|USFDC Home | USF Electronic Theses and Dissertations||| RSS|
This item is only available as the following downloads:
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 001916927
007 cr mnu|||uuuuu
008 071114s2007 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002018
Hogeboom, David L.
The association between Internet use and characteristics of social networking for middle aged and older adults
h [electronic resource] /
by David L. Hogeboom.
[Tampa, Fla.] :
b University of South Florida,
ABSTRACT: BACKGROUND: Studies have shown that strong social networks have a positive effect on physical and psychological well-being. Research suggests that Internet use may affect social networks. However it is not clear if Internet use has a positive or negative effect on social networks. One theory suggests that Internet use displaces face-to-face contacts and off line social participation. Another theory suggests Internet use replaces high quality face-to-face ties with weaker online ties. Other studies however suggest the Internet has a positive effect on social networks. Because older adults have shrinking social networks, but may have more discretionary time than other age groups, the Internet may be a tool that can be used to strengthen social networks for this age group. METHODS: This study uses a sample from the 2004 wave of the Health and Retirement Survey to assess the association between Internet use and social networks.Age is tested for moderation of the association between Internet use and social networks. Oversampling and design effects of the sample are accounted for using weights and special procedures in SAS version 9.1. Univariate, bivariate and linear regression analyses are employed for the examination of associations and moderation. RESULTS: In regression models (n=2,284) considering a number of control variables, frequency of contact with friends, frequency of contact with family, and attendance at organizational meetings (not including religious services), were found to have a significant positive association with Internet use, while in-person contact with family members (other than children) had a significant negative association with Internet use. Age was not found to moderate any of the significant associations between Internet use and measures of social networking.CONCLUSIONS: Results suggest the Internet could be used as a tool in interventions designed to strengthen social networks for older adults and that policies to increase the availability of the Internet should be considered. Internet use is not associated with a decrease in social participation based on attendance of religious services or other organizations. The amount of time spent on Internet use is not considered in this study and is a limitation.
Thesis (M.S.)--University of South Florida, 2007.
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 53 pages.
Advisor: Robert J. McDermott, Ph.D.
x Public Health
t USF Electronic Theses and Dissertations.
The Association Between Internet Use and Characteristics of Social Networking for Middle Aged and Older Adults by David L. Hogeboom A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Public Health Department of Community and Family Health College of Public Health University of South Florida Major Professor: Robert J. Mcdermott, Ph.D. Karen M. Perrin, Ph.D. Julie A. Baldwin, Ph.D. Date of Approval: April 10, 2007 Keywords: Online, Friends, Family, Regression, Complex Sample, HRS Copyright 2007, David L. Hogeboom
Acknowledgements A special thanks to Dr. Robert McDerm ott and Dr. Kay Perrin for their thoughtful guidance on this Thesis and throughout my degr ee. Thanks also to Dr. Julie Baldwin for stepping into my Thesis committee and taking the time to help me improve my work. I also would like to express gratitude to th e following people who helped make my Thesis a reality. Dr. Jeffrey Kromrey for helping develop my initial plans for analysis and sharing information which helped me more clearly understand large complex random samples. Dr. Hana Osman for assisting with editing my drafts and sharing her expertise on older adults. Bethany Bell-Ellison, MPH, fo r taking time to help me work through the details involved with statisti cal analysis of complex random samples. I also want to thank my friends and family for giving me support and encouragement when I needed it.
i Table of Contents List of Tables................................................................................................................. .....ii Abstract....................................................................................................................... .......iii Introduction................................................................................................................... ......1 Scope of the Problem..............................................................................................1 Purpose of Study.....................................................................................................2 Review of Literature........................................................................................................... 3 Social Networks......................................................................................................3 Theories on Social Networks and Health................................................................4 The Internet and Social Networks...........................................................................5 Older Adults, Social Networks, and the Internet....................................................6 Conceptual Framework and Hypothesis.............................................................................8 Methods........................................................................................................................ .....11 Sample...................................................................................................................11 Measures...............................................................................................................13 Statistical Analysis................................................................................................17 Results........................................................................................................................ .......23 Sample Demographics..........................................................................................23 Bivariate Analysis Results....................................................................................24 Regression Results................................................................................................27 Discussion..................................................................................................................... ....34 Conclusions...........................................................................................................34 Limitations............................................................................................................37 Implications for Health Edu cation and Public Health..........................................38 Future Directions..................................................................................................39 References..................................................................................................................... ....40 Appendices..................................................................................................................... ...45 Appendix A: Data Descriptions............................................................................46 Appendix B: Tables of Association......................................................................51
ii List of Tables Table 1: Excerpts of the conceptual models by Berkman & Glass (2000) of how social networks impact health 4 Table 2: Sample description ( N =2284) 24 Table 3: Comparison of Inte rnet users and non-users 26 Table 4: Regression models for research question one: Internet use associations with social networking measures 27 Table 5: Regression model parameters fo r research question one where Internet use is significant 28 Table 6: Regression models for research que stion two: Internet use associations with social networking measures 30 Table 7: Regression model parameters fo r research question two where Internet use is significant 31 Table 8: Tests for interaction be tween age and Internet use 33 Table B-1: Chi-square, Pvalue, and Cramers V for categorical variables 51 Table B-2: Correlation matrix 52
iii The Association Between Internet Use a nd Characteristics of Social Networks for Middle Aged and Older Adults David L. Hogeboom ABSTRACT Background : Studies have shown that strong social networks have a positive effect on physical and psychological well-being. Resear ch suggests that Internet use may affect social networks. However it is not clear if Internet use has a positive or negative effect on social networks. One theory suggests that Internet use displaces face-to-face contacts and off line social participation. Another theory suggests Internet use replaces high quality face-to-face ties with weaker online ties. Other studies however suggest the Internet has a positive effect on social netw orks. Because older adults have shrinking social networks, but may have more discretio nary time than other age groups, the Internet may be a tool that can be used to stre ngthen social networks for this age group. Methods : This study uses a sample from the 2004 wave of the Health and Retirement Survey to assess the association between Intern et use and social networks. Age is tested for moderation of the association between Internet use and social networks. Oversampling and design effects of the sa mple are accounted for using weights and special procedures in SAS version 9.1. Un ivariate, bivariate and linear regression analyses are employed for the examinati on of associations and moderation. Results: In regression models ( n =2,284) considering a number of control variables, frequency of contact with friends, frequency of contact with family, and attendance at
iv organizational meetings (not including religious services ), were found to have a significant positive association with Internet use, while in-person contact with family members (other than children) had a significant negative association with Internet use. Age was not found to moderate any of the si gnificant associations between Internet use and measures of social networking. Conclusions : Results suggest the Inte rnet could be used as a tool in interventions designed to strengthen social networks for ol der adults and that policies to increase the availability of the Internet s hould be considered. Internet use is not asso ciated with a decrease in social participation based on attendance of religious services or other organizations. The amount of time spent on Inte rnet use is not consid ered in this study and is a limitation.
