situation. The allocation of domestic tasks. Lack of congruity between an individuals personality and a particular social role.* When a housewifes need for achievement is frustrated by housewifery. A socially structured insufficiency of resources for role fulfillment. When a single parent cannot find suitable child care facilities. Low reward for role conformity. When a woman feels her home-based work carries less social esteem than employment. Conflict between normative role phenomena. If a father must be at work when other family members are available to be with him. Overload of role obligations. When an individual has too many statuses such as parent, student, child, friend, spouse, worker, and community leader to meet the demands of each status to the satisfaction of all the role partners and th e satisfaction of self. Note: Reproduced from Bohen & Viveros-Long, p. 231. *=not used in developing content of Job-Family Role Strain scale. Table 3 Burley (1989) Work/Family Conflict Scale*. Work Interference with Family Family Interference with Work 1. After work, I come home too tired to do some of the things Id like to do. 1. Im often too tired at work because of the things I have to do at home. 2. On the job I have so much work to do that it takes away from my personal interests. 2. My personal demands are so great that it takes away from my work. 3. My family/friends dislike how often I am preoccupied with my work while I am at home. 3. My superiors and peers dislike how often I am preoccupied with my personal life while at work. 4. My work takes up time that I would like to spend with my family/friends. 4. My personal life takes up time that I would like to spend at work. *=taken from Kopelman, Greenhaus, & Connolly (1983), and reported by Gutek, Searle, & Klepa (1991).
99 Table 4 Carlson et al. Work/F amily Conflict Scale Time-based work interference with family 1. My work keeps me from my family activities more than I would like. 2. The time I must devote to my job keeps me from participating equally in household responsibilities and activities. 3. I have to miss family activities due to the amount of time I must spend on work responsibilities. Time-based family interference with work 4. The time I spend on family responsibilitie s often interferes with my work responsibilities. 5. The time I spend with my family often causes me not to spend time in activities at work that could be helpful to my career. 6. I have to miss work activities due to th e amount of time I must spend on family responsibilities. Strain-based work interference with family 7. When I get home from work I am often too frazzled to participate in family activities/res ponsibilities. 8. I am often so emotionally drained when I get home from work that it prevents me from contributing to my family. 9. Due to all the pressures at work, sometimes when I come home I am too stressed to do the things I enjoy. Strain-based family interference with work 10. Due to stress at home, I am often preo ccupied with family matters at work. 11. Because I am often stressed from family responsibilities, I have a hard time concentrating on my work. 12. Tension and anxiety from my family life often weakens my ability to do my job. Behavior-based work interference with family 13. The problem-solving behaviors I use in my job are not effective in resolving problems at home. 14. Behavior that is effective and necessary for me at work would be counterproductive at home. 15. The behaviors I perform that make me eff ective at work do not help me to be a better parent or spouse. Behavior-based family interference with work 16. The behaviors that work for me at home do not seem to be effective at work. 17. Behavior that is effective and n ecessary for me at home would be counterproductive at work. 18. The problem-solving behaviors that work for me at home do not seem to be as useful at my work.
100 Table 5 Item Intercorrelations for Male s in the United States Sample Item 1 Item 2 Item 3Item 4Item 5Item 6Item 7Item 8 Item 9Item 10 Item 2 .780** Item 3 .653**.679** Item 4 .193 .308** .253* Item 5 .288* .400** .287** .419 Item 6 .267* .416** .337** .428** .726** Item 7 .374**.357** .364** .237* .302** .278* Item 8 .403**.325** .314** .266* .306** .330** .488** Item 9 .402**.300** .357** .225* .332** .329** .464** .726** Item 10 .218* .200 .084 .376** .248* .262* .380** .560** .394** Item 11 .175 .119 .103 .432** .240* .303** .363** .483** .361** .722** Item 12 .210 .187 .129 .277* .464** .437** .293** .520** .363** .574** Item 13 .112 .051 .028 .234* .208 .261* .309** .416** .480** .441** Item 14 .242* .126 .076 .062 .258* .128 .254* .365** .473** .257* Item 15 .342**.198 .105 .182 .288** .247* .320** .502** .457** .528** Item 16 .191 .067 -.043 .080 .182 .235* .139 .494** .379** .404** Item 17 .155 .005 .064 .123 .131 .152 .174 .355** .452** .393** Item 18 .242* .133 .143 .241* .200 .171 .113 .497** .476** .442** Note: N=83; *=p<.05; **=p<.01 Table 5 (continued) Item Intercorrelations for Males in the United States Sample (continued) Item 11 Item 12 Item 13Item 14Item 15Item 16Item 17 Item 12 .701** Item 13 .427** .273* Item 14 .198 .129 .612** Item 15 .396** .394** .544** .565** Item 16 .365** .460** .430** .561** .658** Item 17 .392** .261* .514** .625** .585** .752** Item 18 .355** .270* .601** .605** .519** .694** .718** Note: N=83; *=p<.05; **=p<.01