1 Introduction Scope of the Problem Available evidence suggests that strong social networks help manage stress, reduce depression, and improve health outco mes (Berkman, 1985; Cohen & Syme, 1985; Crawford, 1987; Lubben & Gironda, 1996; Seem an, 1996). It also appears that the Internet is transforming the way people comm unicate, and may affect social networks (Bargh & McKenna, 2004; Coget, Yamauchi & Suman, 2002; Katz & Aspden, 1997; Kraut et al., 2002; Kraut et al., 1998; Nie & Hillygus, 2002). The literature contains contradictory studies on how Internet use may influence social netw orks. Some studies suggest that the Internet dist racts users from their real life social networks, thereby weakening their social ties (K raut et al., 1998; Nie & Hillygu s, 2002). They point to the Internet paradox where people communicate mo re, yet their social networks suffer because of reduced off-line social interacti on. Other studies suggest that the Internet helps people create new social ties and strengthen social networks (Katz & Aspden, 1997; Kraut et al., 2002). The Pew Internet & American Life Projec t found that in the United States 88% of 18-29 year-olds, 84% of 30-49 year-olds, 71% of 50-64 year-olds, and 34% of those 65 and older went online in January 2006 (F ox, 2006). In 2004, only 22% of Americans 65 and older went online (Fox, 2004). Because few ol der adults use the Internet, there is a dearth of studies concerning the influence Inte rnet use may have on their social networks.
2 Considering the challenges older adults f ace in maintaining a strong social network, including reduction in network size, reduced mobility, and health problems associated with aging, the Internet may be of use in strengthening their social network. Older adults have more discretionary time than other age groups (Moss & Lawton, 1982), and may, therefore, not have the time constraints younger age groups face concerning Internet use and off-line social participation. Purpose of Study The problem considered in this study is wh ether or not the Internet can be used as a tool to strengthen the social networks of older adults. To make that determination the association between Internet use and social ne tworks for older adults must be examined. It is not clear in the literature if use of th e Internet strengthens so cial ties, or weakens them. There may be differences between th e older adults and other age groups based on employment status, family stage, and health status. This study adds to the body of literature on social networki ng by examining the association between Internet use and characteristics of social networks of middle ag ed and older adults (51 years or older). To account for potential differences in age groups, age is examined as a potential moderator of the association between Intern et use and social networks.
3 Review of Literature Social Networks Social networks are linkages between pe ople that may provide social support (Heaney & Israel, 2002). Stru ctural network characteristics may explain support, access to jobs, social influence, and health behavi ors (Berkman & Glass, 2000). Whereas social networks are the structural aspects of social integration, social s upport is the functional aspect (Lubben & Gironda, 1996). Social networks consist of elements that describe the network as a whole, and characteristics of the social ties that comprise the network. Social network structures include size, reciprocity, complexity, de nsity, boundedness, homogeneity, reachability, and geographic dispersion (Berkman & Gla ss, 2000; Hall & Wellm an, 1985; Heaney & Israel, 2002; Lubben & Gironda, 1996). Char acteristics of network ties include frequency of contacts, frequenc y of organizational participation, reciprocity of ties, multiplexity, duration, and intimacy (Berkman & Glass, 2000; Hall & Wellman, 1985). Table 1, adapted from Berkman and Gl ass (2000), shows the linkages among social networks, social support, and healt h. Through this conceptual model, Berkman and Glass show how social networks provide opportunities for psychosocial mechanisms, such as social support and social influence, which in turn impact health through health behavioral, psychological, and physiologic pathways.
4 Table 1: Excerpts of the conceptual mo dels by Berkman & Glass (2000) of how social networks impact health Social Networks >> Psychosocial Mechanisms >> Pathways Social network structure: o Size o Range o Density o Boundedness o Proximity o Homogeneity o Reachability Social Support: o Instrumental & financial o Informational o Appraisal o Emotional Health behavioral pathways: o Smoking o Alcohol consumption o Diet o Exercise o Adherence to medical treatments o Help-seeking behavior Social influence: o Constraining/enabling influences on health behaviors o Norms toward helpseeking/adherence o Peer pressure o Social comparison process Psychological pathways: o Self-efficacy o Coping effectiveness o Depression/distress o Sense of well-being Characteristics of network ties: o Frequency of face-toface contact o Frequency of nonvisual contact o Frequency of organizational participation (attendance) o Reciprocity of ties o Multiplexity o Duration o Intimacy Which provide opportunities for Social Engagement o physical/cognitive exercise o Reinforcement of meaningful social roles o Bonding/ interpersonal attachment Which impact health through these Physiologic pathways: o HPA axis response o Allostatic load o Immune system function o Cardiovascular reactivity o Cardiopulmonary fitness o Transmission of infectious disease Theories on Social Networks and Health There are many theories that suggest str ong social networks have positive effects on health outcomes and psychological well -being. The stress-buffering hypothesis suggests that strong social ties reduce the susc eptibility of individuals to stress-related illnesses (Lubben & Gironda, 1996). Another th eory suggests that social isolation may have a physiologic effect on elders, such as impacting immune or cardiovascular functioning (Berkman, 1985; Seeman, 1996). Ye t another theory suggests that health promotion behaviors of the elderly can be a ffected by a strong social support network because of the encouragement that it can provide (Crawford, 1987). During times of illness, social networks provide support that contribute to better adaptation and
5 accelerated recovery time (Cohen & Syme, 1985). According to Seeman (1996) many studies suggest an association between social integration and mortality risk from all causes. Decreasing levels of social integr ation may be associated with increasing mortality risk (Seeman, 1996). Not only do social networks affect health, bu t health status affects ones ability to maintain a social network (Hean ey & Israel, 2002). In addi tion to affecting individual health, there is evidence that social networ k building within communities is associated with enhanced community capac ity (Heaney & Israel, 2002). The Internet and Social Networks The Internet is affecting the way peop le communicate and interact (Bargh & McKenna, 2004; Coget et al., 2002; Katz & Rice 2002). Some authorities argue that this communication transformation has an imp act on characteristics of social networks (Bargh & McKenna, 2004; Coget et al., 2002; Kr aut et al., 2002; Kraut et al., 1998; Nie & Hillygus, 2002). Studies show that Internet use may have a negative impact on social networks such as declines in face-to-face co mmunication with family and smaller social circles, damage to social interaction with fa mily members, or may lead to depression and isolation (Kraut et al., 1998; Nie & Hillygus, 2002). This phenomenon has been coined "the Internet paradox (Kraut et al., 1998) because people use the Internet for communication, and communication generally has a positive eff ect on social involvement. Other studies suggest Internet use does not increase social isolation, but is a source of civic organizational involvement, new pers onal friendships, and has positive effects on communication, social involvement, and wellbeing (Katz & Aspden, 1997; Kraut et al., 2002). The level of influence that Internet use has on a persons social network may
6 depend upon the quality of their In ternet relationships or what they give up to spend time online (Kraut et al., 2002). Older Adults, Social Networks, and the Internet Older adults, those over the age of 64, spend the major ity of their discretionary time at home, and they have more discretiona ry time than persons of younger age groups (Moss & Lawton, 1982). Moreover, they often re port problems with is olation, loneliness, and boredom (Neugarten, 1977). The reduction of social contact due to retirement, death of family members and friends, or residential relocation are, in part causes of decreasing social networks of older adults (Havens, Hall, Sylvestre, & Jivan, 2004; Pillemer & Glasgow, 2000). The disengagement theory sugg ests that older adults withdraw from social roles, but critics sugge st that disengagement may occu r due to lack of opportunity for a meaningful role (Pillemer & Glasgow, 2000). The evidence of decreasing social network strength suggests that older adults are at risk for becoming socially isolated. Pillemer and Glasgow (2000) argue that babyboomers may be at higher risk due to lowe r marriage rates, higher rates of divorce, and fewer offspring. In addition, the length of time an older adult may be without a meaningful role could be increasing with the current trends of earlier retirement, increasing longevity and improving hea lth status (Pillemer & Glasgow, 2000). Some authorities report that e-mail allows older adults to feel less isolated from their family, better informed about health i ssues, and more social ly connected (Malcolm et al., 2001). Similarly, older adults who participat e in online forums and use e-mail feel less isolated and more connected (Law hon, Ennis, & Lawhon, 1996). Retired older adults find roles in online forums by shari ng their knowledge and skills (Lawhon et al.,
7 1996). For older adults to take advantage of the Internet, th ey need to be aware of its capabilities and be able to develop a ba sic knowledge of comput ers. Interactive multimedia computer technology can be effective in teaching important information to older adults (Mercer, Chiri boga, & Sweeney, 1997). Research ers also suggest that older adults are capable and willing to learn how to use new computer technologies (Malcolm et al., 2001), and that most pa rticipants feel less anxious a nd more confident about using computer technology after training (Irizarry, Downing, & West, 2002).
8 Conceptual Framework and Hypothesis The literature demonstrates that a strong so cial network impacts health and health outcomes (Berkman, 1985; Cohen & Syme 1985; Crawford, 1987; Lubben & Gironda, 1996; Seeman, 1996). Social networks may become smaller as people grow older, providing fewer contacts that offer social support (Havens et al., 2004). Whereas some studies suggest that Internet use may increase the strength of the social networks by increasing the number of ties and frequency of contacts, other studies have shown that Internet use may have a nega tive effect on social networks by reducing the amount of time people spend on social act ivities (Bargh & McKenna, 2004; Coget et al., 2002; Katz & Aspden, 1997; Kraut et al., 2002; Kraut et al ., 1998; Nie & Hillygus, 2002). This thesis adds to the body of literature that report s on social networks by investigating the association between Internet use and social networks for middle ag ed and older adults. One theory that looks at Internet use and social networks contends that Internet use replaces high quality faceto-face ties with weaker onlin e ties (Coget et al., 2002; Kraut et al., 1998). One way to evaluate th is possible association is by measuring the number of confidants and the number of cl ose ties reported. A second theory suggests that Internet use displaces so cial activities in that time spent on the Internet is not available for other activities (Coget et al., 2002 ; Kraut et al., 1998; Nie & Hillygus, 2002). If this displacement phenomenon is the case, there should be a reduc tion of participation in organizations and face-to-face c ontacts for Internet users.
9 This study first investigates the null hypothe sis that, for U.S. adults 51 years and older, there is no difference between Intern et users and non-users with regard to the quality of their social ties. The alternative hypothesis is that, for U.S. adults 51 years of age and older, the quality of social ties for Internet users will differ from that of non-users of the Internet. If the null hypothesis is rejected, furthe r investigation may suggest that Internet use replaces high quality social ties with weaker ones. o Research Question 1 : Do Internet users report fewer close social ties (family or friends), confidants, or contacts with close ties than do non-users of the Internet? The second null hypothesis is that for U.S. adults 51 years and older, there is no difference between Internet users and non-user s with regard to frequency of in-person social contact. The alternative hypothesis is that, for U.S. adults 51 years and older, the frequency of in-person social contact for Inte rnet users differs from that of non-users of the Internet. If the null hypothesis is rejected, further investigation may show that Internet use does reduce time available for in-person social participation. o Research Question 2 : Do Internet users report less participation in organizations, attend religious services less often, or have fewer meetings with close friends and family th an do non-users of the Internet? The final null hypothesis in this study is that in the U.S., age does not moderate the association between Internet use and soci al networks for adults 51 and older. The alternative hypothesis is that in the U.S., the association be tween Internet use and social networks for adults 51 years and older is mode rated by age. Some studies suggest time spent on the Internet is time away from famil y, friends, and participating in organizations,
10 thereby reducing social integration (Kraut et al., 1998; Nie & H illygus, 2002). Those time restrictions may not be the case for olde r adults who have more discretionary time than younger adults (Moss & Lawton, 1982). o Research Question 3 : Does age moderate the association between Internet use and the number of close family, nu mber of close friends, contacts with close friends, contacts with fam ily, confidants, participation in organizations, or attendance of religious services? Characteristics commonly used in studies that look at network structure are the number of close friends and relatives, marital status, frequency of contact with family and friends, confidants, and frequency of attend ance at religious and voluntary associations, race, gender, income, education, age, and living situation, (Berkman & Glass, 2000; Oxman & Berkman, 1990; Seeman, 1996; Seem an & Berkman, 1988). This study uses similar measures to evaluate the stated hypotheses.
11 Methods Sample Data for this study were collected from the 2004 wave of the Health and Retirement Study a nationally representative, long itudinal study. The University of Michigan Health and Retirement Study (HRS ), supported by the National Institute on Aging (NIA U01AG009740), surveys more th an 22,000 Americans over the age of 50 in the contiguous United States every two years (Heeringa & Connor, 1996). The HRS studies the later life course and collects detailed information about the respondents demographic background, health, employment, family relations hips, income and wealth. The sample was selected under a multi-stage area probability sample design (Heeringa & Connor, 1996). The study design also include d supplemental oversamples of African Americans, Hispanics and residents of the state of Florida (Heer inga & Connor, 1996). The data collection period for the 2004 in terview was March 2004 through February 2005. Institutionalized persons (i.e., those in prisons, jails, nursing homes, long-term or dependent care facilities) were excluded fr om the survey populat ion (Heeringa & Connor, 1996). The HRS 2004 Core Final Release cont ains data for 20,129 respondents, in 13,645 households ( 2004 HRS final core code book 2006). Each wave of the HRS also includes addi tional modules that are asked only of a portion of the sample. The modules contain ques tions of interest for a specific research issue. The 2004 Psychosocial Leave-Be hind (PLB) module asked questions about
12 loneliness and social support. The PLB module was administered to a random sample of respondents who received interv iews in the 2004 survey ( HRS data description and usage 2006). The final PLB module data set contains 3,273 records ( 2004 HRS final core code book 2006). All of the measures for social ne tworking in this study were gathered from the PLB module. Data were also collected from the RA ND HRS Data file, version F. The RAND HRS file is a cleaned version of data from nine waves of the Health and Retirement Study data. Derived variables cove ring a broad range of measures have been constructed for this dataset. Version F incorpor ates data from 1992, 1993, 1994, 1995, 1996, 1998, 2000, and 2002 final releases, and the 2004 early release of HRS data ( RAND Contributions: RAND HRS Data File (v.F) 2006). The file was developed by the RAND Center for the Study of Aging with funding from the Nationa l Institute on Aging and Social Security Administration. The file incorpor ates only the core interviews. Data from the 2004 HRS PLB module data set were merged with data from the core 2004 HRS datasets, the 2004 Cross-Wave Tr acker file, and the RA ND HRS data file version F, using household ID and person ID numb ers to create a master data file for this study. The 2004 HRS Cross-Wave Tracker v.1. 0 (January 2007) contains variables (stratum, secu, jwgtr) used to account for the complex sampling design and oversampling used in the HRS. Only variables to be used in this study were included from each file in the merges. The master dataset file containe d only respondents that participated in the 2004 HRS Psychosocial Leave-Behind module a nd consisted of 3,273 cases in 56 strata and 111 clusters.