101 Table 6 Means, Standard Deviations, Reliabilities, and Inte rcorrelation of Scales for Males in the United States Mean SD WIF Time FIW Time WIF Strain FIW Strain WIF Behavior FIW Behavior WIF Time 2.90 1.14 .877 FIW Time 1.77 0.81 .412** .769 WIF Strain 2.45 1.00 .472** .417** .791 FIW Strain 1.72 0.71 .205 .459** .559** .849 WIF Behavior 2.49 1.01 .191 .296** .558** .453** .801 FIW Behavior 2.36 1.04 .133 .225* .451** .471** .741** .885 Note: N=83; *=p<.05; **=p<.01; WIF = work interf erence with family; FIW = family interference with work; Diagonal contains reliabilities. Table 7 Item Intercorrelations for Males in the Australia/New Zealand Sample Item 1 Item 2 Item 3Item 4Item 5Item 6Item 7Item 8 Item 9Item 10 Item 2 .772** Item 3 .753**.742** Item 4 .150 .156 .310** Item 5 .068 .182* .205* .541** Item 6 .039 .113 .237** .624** .706** Item 7 .337**.359** .419** .136 .143 .149 Item 8 .232**.222** .270** .075 .082 .072 .754** Item 9 .329**.344** .351** .139 -.001 .050 .747** .695** Item 10 -.048 -.073 -.059 .409** .307** .409** .055 .147 .132 Item 11 -.027 -.053 .038 .348** .192* .317** .149 .139 .258** .736** Item 12 .054 .014 .104 .309** .214* .272** .229** .198* .301** .631** Item 13 .206* .205* .286** .089 .163* .167* .324** .301** .233** .151 Item 14 .169 .102 .203* .150 .258** .199* .243** .299** .171* .214* Item 15 .124 .141 .254** .167* .241** .275** .331** .359** .215* .179* Item 16 .138 .060 .158 .174* .170* .329** .347** .365** .284** .176* Item 17 .174* .041 .142 .167* .256** .271** .331** .314** .266** .140 Item 18 .189* .211* .270** .175* .191* .278** .357** .319** .328** .185* Note: N=143; *=p<.05; **=p<.01
102 Table 7 (continued) Item Intercorrelations for Males in th e Australia/New Zealand Sample (continued) Item 11 Item 12 Item 13Item 14Item 15Item 16Item 17 Item 12 .875** Item 13 .325** .427** Item 14 .211* .272** .540** Item 15 .218** .313** .634** .776** Item 16 .245** .316** .473** .614** .687** Item 17 .178* .251** .438** .645** .614** .761** Item 18 .220** .317** .537** .616** .702** .750** .723** Note: N=143; *=p<.05; **=p<.01 Table 8 Means, Standard Deviations, Reliabilities, and In tercorrelation of Scales for Males in Australia/New Zealand. Mean SD WIF Time FIW Time WIF Strain FIW Strain WIF Behavior FIW Behavior WIF Time 3.55 1.33 .903 FIW Time 2.14 0.99 .204* .831 WIF Strain 3.10 1.20 .384** .118 .891 FIW Strain 1.64 0.80 -.006 .387** .218**.898 WIF Behavior 2.66 1.07 .233** .254** .345**.321**.848 FIW Behavior 2.54 1.04 .182* .283** .391**.272**.742** .896 Note: N=143; *=p<.05; **=p<.01; WIF = work interfe rence with family; FIW = family interference with work; Diagonal contains reliabilities.
103 Table 9 Item Means and Standard Deviations for Male s in the United States and Australia/New Zealand Samples. US Males ANZ Males Mean St. Dev. Mean St. Dev. Item 1 3.31 1.33 3.76 1.47 Item 2 2.78 1.30 3.66 1.43 Item 3 2.61 1.18 3.22 1.46 Item 4 1.73 0.94 2.08 1.08 Item 5 1.88 1.05 2.31 1.25 Item 6 1.71 0.94 2.04 1.12 Item 7 2.40 1.19 3.22 1.31 Item 8 2.43 1.15 2.82 1.31 Item 9 2.51 1.22 3.25 1.36 Item 10 1.86 0.91 1.71 0.89 Item 11 1.64 0.73 1.62 0.85 Item 12 1.65 0.79 1.59 0.91 Item 13 2.41 1.18 2.58 1.18 Item 14 2.60 1.26 2.69 1.25 Item 15 2.46 1.15 2.73 1.22 Item 16 2.43 1.23 2.60 1.23 Item 17 2.32 1.10 2.55 1.15 Item 18 2.33 1.12 2.47 1.06 Note: US males: N=83; ANZ males: N=143. Likert for US Males 1 to 5, ANZ males 1 to 6.
104 Table 10 Item Intercorrelations for Fema les in the United States Sample Item 1 Item 2 Item 3Item 4Item 5Item 6Item 7Item 8 Item 9Item 10 Item 2 .603** Item 3 .693** .699** Item 4 .115 .150 .082 Item 5 .202* .102 .190* .623** Item 6 .144 .063 .195* .568** .615** Item 7 .271** .395** .516** .194* .305** .343** Item 8 .372** .516** .537** .254** .269** .310** .706** Item 9 .304** .455** .543** .157 .279** .326** .709** .747** Item 10 .158 .055 .208* .415** .376** .554** .203* .166* .148 Item 11 .169* .171* .240** .584** .485** .649** .337** .345** .379** .645** Item 12 .191* .096 .251** .617** .568** .635** .242** .306** .293** .648** Item 13 .081 .161 .136 .169* .136 .288** .353** .208* .234** .264** Item 14 .225** .265** .244** .226** .218** .345** .287** .284** .332** .246** Item 15 .252** .279** .278** .200* .179* .259** .237** .325** .225** .275** Item 16 .303** .304** .332** .097 .171* .165* .280** .357** .383** .282** Item 17 .285** .300** .332** .106 .188* .163 .205* .334** .363** .305** Item 18 .120 .227** .155 .209* .113 .192* .208* .279** .220** .309** Note: N=139; *=p<.05; **=p<.01 Table 10 (continued) Item Intercorrelations for Females in the United States Sample (continued) Item 11 Item 12 Item 13Item 14Item 15Item 16Item 17 Item 12 .848** Item 13 .333** .249** Item 14 .321** .274** .571** Item 15 .270** .248** .470** .503** Item 16 .266** .258** .447** .438** .598** Item 17 .336** .304** .512** .502** .499** .780** Item 18 .335** .302** .380** .243** .411** .540** .496** Note: N=139; *=p<.05; **=p<.01
105 Table 11 Means, Standard Deviations, Reliabilities, and In tercorrelation of Scales for Females in the United States Mean SDWIF Time FIW Time WIF Strain FIW Strain WIF Behavior FIW Behavior WIF Time 2.84 1.08.855 FIW Time 2.00 0.90.184* .819 WIF Strain 2.82 1.12.544** .351** .886 FIW Strain 1.92 0.83.214* .698** .328** .878 WIF Behavior 2.62 0.95.293** .316** .371** .373** .760 FIW Behavior 2.67 0.95.346** .212* .378** .389** .635** .821 Note: N=139; *=p<.05; **=p<.01; WIF = work interfe rence with family; FIW = family interference with work; Diagonal contains reliabilities. Table 12 Item Intercorrelations for Females in the Australia/New Zealand Sample Item 1Item 2Item 3Item 4Item 5Item 6Item 7Item 8Item 9Item 10 Item 2.741** Item 3.762** .729** Item 4.247** .212** .293** Item 5.187** .046 .234** .454** Item 6.101 .014 .161* .460** .791** Item 7.484** .627** .561** .098 .109 -.008 Item 8.471** .629** .544** .046 .086 .049 .865** Item 9.485** .543** .523** .176** .074 -.009 .728** .731** Item 10.077 .088 .119 .122* .190** .246** .115 .125* .135** Item 11.032 .071 .073 .148* .214* .233* .093 .096 .108 .798** Item 12.038 .035 .083 .128* .149* .258** .024 .021 .059 .747** Item 13.139* .112 .157* .060 .087 .098 .172** .177** .212** .422** Item 14.093 .109 .163* .038 .093 .067 .198** .217** .237** .227** Item 15.252** .228** .296** .090 .085 .053 .286** .314** .308** .210** Item 16.187** .212** .256** .097 .047 .075 .258** .298** .294** .183** Item 17.128* .125* .196** .017 .033 .024 .244** .256** .234** .183** Item 18.140* .150* .201** .069 .077 .083 .259** .309** .305** .205** Note: N=249; *=p<.05; **=p<.01
106 Table 12 (continued) Item Intercorrelations for Females in the Australia/New Zealand Sample (continued) Item 11Item 12Item 13Item 14Item 15Item 16Item 17 Item 12 .790** Item 13 .391** .402** Item 14 .260** .179** .387** Item 15 .227** .167** .437** .652** Item 16 .249** .199** .402** .598** .775** Item 17 .223** .170** .339** .702** .644** .688** Item 18 .246** .177** .469** .589** .688** .720** .703** Note: N=249; *=p<.05; **=p<.01 Table 13 Means, Standard Deviations, Reliabilities, and In tercorrelation of Scales for Females in Australia/New Zealand Mean SD WIF Time FIW Time WIF Strain FIW Strain WIF Behavior FIW Behavior WIF Time 2.90 1.29 .896 FIW Time 1.94 0.91 .212** .800 WIF Strain 3.12 1.33 .644** .089 .912 FIW Strain 1.60 0.77 .082 .240** .103 .912 WIF Behavior 2.52 1.02 .234** .109 .316** .365** .745 FIW Behavior 2.59 1.04 .216** .075 .330** .246** .797** .875 Note: N=249; *=p<.05; **=p<.01; WIF = work interfe rence with family; FIW = family interference with work; Diagonal contains reliabilities.
107 Table 14 Item Means and Standard Deviations for Females in the United States and Australia/New Zealand Samples US Females ANZ Females Mean Std. Dev.Mean Std. Dev. Item 1 3.16 1.27 3.13 1.45 Item 2 2.71 1.22 2.92 1.48 Item 3 2.64 1.19 2.63 1.34 Item 4 1.96 0.99 1.84 1.02 Item 5 2.11 1.12 2.06 1.15 Item 6 1.95 1.04 1.91 1.04 Item 7 2.74 1.22 3.04 1.44 Item 8 2.84 1.24 3.03 1.46 Item 9 2.89 1.27 3.29 1.45 Item 10 2.08 0.99 1.68 0.89 Item 11 1.88 0.91 1.57 0.81 Item 12 1.81 0.88 1.55 0.81 Item 13 2.73 1.19 2.31 1.22 Item 14 2.52 1.13 2.55 1.22 Item 15 2.59 1.17 2.69 1.32 Item 16 2.68 1.13 2.58 1.17 Item 17 2.63 1.09 2.67 1.22 Item 18 2.70 1.12 2.51 1.09 Note: US females: N=139; ANZ females: N=249. Likert for US females 1 to 5, ANZ females 1 to 6.
108 Table 15 Item Intercorrelations Between Males in the United States and Australia/New Zealand Item 1 Item 2 Item 3Item 4Item 5Item 6Item 7Item 8 Item 9Item 10 Item 2 .776** Item 3 .729** .736** Item 4 .184** .241** .317** Item 5 .158* .286** .257** .519** Item 6 .129* .238** .287** .576** .719** Item 7 .374** .415** .438** .208** .233** .222** Item 8 .302** .284** .304** .155* .171** .167* .677** Item 9 .376** .383** .387** .202** .142* .168* .683** .711** Item 10 .032 -.002 -.029 .377** .267** .342** .136* .272** .193** Item 11 .033 -.003 .052 .366** .200** .308** .202** .238** .276** .727** Item 12 .097 .056 .102 .290** .278** .313** .228** .288** .300** .610** Item 13 .182** .165* .213** .147* .186** .205** .324** .344** .324** .252** Item 14 .197** .114 .165* .124 .258** .178** .245** .321** .272** .227** Item 15 .209** .184** .225** .186** .269** .278** .343** .414** .311** .292** Item 16 .164* .079 .105 .151** .182** .303** .282** .412** .321** .254** Item 17 .180** .056 .135* .166* .229** .243** .294** .337** .341** .222** Item 18 .214** .194** .234** .204** .201** .247** .276** .384** .382** .277** Note: N=226; *=p<.05; **=p<.01 Table 15 (continued) Item Intercorrelations Between Males in the United States and Australia/New Zealand (continued) Item 11 Item 12 Item 13Item 14Item 15Item 16Item 17 Item 12 .823** Item 13 .356** .371** Item 14 .205** .222** .567** Item 15 .271** .333** .605** .700** Item 16 .282** .360** .460** .595** .678** Item 17 .244** .249** .469** .638** .608** .759** Item 18 .263** .296** .563** .612** .636** .730** .722** Note: N=226; *=p<.05; **=p<.01