13 Measures Measures used in this study are detailed in appendix A. The dependent variables are measures of social networking characteri stics that are examined in theories that explain Internet effects on so cial networks. These variable s include measures of network size, confidants, frequency of contact w ith friends and family, and organization attendance. Internet use is th e independent variable of inte rest and is investigated for associations with char acteristics of social networks. Age is examined as a potential moderator of the association between Internet use and meas ures of social networking. Control variables include meas ures commonly used in invest igations of social network structures and Internet use including mar ital status, race, ethnicity, gender, income, education, living status, current employment status, occupa tion, number of children, and health status. The size of a social network is one of the most commonly used variables when looking at social networks (L itwin, 1996). Some studies argue that the quantity of contacts may be associated with risk of dying (Mullins., Elston, & Gutkowski, 1996). Other studies suggest that networ k size is not as important as quality of ties and that larger networks may bring increased demands and potential for damaging interactions (Stokes, 1983). Kraut et al. (1998) suggests that Internet us ers replace in-person contact with online virtual relationships and that online relationships are not as deep as off-line relationships. This supposed l ack of depth effectively reduces the quality of relationships while possibly increasing breadth. This study lo oks at the number of close relationships each respondent has by asking participants to estimate the number of close relationships they have with children, other family, and friends.
14 Some studies have suggested that having at least one confidant may be the most important indicator of a supportive social network (Stokes, 1983). However, the literature on the Internet and confidants is mixed. There are st udies that suggest relationships created or primarily maintained online are lower quality than those maintained by other means (Cummings, Lee, & Kraut, 2006). Other studies argue online relationships can be as strong as thos e developed by other means given enough time (Cummings et al., 2006). This study looks at confidants reported by respondents. Respondents were asked how much they can open up to and talk about your worries with their spouse/partner, children, other fam ily, and friends. Responses are measured on a four-point scale from not at all" to "a lot." There is debate in the literature whether or not Internet use in creases contact with family and friends, or reduces the time av ailable for such contacts (Coget et al., 2002; Kraut et al., 1998; Nie & Hillygus 2002). Studies are also mi xed about whether Internet use increases or decreases the number of in-per son contacts with family and friends (Katz & Rice, 2002; Kraut et al., 1998; Nie & H illygus, 2002). Frequency of contact is examined in this study by asking respondents how often they have contact with family and friends either in-person, by phone, or by mail or email. All forms of contact are measured on a six-point scale ranging from "le ss than once a year or never" to "three or more times a week." In-person contact is also examined separately by looking specifically at how often respondents repor ted meeting with family and friends. Participation in social groups is m easured by studying religious services attendance and participation in meetings ot her than religious services. One theory suggests participation in social groups is being replaced by time spent online (Kraut et al.,
15 1998; Nie & Hillygus, 2002). Some studies sugg est Internet users are more likely to belong to religious, leisure, and commun ity organizations (Bargh & McKenna, 2004; Katz, Rice, & Aspden, 2001). This study looks at attendance of religious services measured on a five-point scale ranging from not at all" to "more than once a week." Attendance in leisure or community meetings (not including attendance of religious services) is examined using a six-point scal e ranging from "never" to "more than once a week." Internet use is a dichotom ous variable in the datase t and age is a continuous variable. Internet use is de fined in the HRS survey as regular use of the World Wide Web, or the Internet, for sendi ng and receiving e-mail or for any other purpose, such as making purchases, searching for information, or making travel reservations. Age is the reported age of the respondent at the time of the interview. Respondents range from 51 to 101 years of age. The association be tween Internet use and social networking measures may vary at different ages, thus ag e is explored as a potential moderator. A number of control measures are included in this study based on controls used in other studies on Internet use and social ne tworks. Control measures include marital status, race, ethnicity, gender, income, education, liv ing status, current employment status, occupation, having children, and he alth status. These variable s were selected as control measures because it has been suggested they may be related to Internet use and the dependent variables (Coget et al., 2002; Fox, 2006; Glass, F., Seeman, & Berkman, 1997; Levy et al., 2000; Nie & Hillygus, 2002).