109 Table 16 Means, Standard Deviations, and Intercorrelati on of Scales for Males in the United States and Australia/New Zealand Mean SD WIF Time FIW Time WIF Strain FIW Strain WIF Behavior FIW Behavior WIF Time 3.31 1.30 -FIW Time 2.01 0.95 .297**-WIF Strain 2.86 1.17 .447**.242** -FIW Strain 1.67 0.77 .045 .391** .294**-WIF Behavior 2.60 1.05 .232**.276** .416**.359**-FIW Behavior 2.47 1.04 .180**.274** .415**.333**.743** -Note: N=226; *=p<.05; **=p<.01; WIF = work interfe rence with family; FIW = family interference with work. Table 17 Item Intercorrelations Between Females in the United States and Australia/New Zealand Item 1 Item 2 Item 3Item 4Item 5Item 6Item 7Item 8 Item 9Item 10 Item 2 .698** Item 3 .741** .718** Item 4 .205** .188** .224** Item 5 .192** .062 .220** .513** Item 6 .115* .028 .172** .498** .729** Item 7 .418** .565** .544** .121* .167** .101* Item 8 .440** .599** .540** .107* .142** .131** .820** Item 9 .425** .521** .523** .160** .136** .096 .726** .736** Item 10 .104* .060 .147** .237** .258** .358** .118* .122* .106* Item 11 .080 .090 .130** .316** .314** .389** .154** .167** .173** .745** Item 12 .091 .044 .140** .312** .304** .398** .080 .106* .116* .715** Item 13 .120* .113* .149** .106* .106* .165** .206** .173** .190** .383** Item 14 .134** .156** .189** .099* .135** .161** .225** .237** .265** .226** Item 15 .251** .245** .290** .123* .115* .120* .274** .318** .285** .219** Item 16 .224** .236** .280** .099* .090 .107** .259** .313** .313** .223** Item 17 .175** .177** .238* .045 .083 .070 .233** .279** .273** .217** Item 18 .134** .167** .185** .123* .091 .124** .231** .291** .261** .255** Note: N=388; *=p<.05; **=p<.01
110 Table 17 (continued) Item Intercorrelations Between Females in the United States and Australia/New Zealand (continued) Item 11 Item 12 Item 13Item 14Item 15Item 16Item 17 Item 12 .818** Item 13 .386** .361** Item 14 .275** .208** .440** Item 15 .231** .186** .434** .605** Item 16 .257** .224** .418** .544** .715** Item 17 .255** .211** .387** .639** .600** .716** Item 18 .289** .233** .442** .466** .588** .656** .628** Note: N=388; *=p<.05; **=p<.01 Table 18 Means, Standard Deviations, and Intercorrelati on of Scales for Females in the United States and Australia/New Zealand Mean SD WIF Time FIW Time WIF Strain FIW Strain WIF Behavior FIW Behavior WIF Time 2.87 1.22 -FIW Time 1.96 0.90 .202**-WIF Strain 3.01 1.27 .615**.166** -FIW Strain 1.71 0.81 .118* .409** .151** -WIF Behavior 2.55 1.00 .250**.181** .325** .369** -FIW Behavior 2.62 1.01 .254**.122* .337** .297** .745** -Note: N=388; *=p<.05; **=p<.01; WIF = work interfe rence with family; FIW = family interference with work.
111 Table 19 Level of Overall Fit Across Multiple Models of Invariance 2 2 RMSEA SRMRNNFI Males Variance/Covariance Invariance 282.03 n/a .062 .100 .92 Factor Pattern Invariance (2uncon) 445.30 n/a .079 .070 .90 Factor Pattern and Loading Invariance 475.67 30.37*.081 .140 .84 Partial Measurement Invariance 460.44 15.14 .078 .100 .91 Females Variance/Covariance Invariance 427.29 n/a .080 .120 .90 Factor Pattern Invariance (2uncon) 464.76 n/a .066 .049 .94 Factor Pattern and Loading Invariance 516.33 51.57*.068 .12 .93 Partial Measurement Invariance 473.30 8.54 .065 .064 .94 Note: *=p<.05; RMSEA=Root Mean Square Error of Approximation; SRMR=Standardized Root Mean Square Residual; NNFI=Non-Normed Fit Index. Table 20 Steps in Statistical Identification of the Model for Males and Females Step Identification? WIF Time FIW Time WIF Strain FIW Strain WIF Behavior FIW Behavior Males 1 Yes Ref. Ref. Ref. Ref. Ref. Ref. 2 Yes Var. Ref. Ref. Ref. Ref. Ref. 3 Yes Var. Var. Ref. Ref. Ref. Ref. 4 Yes Var. Var. Var. Ref. Ref. Ref. 5 Yes Var. Var. Var. Var. Ref. Ref. 6 Yes Var. Var. Var. Var. Var. Ref. 7 Yes Var. Var. Var. Var. Var. Var. Females 1 Yes Ref. Ref. Ref. Ref. Ref. Ref. 2 Yes Var. Ref. Ref. Ref. Ref. Ref. 3 Yes Var. Var. Ref. Ref. Ref. Ref. 4 Yes Var. Var. Var. Ref. Ref. Ref. 5 No Var. Var. Var. Var. Ref. Ref. 6 Yes Var. Var. Var. Ref. Var. Ref. 7 Yes Var. Var. Var. Ref. Var. Var. Note: WIF=Work Interference with family; FIW=Family interference with work; Ref.=Reference variable used in latent variable scaling; Var.=start va lue used in direct latent variable scaling.