16 Marital status is a dichotomous variable with one value indi cating the respondent is currently married or partnered and the other not married. Race has three values, Caucasian, African American, or other. Ethnicity is a dichot omous variable with a value for Hispanic and the other for non-Hispanic. Gender is given by the respondent at the time of the interview and is a dichotomous variable with values for male or female. Income is a variable that is the total household income in dollars for the respondent and spouse and is calculated in the Rand datase t using a number of variables from the 2004 HRS core data. Education is an interval leve l variable that holds the number of years of education the respondent has completed, w ith 17 being the highest possible value, representing graduate education level. Living status is an interval level variable that tracks the number of people living in the res pondent's home. A variable to track if the respondent had any children was dichotomous with values for yes and no. Current employment status is dichot omous with values for currently working and not currently working. Occupation of longest tenure was coded in to three categories. White and blue collar occupation categories were coded using the same method as other studies using the HRS dataset and the occupation variable (Bovbjerg, 1998; Wu & Prorell, 2000). Managerial, professional specialty/technical support, sales, and clerical/administrative support occupations were classified as wh ite collar. Cleanin g/building services, protection services, food prep aration services, health services, personal services, farming/forestry/fishing, mechanics/repair construction trade, precision production, machine operators, transport operators and handl ers occupations were classified as blue
17 collar. Respondents in the armed forces or missing data for occupation were categorized as unknown. It has been suggested that social ne tworks and health have a reciprocal relationship. Supportive ties may enhance well -being and health, and health status may influence the size and strength of a social ne twork one is able to maintain (Heaney & Israel, 2002). Health status may also aff ect Internet use. Bargh and McKenna (2004) discuss the possibility that those who suffer fr om a stigmatized illness or lack of mobility may be especially likely to turn to the Inte rnet. Two self-report m easures were selected for health status. Health status is reported on a five-point scale ra nging from excellent to poor. Health barriers measure how often hea lth stops the respondent from doing things they want to do, measured on a four-poi nt scale from often to never. Statistical Analysis This analysis examines the dataset constructed from the HRS data for associations between Internet use and measures of social networ king, while controlling for certain demographic, health, and occupation variable s. The potential for moderation by age is also examined. All of the analysis was r un on SAS version 9.1. Variab les from the tracker file were used to account for the complex sa mpling design and oversampling including a respondent level weight variable (jwgtr), strata variable (s tratum), and the stratum halfsample code variable (secu). The stratum and secu variables used in conjunction with specialized SAS survey procedures to acc ount for the complex random sample design effects of the HRS sample. Sample weights were normalized (relative) to the size of the dataset used. For analysis of the PLB samp le of the 2004 HRS, weights were normalized
18 to the 3,273 cases in that datase t. For analysis of the final dataset, weights were normalized to the 2,284 cases retained. SAS has no tools to adjust weights based on reduction of sample size (due to excluded cases). Before the final sample for study was cr eated, some variables were manipulated to reduce missing data. Data that were mi ssing from the PLB module dataset were crosschecked with other variables from within th at dataset as well as from the HRS Core, Tracker, and RAND data. The missing data in the PLB dataset were replaced if valid data were found in other data sets. As an example, if a case in the PLB dataset was missing data for number of children, and a nother dataset had data that stated the respondent had no children, then the number of children variable was updated from missing to zero. Some variables were reve rse coded to facilitate interpretation. Appendix A details modifications to data. To decrease the number of social networki ng variables, it was proposed that some variables be combined. However, due to weak Cronbach values, the only variables that were combined were frequency of contact m easures. The two new variables created were contact with family ( = 0.69) and contacts with friends ( = 0.71). The forms of contacts combined were in-person meeti ngs, contact by phone, and contact by mail or email. There is some evidence that Internet use may affect interactions with close family differently than with friends (Cummings et al., 2006). In addition, some studies suggest friends and family may provide different levels of support (Fiori, Toni, & Cortina, 2006). Litwin reports that older adults generally have more frequent contact with family members than friends (Litwin, 1996). For these reasons, measures of family are separated from measures of friends.
19 Univariate analysis revealed some variab les with large kurtosis values. These variables all had a few cases with values fa r greater than their m ean. To reduce their kurtosis value and bring them to a more normal distribution their values were truncated at their 95th percentile. Variables truncated were number of close other family (95th percentile =12), number of close friends (95th percentile =14), household residents (95th percentile =4), household income (95th percentile =187500). Because there were an unusual number of cases missing data for "occupation" (359 weighted cases), missing data were recoded to "unknown." Most cas es in the new category of unknown had a current job status of retired, homemak er, or were in the Armed Forces. To determine correlations between interval level variables and other interval or dichotomous variables, the surveyreg procedure in SAS was use d. The correlations were calculated using the square root of the R-squa red value of simple regression models (one dependent, one independent variable) resulti ng in Spearman rank correlations for interval level independent variables, and point-biserial correlati on for dichotomous independent variables. Where the independent variable was a categorical va riable with more than two categories, dummy coded variab les were used in the model and the Eta was derived. To determine bivariate associations between categorical variables, the SAS procedure surveyfreq was used to determine the chi-square value. Cramers V was calculated from the results. SAS procedures used were capab le of using variables included in the HRS dataset (stratum, secu, jwgtr) to take in to account the complex sampling design and oversampling used in the HRS sample. Cases missing data in variables to be used in the bivariate an d regression tests or with non-positive weights were then dropped from the dataset. Those cases retained were
20 compared to those dropped. Seventy percent ( n =2284) of the cases from the PLB dataset were retained. There was no signifi cant difference in work status 2(1, N =3273) = 3.02, p = .232, or gender, 2(1, N =3273) = 3.718, p =.0547 between the two sets. Measures with statistically significant differences, but small effect sizes, between retained and dropped cases were race, 2(2, N =3269) = 29.509, p < .0001, marriage status, 2(1, N =3273) = 7.10, p =.0350, have children, 2(1, N =3261) = 22.03, p =.0001, Hispanic, 2(1, N =3272) = 43.94, p <.0001, Internet use, 2(1, N =3155) = 60.34, p <.0001, age, rpb (56)=.1178, p <.0001, household residents, rpb(56)=.0629, p <.001, household income, rpb (56)=.0805, p <.0001, health status, rpb (56)=.1117, p <.0001, and health barriers, rpb (56)=.0948, p <.0001. Measures with statistically significan t differences, and moderate effect sizes, between retained and dropped cases were occupation and edu cation. Retained cases were more likely to have had a white collar occupation (54.6%) than dropped cases (34.9%), 2(2, N =3273) = 80.40, p <.0001, and more years of education ( M =13.21, SE =.111) than dropped cases ( M =12.02, SE =.215), rpb (56)=.1740, p <.0001. There were 2,284 cases, 56 strata and 110 clusters in the final sample of retained cases. Bivariate analyses were run on the reta ined cases. Where an association was found between Internet use and measures of so cial networking in th e bivariate analysis, further examination was done with linear regr ession. The first set of regression models examined research question one which examines potential associations between Internet use and close ties. Regression models were built for the dependent variables number of close children, number of close other family, number of close friends, children confidants, other family confidants, friend confidants, cont act with family, and contact with friends. Each model contained the independent variable of interest, Internet use, and controls for
21 race, gender, Hispanic status, marriage stat us, currently working status occupation of longest tenure, whether the respondent had ch ildren, age, number of household residents, annual income, years of education, health barr iers, and health status. The Bonferroni method to control for Type-I errors adjust ed the alpha level to .00625 (desired alpha level .05/number of models 8). The next set of regression models examined research question two which examines attendance at organizations and in-p erson meetings with cl ose ties. Regression models were built for the dependent variables in-person meetings with children, in-person meetings with other famil y, in-person meetings with friends, and attendance of organizations or clubs other than religious services. Th ese models also included the Internet use measure and all of the control vari ables. The Bonferroni adjusted alpha level became .0125. Models were examined to make sure the assumptions of regression modeling were met and diagnostics were performed to de tect colinearity and outliers. Outliers are defined in this study as any case with a Cooks D value greater than one (Stevens, 1996). Using this criterion, there were no out liers and no colinearity was found. During examination of the assumptions of regression a violation of the assumption of equal variance of errors was detected. To preserve the assumption of equal variance of errors, the control variable 'have children' was removed on models where the dependent variable regarded children. The final regression tests looked for m oderation of the a ssociation between Internet use and social networking measures by age. Only full models that were statistically significant and had a statistically significant Internet use parameter were
22 tested for moderation by age. An interacti on term of Internet use by age was added to each full model that met the requirements and the statistical significance of the interaction term was examined. Because four models were tested, the Bonferroni method to control for Type I error adju sted the alpha level to .0125.
23 Results Sample Demographics The sample is described in detail in table 2. The cleaned dataset ( N =2284) consisted of all cases with no missing data as described in the methods. Approximately half of the respondents were Internet user s (49.8%). Most res pondents were Caucasian (91.5%), married (72.7%), had children (87.5%), and slightly more than half were women (55.4%). Less than half worked at the tim e of the interviews (44.5%) and most had a white collar occupation (54.6%). The mean age was 63 with an average of two people living in a household and an average annual income of $62,0 00. The mean of the selfrated health of respondents was 3.28 and the mean score for health barriers was 2.14. Respondents had contact with family ( M =3.56, SE =0.02) about as frequently as with friends ( M =3.76, SE =0.03). Respondents met in person most often with friends ( M =4.16, SE =0.04), a bit less frequently with children ( M =3.81, SE =0.06), and much less frequently with other family members ( M =3.36, SE =0.04). They reported having more close friends ( M =4.41, SE =0.11) than close other family members ( M =3.70, SE =0.10) and twice as many close friends as close children ( M =2.18, SE =0.06). However, when asked who they can open up to, children ( M =2.93, SE =0.04) and friends ( M =2.91, SE =0.03) were rated equally high, with other family ( M =2.80, SE =0.03) rated a little lower. Respondents attended religious services about two or three times a month
24 ( M =2.80, SE =0.04) and attended other meetings, clubs or organizations they belong to about once a month ( M =2.62, SE =0.05). Table 2: Sample description ( N =2284) Categorical Variables Interval Variables Variable % Variable Range M SE Internet use (yes) 49.80% Age 51-101 63.27 0.45 Race Household residents 1-4 2.14 0.03 (Caucasian) 91.50% Income (in thousands) 0-187.5 62.16 2.17 (African American) 4.20% Education 0-17 13.21 0.11 (Other) 4.30% Health barriers 1-4 2.14 0.03 Gender (Female) 55.40% Health status 1-5 3.28 0.04 Hispanic (Yes) 5.60% Contacts w/family 1-6 3.56 0.02 Married (Yes) 72.70% Contacts w/friends 1-6 3.76 0.03 Have children (Yes) 87.50% In-person w/child 1-6 3.81 0.06 Working now (Yes) 44.50% In-person w/other family 1-6 3.36 0.04 Occupation In-person w/friends 1-6 4.16 0.04 (Blue Collar) 31.50% Child confidant 1-4 2.93 0.04 (White collar) 54.60% Other family confidant 1-4 2.8 0.03 (Unknown) 13.90% Friend confidant 1-4 2.91 0.03 Number close children 0-13 2.18 0.06 Number close other family 0-12 3.7 0.1 Number close friends 0-14 4.41 0.11 Attend meetings 1-6 2.62 0.05 Attend religious services 1-5 2.8 0.04 Bivariate Analysis Results The bivariate results between the primary independent variable, Internet use, and all other variables are shown in table 3. Ther e were no statistically significant differences between Internet users and non-us ers with respect to gender, 2(1,2282) = .07, p =.4285. A significantly higher percentage of Internet users were Caucasian 2(2, 2282) = 25.3, p < .001, married, 2(1,2515) = 67.2, p < .0001, working at the time of the interview, 2(1, 2282) = 190.2, p < .0001, and had a white collar occupa tion as the occupation of longest tenure, 2 (2, 2282) = 205.2, p < .0001, when compared to non-users of the Internet. A significantly lower percentage of Internet users were Hispanic 2(1, 2282) = 21.5, p < .0001, and had children, 2(1, 2282) = 9.2, p < .05, when compared to non-users of the
25 Internet. Internet users were younger, rpb (55) =-.323, p <.0001, than non-users. Of those under 65 years of age, 62% were Internet us ers whereas only 33% of those 65 and over were Internet users, 2 (1, 2282) = 178.0, p < .0001. Internet users had more years of education, rpb (55) =.399, p <.0001, a higher annual income, rpb (55) =.358, p <.0001, and more household residents, rpb (55) =.112, p <.0001, than non-users. Internet users rated their health status higher, rpb (55) =.201, p <.0001, and had fewer health barriers, rpb (55) =-.146, p <.0001 than non-users. Internet use was significantly associated with all measures of social networking except attendance of religious services, rpb(55) =.000, ns, and the number of close friends, rpb(55) =-.039, p =.0624. Internet users have more contacts of all kinds with family, rpb(55) 089, p <.0001, and friends, rpb(55) =.237, p <.0001. Internet users reported more in-person contact with friends, rpb(55) =.052, p <.001, but fewer in-per son contact with close children, rpb(55) =-.076, p <.001, and close other family, rpb(55) =-.117, p <.0001, than non-users. Internet users felt they could open up more to friends, rpb(55) =.051, p <.05, but could not open up as much to children, rpb(55) =-.061, p <.001, or other family members, rpb(55) =-.042, p <.05, as non-users could. Intern et users reported fewer close children, rpb(55) =-.090, p <.0001, and close other family members, rpb(55) =-.093, p <.0001, than non-users of the Internet. Intern et users reported attending more meetings, clubs, and organizations (not including religious services), rpb(55) =.107, p <.0001, than non-users of the Internet. Appendix B contains details of the biva riate results between each of the variables used in this study.