112 Table 21 Partial Measurement Invariance Fit Statistics for Males. 2 2 RMSEARMSEA SRMRSRMR NNFI NNFI (2uncon) 445.30 n/a .079 n/a .070 n/a 0.90 n/a Item 1 446.05 0.75 .079 0.000 .070 0.000 .90 0.00 Item 2 445.52 0.22 .079 0.000 .070 0.000 .90 0.00 Item 3 450.72 5.42* .080 +.001 .082 +.012 .90 0.00 Item 4 448.90 3.60 .080 +.001 .073 +.003 .90 0.00 Item 5 445.60 0.30 .079 0.000 .070 0.000 .90 0.00 Item 6 447.99 2.69 .079 0.000 .074 0.004 .90 0.00 Item 7 451.34 6.04* .081 +.002 .077 +.007 .90 0.00 Item 8 445.64 0.34 .079 0.000 .070 0.000 .90 0.00 Item 9 445.39 0.09 .079 0.000 .070 0.000 .90 0.00 Item 10 445.74 0.44 .079 0.000 .074 +.004 .90 0.00 Item 11 450.36 5.06* .080 +.001 .091 +.021 .90 0.00 Item 12 448.74 3.44 .079 0.00 .082 +.012 .90 0.00 Item 13 445.44 0.14 .079 0.00 .072 +.002 .90 0.00 Item 14 445.55 0.25 .079 0.00 .071 +.001 .90 0.00 Item 15 448.11 2.81 .079 0.00 .083 +.013 .90 0.00 Item 16 445.37 0.07 .079 0.00 .070 0.000 .90 0.00 Item 17 445.30 0.00 .079 0.00 .070 0.000 .90 0.00 Item 18 445.30 0.00 .079 0.00 .071 +.001 .90 0.00 Note: *=p<.05; RMSEA=Root Mean Square Error of Approximation; SRMR=Standardized Root Mean Square Residual; NNFI=Non-Normed Fit Index.
113 Table 22 Partial Measurement Invariance Fit Statistics for Females 2 2 RMSEARMSEA SRMRSRMR NNFI NNFI (2uncon) 464.76 n/a .066 n/a .049 n/a .94 n/a Item 1 470.11 5.35* .067 +0.001 .064 +.015 .94 0.00 Item 2 471.60 6.84* .067 +0.001 .067 +.018 .93 -0.01 Item 3 465.11 0.35 .066 0.000 .050 +.001 .94 0.00 Item 4 469.25 4.49* .066 0.000 .059 +.010 .94 0.00 Item 5 467.41 2.65 .066 0.000 .057 +.008 .94 0.00 Item 6 465.43 0.67 .066 0.000 .051 +.002 .94 0.00 Item 7 473.06 8.30* .067 +0.001 .080 +.031 .93 -0.01 Item 8 470.53 5.77* .067 +0.001 .072 +.023 .93 -0.01 Item 9 464.95 0.19 .066 0.000 .050 +.001 .94 0.00 Item 10 468.71 3.95* .066 0.000 .052 +.003 .94 0.00 Item 11 n/a n/a n/a n/a n/a n/a n/a n/a Item 12 465.23 0.47 .066 0.000 .049 0.000 .94 0.00 Item 13 467.62 2.86 .066 0.000 .056 +.007 .94 0.00 Item 14 465.49 0.73 .066 0.000 .050 +.001 .94 0.00 Item 15 470.64 5.88* .066 0.000 .065 +.017 .93 -0.01 Item 16 464.77 0.01 .066 0.000 .049 0.000 .94 0.00 Item 17 464.88 0.08 .066 0.000 .049 0.000 .94 0.00 Item 18 469.75 4.99* .066 0.000 .060 +.011 .94 0.00 Note: *=p<.05; All 2 based on 1 degree of freedom; RMSEA=Root Mean Square Error of Approximation; SRMR=Standardized Root Mean Square Residual; NNFI=Non-Normed Fit Index. Table 23 Within-Group Completely Standardize d Manifest Variable Loadings and Modification Indices for Males US ANZ MI US ANZMI US ANZMI Item 1 0.88 0.86 0.32 Item 2 0.91 0.87 0.03 Item 3 0.78 0.86 n/a Item 4 0.65 0.66 1.78 Item 5 0.85 0.77 0.24 Item 6 0.89 0.89 0.98 Item 7 0.61 0.90 n/a Item 8 0.89 0.83 0.03 Item 9 0.85 0.83 0.25 Note: US=United States; ANZ=Australia/New Zealand; MI=Modification Index.
114 Table 23 (continued) Within-Group Completely Standardized Mani fest Variable Loadings and Modification Indices for Males (continued) US ANZ MI US ANZMI US ANZMI Item 10 0.76 0.76 3.66 Item 11 0.89 0.99 n/a Item 12 0.85 0.87 4.86 Item 13 0.70 0.69 0.81 Item 14 0.78 0.83 0.00 Item 15 0.83 0.91 2.76 Item 16 0.86 0.88 0.01 Item 17 0.87 0.85 0.09 Item 18 0.82 0.86 0.19 Note: US=United States; ANZ=Australia/New Zealand; MI=Modification Index. Table 24 Within-Group Completely Standardized Phi Solution for Males US ANZUS ANZUS ANZUS ANZ US ANZ WIF Time 1.00 1.00 FIW Time 0.54 0.19 1.001.00 WIF Strain 0.55 0.43 0.540.13 1.00 1.00 FIW Strain 0.30 -0.020.560.34 0.69 0.20 1.001.00 WIF Behavior 0.32 0.22 0.430.30 0.73 0.38 0.580.27 1.00 1.00 FIW Behavior 0.20 0.19 0.320.33 0.61 0.43 0.560.25 0.89 0.84 Note: WIF=Work Interference with Family; FIW=Fa mily Interference with Work; US=United States; ANZ=Australia/New Zealand.
115 Table 25 Within-Group Completely Standardized Theta-Delta Error Terms for Males US ANZ Item 1 0.23 0.26 Item 2 0.17 0.24 Item 3 0.39 0.26 Item 4 0.58 0.57 Item 5 0.29 0.40 Item 6 0.20 0.21 Item 7 0.63 0.19 Item 8 0.21 0.31 Item 9 0.28 0.32 Item 10 0.42 0.43 Item 11 0.21 0.02 Item 12 0.28 0.24 Item 13 0.51 0.52 Item 14 0.38 0.30 Item 15 0.31 0.18 Item 16 0.26 0.22 Item 17 0.24 0.28 Item 18 0.33 0.26 Note: US=United States; ANZ=Australia/New Zealand. Table 26 Within-Group Completely Standardize d Manifest Variable Loadings and Modification Indi ces for Females US ANZ MI US ANZMI US ANZMI Item 1 0.75 0.85 n/a Item 2 0.78 0.86 n/a Item 3 0.93 0.86 0.20 Item 4 0.77 0.51 n/a Item 5 0.78 0.86 1.69 Item 6 0.84 0.90 0.07 Item 7 0.82 0.93 n/a Item 8 0.87 0.93 n/a Item 9 0.87 0.79 0.04 Note: US=United States; ANZ=Australia/New Zealand; MI=Modification Index.
116 Table 26 (continued) Within-Group Completely Standardized Mani fest Variable Loadings and Modification Indices for Females (continued) US ANZ MI US ANZMI US ANZMI Item 10 0.72 0.87 n/a Item 11 0.92 0.92 0.46 Item 12 0.93 0.86 0.79 Item 13 0.63 0.57 4.11 Item 14 0.76 0.74 1.24 Item 15 0.73 0.86 n/a Item 16 0.89 0.86 0.00 Item 17 0.89 0.81 0.12 Item 18 0.59 0.84 n/a Note: US=United States; ANZ=Australia/New Zealand; MI=Modification Index. Table 27 Within-Group Completely Standardized Phi Solution for Females US ANZUS ANZUS ANZUS ANZ US ANZ WIF Time 1.00 1.00 FIW Time 0.24 0.16 1.001.00 WIF Strain 0.67 0.70 0.440.06 1.00 1.00 FIW Strain 0.29 0.09 0.850.27 0.42 0.10 1.001.00 WIF Behavior 0.38 0.30 0.440.10 0.47 0.36 0.460.35 1.00 1.00 FIW Behavior 0.43 0.25 0.240.08 0.45 0.35 0.390.28 0.79 0.98 Note: WIF=Work Interference with Family; FIW=Fa mily Interference with Work; US=United States; ANZ=Australia/New Zealand.
117 Table 28 Within-Group Completely Standardized Theta-Delta Error Terms for Females US ANZ Item 1 0.44 0.28 Item 2 0.40 0.26 Item 3 0.13 0.25 Item 4 0.41 0.74 Item 5 0.39 0.27 Item 6 0.29 0.18 Item 7 0.32 0.14 Item 8 0.24 0.14 Item 9 0.24 0.38 Item 10 0.48 0.24 Item 11 0.15 0.16 Item 12 0.14 0.26 Item 13 0.60 0.68 Item 14 0.43 0.45 Item 15 0.47 0.26 Item 16 0.21 0.25 Item 17 0.21 0.35 Item 18 0.65 0.30 Note: US=United States; ANZ=Australia/New Zealand.
118 Table 29 Exploratory Analysis Loadings and Pattern s in a Forced 6-Factor Solution for Australia/New Zealand Males Item Factor 1 Factor 2 Factor 3Factor 4Factor 5Factor 6 Item 18 .829 Item 17 .875 Item 16 .878 Item 15 .805 .432 Item 14 .770 Item 13 .523 .359 Item 12 .870 Item 11 1.018 Item 10 .665 Item 9 .811 Item 8 .879 Item 7 .887 Item 6 .873 Item 5 .812 Item 4 .648 Item 3 .816 Item 2 .854 Item 1 .921 Note: ANZ = Australian/New Zealand; all loadings less than .300 suppressed.
119 Table 30 Exploratory Analysis Loadings and Patterns in a Forced 5-Factor Solution between US and Australia/New Zealand Males 1 2 3 4 5 Item USA ANZ USA ANZUSAANZUSAANZUSAANZ Item 18 .809 .813 Item 17 .861 .803 Item 16 .859 .815 -.504 Item 15 .617 .879 Item 14 .931 .835 -.324 Item 13 .645 .599 .353 Item 12 .775 .874 Item 11 .967 1.013 Item 10 .808 .662 Item 9 .417 .834 Item 8 .439 .844 Item 7 .876.324 Item 6 .801 .893 Item 5 .884 .776 Item 4 .308 .331 .660 Item 3 .837.763 Item 2 .873.865 Item 1 .874.939 Note: ANZ = Australian/New Zealand; all loadings less than .300 suppressed.
120 Table 31 Exploratory Analysis Loadings and Patterns in a Forced 6-Factor Solution between US and Australia/New Zealand Females 1 2 3 4 5 6 Item USA ANZ USA ANZUSAANZUSAANZUSAANZUSA ANZ Item 18 .822 .554 Item 17 .683 .757 .568 Item 16 .873 .991 Item 15 .901 .412 .350 Item 14 .687 .664 Item 13 .436 .328 .758 Item 12 .926 .885 Item 11 .922 .896 Item 10 .735 .881 Item 9 .877.666 Item 8 .785.966 Item 7 .844.917 Item 6 .588 .896 Item 5 .342 .900.678 Item 4 .541 .475.384 Item 3 .818.809 Item 2 .693.748 Item 1 .888.916 Note: ANZ = Australian/New Zealand; all loadings less than .300 suppressed.
121 Table 32 Exploratory Analysis Loadings and Patterns in a Forced 5-Factor Solution between US and Australia/New Zealand Females 1 2 3 4 5 Item USA ANZ USA ANZUSAANZUSAANZUSAANZ Item 18 .828 .629 Item 17 .843 .801 Item 16 .856 .852 Item 15 .846 .437 .354 Item 14 .767 .646 Item 13 .382 .340 .349 .540 Item 12 .909 .886 Item 11 .804 .891 Item 10 .665 .882 Item 9 .874.659 Item 8 .779.965 Item 7 .851.920 Item 6 .762 .897 Item 5 .674 .900 Item 4 .757 .474 Item 3 .771.797 Item 2 .681.745 Item 1 .902.914 Note: ANZ = Australian/New Zealand; all loadings less than .300 suppressed.