26 Table 3: Comparison of Internet users and non-users Internet Users Non Users Cases 1136 (49.75%) 1148 (50.25%) Variable % % 2 V Internet use (yes) Race (Caucasian) 94.5% 88.6% (African American) 2.7% 5.8% (Other) 2.8 5.7% 25.348** 0.100 Gender (Female) 54.5% 56.2% 0.699 0.017 Hispanic (Yes) 3.3% 7.8% 21.463*** 0.092 Married (Yes) 80.4% 65.1% 67.247*** 0.163 Have children (Yes) 85.4% 89.6% 9.168* 0.060 Working now (Yes) 58.9% 30.3% 190.181*** 0.275 Occupation (Blue Collar) 21.8% 41.0% (White collar) 69.5% 39.8% (Unknown) 8.6% 19.2% 205.190*** 0.286 M SE M SE rpb Age 60.07 0.45 66.43 0.46 -0.323*** Household residents 2.23 0.04 2.05 0.04 0.112*** Income (in thousands) 80.55 2.69 43.95 1.88 0.358*** Education 14.29 0.08 12.13 0.15 0.399*** Health barriers 1.99 0.03 2.28 0.04 -0.146*** Health status 3.51 0.04 3.05 0.05 0.201*** Contacts w/family 3.66 0.04 3.47 0.04 0.089*** Contacts w/friends 4.05 0.04 3.47 0.05 0.237*** In-person w/child 3.68 0.08 3.95 0.07 -0.076** In-person w/other family 3.18 0.05 3.55 0.05 -0.117*** In-person w/friends 4.24 0.05 4.08 0.06 0.052* Child confidant 2.85 0.05 3.00 0.04 -0.061** Other family confidant 2.75 0.04 2.85 0.04 -0.042* Friend confidant 2.97 0.04 2.86 0.04 0.051* Number close children 2.03 0.06 2.33 0.09 -0.090*** Number close other family 3.41 0.09 3.99 0.15 -0.093*** Number close friends 4.25 0.15 4.56 0.17 -0.039 Attend meetings 2.81 0.07 2.44 0.05 0.107*** Attend religious services 2.80 0.05 2.79 0.06 0.000 SE=Standard error of the mean p<.05, **p<.001, ***p<.0001
27 Regression Results The first regression models were built to answer research question one which asked if Internet uses report fewer close social ties, confidants, or c ontacts with close ties than do non-users of the Internet A Bonferroni adjusted al pha level of .0071 was used to determine statistical significance for these m odels. The models for the social networking dependent variables of number of close ch ildren, number of close other family, children confidants, other family confid ants, and friend c onfidants were all statistically significant, but the independent variable of primar y interest, Inte rnet use, was not statistically significant, as shown in table 4. For these models, the bivariate association between Internet use and the m easures of social ne tworking did not hold when considering the control variables. Table 4: Regression models for research que stion one: Internet use associations with social networking measures Models (strata=56, strata collapsed=2, clusters=110) Internet Use Parameter Dependent Variable Model F* Adj. R2 b SE t Pr>|t| Number close children (15, 55) = 26. 19 0.1340-0.0202 0.0799 -0.25 0.8009 Number close other family (16, 55) = 11.88 0.0450-0.2808 0.1710 -1.64 0.1063 Children confidants (15, 55) = 19. 84 0.1095-0.0552 0.0596 -0.93 0.3585 Other family confidants (16, 55) = 11.34 0.0263-0.0623 0.0562 -1.11 0.2730 Friend confidants (16, 55) = 11.07 0.07280.0135 0.0618 0.22 0.8282 Contact w/family (15, 55) = 18. 71 0.09430.2344 0.0592 3.96 0.0002 Reduced Model (14, 55) = 15.77 0.0854 R2 Difference (1,2268) = 22.14 0.0089 Contacts w/friends (16, 55) = 16. 83 0.13690.5285 0.0638 8.28 <.0001 Reduced Model (15, 55) = 12.01 0.1034 R2 Difference (1,2267) = 87.99 0.0335 Note: Models include Internet use and controls for race, gender, Hispanic status, marriage status, currently working status occupatio n, whether the respondent has children, age, number of household residents, annual income, years of education, health barriers, health status, except were noted. The Bonferroni adjusted alpha level was .0071. = Variable 'have-children' was not included in these models. = p<.0001
28 However, the model for contacts with family, adjusted R2 = 0.09427, F (15,55) = 18.71, p < .0001, and the Internet use variable in the model, t = 3.96, p = 0.0002, were both statistically significant. An R2 difference test revealed that Internet use accounted for a statistically significant and unique amount of the vari ation in the contacts with family measure, R2 = 0.0089, F (1,2268) = 22.14, p < .0001. Other parameters that accounted for a statistically significant amount of variation in the contacts with family measure were gender, b = .4755, p < .0001, age, b = .0127, p < .00625, and the number of household residents, b = .1481, p < .00625. All the parameters for the model are shown in table 5. Table 5: Regression model parameters fo r research question one where Internet use is significant Contacts w/Friends Model Contacts w/Family Model Model F (16, 55) = 16.83 ** Adj. R2 = 0.1369 Model F (15, 55) = 18.71 ** Adj. R2 = 0.0943 Parameter Estimate SE t Estimate SE t Intercept 2.0981 0.4577 4. 584** 2.0328 0.3465 5.866** Internet Use 0.5285 0.0638 8.283** 0.2344 0.0592 3.961* Race African Am. -0.1020 0.186 9 -0.546 -0.1670 0.1257 -1.328 Race Other -0.1065 0.2238 -0 .476 -0.0448 0.1316 -0.340 Gender 0.4811 0.0626 7.687* 0.4755 0.0589 8.076** Hispanic -0.2250 0.1605 -1. 402 -0.0516 0.1151 -0.449 Married -0.1184 0.0775 -1 .528 0.1495 0.0703 2.125 Working Now -0.1691 0.0832 -2 .032 -0.0416 0.0659 -0.631 Occupation Blue Collar 0.0039 0.0807 0.049 0.0575 0.0566 1.017 Occupation Other -0. 0940 0.0857 -1.096 0.0107 0.0624 0.171 Have Children -0.1052 0.1166 -0.903 Age 0.0129 0.0042 3.053* 0.0127 0.0032 3.932* Household Residents -0.0624 0. 0495 -1.259 0.1481 0.0426 3.477* Household Income 0.0021 0.000 7 3.025* 0.0011 0.0007 1.540 Education 0.0364 0.0171 2. 121 -0.0271 0.0121 -2.242 Health Barriers -0.0366 0.032 7 -1.117 0.0045 0.0345 0.132 Health Status 0.0617 0.030 1 2.051 0.0631 0.0311 2.032 Note: Models include Internet use and controls for race, gender, Hispanic status, marriage status, currently working status occupation, whether the respondent has children, age, number of household residents, annual income, years of educat ion, health barriers, health status, except where noted. Bonferroni adjusted alpha was .0071. = Variable 'have-children' was not included in these models. p<.0071 ** p<.0001
29 The model for contacts with friends, adjusted R2 = 0.1369, F (16,55) = 16.83, p < .0001 and the Internet use variable in the model, t = 8.28, p < .0001, were also both statistically significant. An R2 difference test revealed that Internet use accounted for a statistically significant and uni que amount of the variation in the contacts with friends measure, R2 = 0.0335, F (1, 2267) = 87.99 p < .0001. Other parameters that accounted for a statistically significant amount of variat ion in the contacts with friends measure were gender, b = .4811, p < .0001, age, b = .0129, p < .00625, and household income, b = .0021, p < .00625. All the parameters for the model are shown in table 5. For the contacts with friends and contact with fa mily measures of social networking, the association with Internet use did hold even when considering the control variables. The next regression models were built to answer research question two which asked if Internet users report less participation in organizatio ns, attend religious services less often, or have fewer meetings with clos e ties than non-users of the Internet. A Bonferroni adjusted alpha leve l of .0125 was used to determin e statistical significance for these tests. The models for the social netw orking measures of in -person meetings with close children and in-person meetings with close friends were significant, but the independent variable of primary interest, Internet use, was not statistically significant, as shown in table 6. For those models, the biva riate association betw een Internet use and the measures of social networking did not hold when considering the control variables.