122 Table 33 Principal Components Analysis Loadings and Patterns in a Forced 6-Component Solution between US and Australia/New Zealand Males 1 2 3 4 5 6 Item USA ANZ USA ANZUSAANZUSAANZUSAANZUSA ANZ Item 18 .929 .858 Item 17 .930 .965 Item 16 .623 .943 Item 15 .848 .548 .496 Item 14 .855 .594 -.361 .420 Item 13 .583 .377 .796 Item 12 .836 .904 Item 11 .866 .967 Item 10 .773 .791 Item 9 .848.641 Item 8 .396 .944.893 Item 7 .904.505 Item 6 .861 .873 Item 5 .913 .938 Item 4 .718.826 Item 3 .852.833 Item 2 .898.876 Item 1 .956.909 Note: ANZ = Australian/New Zealand; all loadings less than .300 suppressed.
123 Table 34 Principal Components Analysis Loadings and Patterns in a Forced 5-Component Solution between US and Australia/New Zealand Males 1 2 3 4 5 Item USA ANZ USAANZUSAANZUSAANZUSA ANZ Item 18 .867 .835 Item 17 .910 .841 Item 16 .913 .839 -.336 Item 15 .654 .892 Item 14 .836 .876 -.361 Item 13 .507 .663 -.321 .492 Item 12 .818 .915 Item 11 .871 .967 Item 10 .778 .777 Item 9 .882 .584 Item 8 .390 .913 .416 Item 7 .881 .835 Item 6 .869 .876 Item 5 .892 .871 Item 4 .558 .763 Item 3 .869.805 Item 2 .912.860 Item 1 .918.914 Note: ANZ = Australian/New Zealand; all loadings less than .300 suppressed.
124 Table 35 Principal Components Analysis Loadings and Patterns in a Forced 6-Component Solution between US and Australia/New Zealand Females 1 2 3 4 5 6 Item USA ANZ USA ANZUSAANZUSAANZUSAANZUSA ANZ Item 18 .950 .774 Item 17 .654 .950 Item 16 .769 .868 Item 15 .427 .821 .473 Item 14 .860 .920 Item 13 .811 .952 Item 12 .754.914 Item 11 .780.933 Item 10 .978.898 Item 9 .786 .894 Item 8 .938 .819 Item 7 .915 .925 Item 6 .473 .407.925 Item 5 .870.939 Item 4 .402.885.618 Item 3 .820 .813 Item 2 .788 .753 Item 1 .929 .986 Note: ANZ = Australian/New Zealand sample; all loadings less than .3 suppressed.
125 Table 36 Principal Components Analysis Loadings and Patterns in a Forced 5-Component Solution between US and Australia/New Zealand Females 1 2 3 4 5 Item USA ANZ USAANZUSAANZUSAANZUSA ANZ Item 18 .857 .894 Item 17 .882 .650 .315 Item 16 .878 .705 Item 15 .867 .567 Item 14 .827 .902 Item 13 .411 .443 .768 Item 12 .872 .930 Item 11 .791 .910 Item 10 .715 .921 .337 Item 9 .880 .783 Item 8 .809 .937 Item 7 .909 .914 Item 6 .795 .921 Item 5 .763 .934 Item 4 .833 .400 .621 Item 3 .790 .819 Item 2 .773 .785 Item 1 .984 .933 Note: ANZ = Australian/New Zealand sample; all loadings less than .3 suppressed. Table 37 Comparison of Overall Fit Ac ross Multiple Models for Me asurement Invariance between Australian/New Zealand Males and Females 2 2 RMSEA SRMRNNFI Males Variance/Covariance Invariance 322.74 n/a .062 .081 .94 Factor Pattern Invariance (2 uncon) 497.98 n/a .070 .060 .93 Factor Pattern and Loading Invariance 529.04 31.06*.071 .083 .93 Partial Measurement Invariance 519.71 22.23 .070 .075 .93 Note: *=p<.05; RMSEA=Root Mean Square Error of Approximation; SRMR=Standardized Root Mean Square Residual; NNFI=Non-Normed Fit Index.
126 Table 38 Partial Measurement Invariance Fit Sta tistics between ANZ Males and Females 2 2 RMSEARMSEA SRMRSRMR NNFI NNFI (2uncon) 497.98 n/a .070 n/a .060 n/a .93 n/a Item 1 498.11 0.13 .070 0.000 .060 0.000 .93 0.00 Item 2 498.02 0.04 .070 0.000 .059 -.001 .93 0.00 Item 3 498.65 0.67 .070 0.000 .062 +.002 .93 0.00 Item 4 502.26 4.28* .071 +.001 .068 +.008 .93 0.00 Item 5 498.11 0.13 .070 0.000 .059 -.001 .93 0.00 Item 6 498.49 0.51 .070 0.000 .061 +.001 .93 0.00 Item 7 499.57 2.09 .070 0.000 .062 +.002 .93 0.00 Item 8 502.67 4.69* .071 +.001 .069 +.009 .93 0.00 Item 9 497.98 1.00 .070 0.000 .059 -.001 .93 0.00 Item 10 500.03 2.05 .070 0.000 .059 -.001 .93 0.00 Item 11 500.24 2.16 .070 0.000 .067 +.007 .93 0.00 Item 12 500.34 2.26 .070 0.000 .069 +.009 .93 0.00 Item 13 500.03 2.05 .070 0.000 .067 +.007 .93 0.00 Item 14 499.41 1.43 .070 0.000 .065 +.005 .93 0.00 Item 15 497.98 0.00 .070 0.000 .059 -.001 .93 0.00 Item 16 498.48 0.50 .070 0.000 .063 +.003 .93 0.00 Item 17 498.02 0.04 .070 0.000 .059 -.001 .93 0.00 Item 18 497.98 0.00 .070 0.000 .060 0.000 .93 0.00 Note: *=p<.05; RMSEA=Root Mean Square Error of Approximation; SRMR=Standardized Root Mean Square Residual; NNFI=Non-Normed Fit Index.
127 Table 39 Within-Group Completely Standardized Mani fest Variable Loadings and Modification Indices for ANZ Males and Females Males Females MI MalesFemalesMI Males Females MI Item 1 0.86 0.86 0.08 Item 2 0.88 0.86 0.42 Item 3 0.84 0.88 0.99 Item 4 0.70 0.52 n/a Item 5 0.79 0.87 0.45 Item 6 0.88 0.91 0.58 Item 7 0.92 0.92 2.49 Item 8 0.84 0.93 n/a Item 9 0.83 0.79 0.35 Note: MI=Modification Index. Table 39 (continued) Within-Group Completely Standardized Mani fest Variable Loadings and Modification Indices for ANZ Males and Females (continued) Males Females MI Males FemalesMI Males FemalesMI Item 10 0.78 0.85 10.03 Item 11 0.97 0.94 3.14 Item 12 0.87 0.87 2.57 Item 13 0.63 0.56 1.87 Item 14 0.81 0.77 1.50 Item 15 0.93 0.86 0.86 Item 16 0.87 0.87 0.68 Item 17 0.85 0.81 0.52 Item 18 0.86 0.84 0.12 Note: MI=Modification Index. Table 40 Within-Group Completely Standardized Phi Solution for ANZ Males and Females M F M F M F M F M F WIF Time 1.00 1.00 FIW Time 0.20 0.18 1.00 1.00 WIF Strain 0.44 0.69 0.14 0.07 1.00 1.00 FIW Strain 0.01 0.09 0.36 0.29 0.21 0.11 1.00 1.00 WIF Beh. 0.22 0.30 0.30 0.12 0.38 0.36 0.27 0.36 1.00 1.00 FIW Beh. 0.20 0.25 0.33 0.09 0.44 0.36 0.26 0.30 0.83 0.98 Note: WIF=Work Interference with Family; FIW= Family Interference with Work; Beh.=Behavior
128 Table 41 Within-Group Completely Standardized Theta-Delta Error Terms for ANZ Males and Females Males Females Item 1 0.25 0.27 Item 2 0.23 0.27 Item 3 0.29 0.23 Item 4 0.51 0.73 Item 5 0.38 0.25 Item 6 0.22 0.16 Item 7 0.15 0.15 Item 8 0.30 0.14 Item 9 0.31 0.37 Item 10 0.39 0.28 Item 11 0.05 0.13 Item 12 0.24 0.24 Item 13 0.61 0.68 Item 14 0.34 0.40 Item 15 0.13 0.27 Item 16 0.24 0.24 Item 17 0.28 0.34 Item 18 0.25 0.30 Table 42 Mean Group Comparisons Between Austra lian/New Zealand Males and Females Gender N Mean SD t df WIF Time Males 143 3.547791.331224.717**289.110 Females 249 2.895581.29369 FIW Timea Males 143 2.144520.993372.054* 274.49 Females 249 1.937080.90699 WIF Straina Males 143 3.095571.20421-0.200 321.05 Females 249 3.121821.33262 FIW Strain Males 143 1.641030.804020.513 286.082 Females 249 1.598390.77151 WIF Behavior Males 143 2.664341.067001.328 284.946 Females 249 2.518071.01898 FIW Behavior Males 143 2.540791.04425-0.441 294.651 Females 249 2.589021.03835 Note: *=p<.05; **p
129 Figure 1. Carlson et al. (2000) model of work/family conflict.
130 6,1 4,2 4,1 2,1 6,2 5,2 5,1 6,4 6,3 5,3 6,5 5,4 4,3 5 6 4 3 2 1 1,1 2,2 3,3 4,4 5,5 6,6 x1 1,1 x2 2,2 x3 3,3 x4 4,4 x5 5,5 x6 6,6 x7 7,7 x8 8,8 x9 9,9 x10 10,10 x11 11,11 x12 12,12 x13 13,13 x14 14,14 x15 15,15 x16 16,16 x17 17,17 x18 18,18 1 x2, 1 x3, 11 x5, 2 x6, 21 x8, 3 x9, 31 x11, 4 x12, 41 x14, 5 x15, 51 x17, 6 x18, 6 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10,1 0 11,1 1 12,1 2 13,1 3 14,1 4 15,1 5 16,1 6 17,1 7 18,1 8
131 About the Author David Herst began his career in psychol ogy as a Licensed Psychiatric Technician, graduating as valedictorian of his 1992 class from Mt. San Antonio Community College. He then received a Bachelors of Science degree from the Univer sity of Wyoming in 1995, and his Masters Degree in Industrial /Organizational Psychology in 2001 from the University of South Florida on route to his Ph.D. While in the Ph.D. program at the Univer sity of South Florid a, Mr. Herst taught extensively in the undergraduate program in the psychology department. He has presented at numerous conferences, includi ng the Society for Indus trial/Organizational Psychology, Academy of Management, and Amer ican Psychological Association. He is a published author, and has repeatedly served as a reviewer for the Southern Management Associations annual conference.
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 001441507
007 cr mnu|||uuuuu
008 031203s2003 flua sbm s000|0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0000181
Herst, David Evan Loran.
Cross-cultural measurement invariance of work/family conflict scales across English-speaking samples
h [electronic resource] /
by David Evan Loran Herst.
[Tampa, Fla.] :
University of South Florida,
Thesis (Ph.D.)--University of South Florida, 2003.
Includes bibliographical references.
Text (Electronic thesis) 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 141 pages.
ABSTRACT: The factor structure of the work/family conflict scale developed by Carlson, Kacmar, & Williams (2000) was analyzed for measurement invariance between a US and an Australian/New Zealand (ANZ) sample using a multisample confirmatory factor analysis procedure. Results indicated that factor pattern fit between the female samples on the common model was good-to-mediocre, and factor pattern fit between the male samples and the common model was mediocre-to-poor. Both samples exhibited significant changes in chi square when testing for the more restrictive factor loading equivalence. Partial measurement invariance revealed a better fit between the male samples when three of the items were unconstrained, and when eight items were unconstrained between the female samples. Finally, males and females in the ANZ sample exhibited factor pattern invariance, but required two items to be unconstrained before factor loading invariance was achieved. Mean differences on the six scales revealed higher levels of time-based work interference with family and family interference with work for males than for females in the ANZ sample.
Adviser: Brannick, Michael
t USF Electronic Theses and Dissertations.