30 Table 6: Regression models for research que stion two: Internet use associations with social networking measures Models (strata=56, strata collapsed=2, clusters=110) Internet Use Parameter Dependent Variable Model F Adj. R2 b SE t Pr>|t| In-person meetings w/children (15, 55) = 10.27 ** 0.0624-0.2294 0.098 6 -2.33 0.0237 In-person meetings w/friends (16, 55) = 8.74 ** 0.05530.0575 0. 0708 0.81 0.4202 In-person meetings w/other family (16, 55) = 7.61 ** 0.0407-0.3025 0. 0889 -3.40 0.0013 Reduced Model (15, 55) = 5.98 ** 0.0342 R2 Difference (1, 2267) = 15.22 ** 0.0064 Attend meetings (16, 55) = 13.02 ** 0.08460.2704 0.0813 3.33 0.0016 Reduced Model (15, 55) = 12.75 ** 0.0802 R2 Difference (1, 2267) = 10.95 0.0044 Note: Models include Internet use and controls for race, gender, Hispanic status, marriage status, currently working status occupatio n, whether the respondent has children, age, number of household residents, annual income, years of education, health barriers, health status, except were noted. The Bonferroni adjusted alpha level is .0125. = Variable 'have-children' was not included in these models. p<.001 ** p<.0001 The model for in-person meetings w ith close other family, adjusted R2 = 0.0407, F (16,55) = 7.61, p < .0001, and the Internet use va riable in that model, t = -3.4, p = .0013, were both statistically significant. An R2 difference test revealed that Internet use accounted for a statistically significant and unique amount of the va riation in the inperson meetings with close other family measure, R2 = 0.0064, F (1, 2284) = 15.22, p < .0001. Other parameters that accounted for a statistically significant amount of variation in the in-person meetings with close other family members measure were gender, b =.3137, p < .0001, having children, b =.3753, p < .0125, and education, b = .0466, p < .0125. All the parameters for the model are shown in table 7.
31 Table 7: Regression model parameters fo r research question two where Internet use is significant In-person w/Other Family Model Meeting Attendance Model Model F (16, 55) = 7.61 ** Adj. R2 = 0.0407 Model F (16, 55) = 13.02 ** Adj. R2 = 0.0846 Parameter Estimate SE t Estimate SE t Intercept 3.8953 0.4110 9. 477** -0.4648 0.5569 -0.835 Internet Use -0.3025 0.0889 -3.401* 0.2704 0.0813 3.326* Race African Am. -0.2363 0. 1878 -1.258 0.7716 0.1900 4.061* Race Other -0.3655 0.171 3 -2.134 0.1599 0.2304 0.694 Gender 0.3137 0.0704 4. 454** 0.1633 0.0993 1.645 Hispanic -0.0932 0.1763 -0.528 0.2975 0.2072 1.436 Married -0.0248 0.1096 -0. 226 -0.0509 0.1266 -0.402 Working Now 0.0033 0.0892 0. 037 -0.3306 0.0926 -3.571* Occupation Blue Collar 0.1787 0.0737 2.423 -0.126 1 0.0935 -1.349 Occupation Other -0. 1331 0.1272 -1.047 -0 .1944 0.1240 -1.568 Have Children 0.3753 0.121 0 3.102* 0.1641 0.1132 1.449 Age -0.0066 0.0050 -1. 326 0.0229 0.0049 4.647** Household Residents -0.0010 0. 0628 -0.015 -0.0592 0.0566 -1.046 Household Income 0.0001 0. 0011 -0.004 0.0023 0.0009 2.421 Education -0.0466 0.0151 -3 .086* 0.1028 0.0184 5.591** Health Barriers 0.0128 0.045 2 0.283 -0.0684 0.0401 -1.704 Health Status 0.0408 0. 0407 1.003 0.0729 0.0454 1.606 Note: Models include Internet use and controls for race, gender, Hispanic status, marriage status, currently working status occupation, whether the respondent has children, age, number of household residents, annual income, years of educat ion, health barriers, health status, except where noted. Bonferroni adjusted Alpha was .0125 p<.0125 ** p<.0001 The model for attend meetings and organi zations other than religious services, adjusted R2 = 0.0846, F (16,55) = 13.02, p < .0001, and the Internet us e variable in that model, t =3.33, p =.0016, were also both statis tically significant. An R2 difference test showed that Internet use account ed for a statistically significant and unique amount of the variation in the attend meetings measure, R2 = .0044, F (1, 2284) = 10.95, p < .001. Other parameters that accounted for a statistically significant amount of va riation in the attend meetings measure were race African American, b = .7716, p < .0125, working now, b = 0.3306, p < .0125, age, b = .0229, p < .0001, and education, b = .1028, p < .0001. All the parameters for the model are shown in table 7. For in-person meetings with close other
32 family members and attend meetings, the asso ciation with Internet use did hold even when considering the control variables. The last set of regression models were built to answer rese arch question three which asked if age moderates any associati on Internet use has on measures of social networks. Four models were built, one for each measure of social networking that was associated with Internet use in the full regr ession models. As shown in table 8, with a Bonferroni adjusted alpha of .0125, none of the interaction terms were statistically significant. Age does not moderate any of the associations found between Internet use and measures of social networking while consid ering the control variables in this sample. The hypothesis that there is no differen ce between Internet users and non-users with regard to their quality of social tie s is not supported based on the finding that Internet users have more frequent contact wi th friends and family than non-users. The hypothesis that there is no diffe rence between Internet users and non-users with regard to the frequency of in-person social contact for a dults 51 years and older in the U.S. is also not supported by these data. Internet users have fewer in-person contacts with close family members not including children a nd they participate more in clubs and organizational meetings, excluding religious se rvices. The hypothesi s that there is no difference in the association between Internet use and social networks based on age for adults 51 years and older in the U.S. is suppor ted by the results of th is study as no tests for moderation were statistically significant.
33 Table 8: Tests for interaction between age and Internet use Model (strata=56, strata collapsed=2, clusters=110) Interaction Term Dependent Variable F p R2 b SE t p Contact w/family (16, 55) = 19.29 <.0001 0. 09420.0043 0.0064 0.67 0.5077 Contacts w/friends (17, 55) = 118.51 <.0001 0.136 6-0.0019 0.0071 -0.27 0.7897 In-person w/other family (17, 55) = 17.21 <.0001 0.040 2-0.0014 0.0075 -0.19 0.8488 In-person w/friends (17, 55) = 19.47 <.0001 0. 05500.0027 0.0092 0.30 0.7674 Note: Models include Internet use and controls for race, gender, Hispanic status, marriage status, currently working status occupatio n, whether the respondent has ch ildren, age, number of household residents, annual income, years of education, health barriers, health status, except were noted. The Bonferroni adjusted alpha level is .0125. = Variable 'have-children' was not included in these models.
34 Discussion Conclusions This study investigates the pot ential for the Internet to be used as a tool to strengthen the social networks of older adults by examining the association between Internet use and social networ ks for middle aged and older. Part of the investigation utilizes Bonferroni adjustments to reduce the possibility of a type I error. Using the Bonferroni adjustment also increases the chan ce of a type II error being committed. This could affect some of the results in this thesis by not finding significance where tests actually are significant. However, this st udy examines three null hypotheses, and each hypothesis is tested with many regression models, and since only one of the tests need be significant to reject a null hypothesis, using a co nservative approach to control for type I error is appropriate. Internet users in this study were similar to those of other stud ies on Internet use, with Internet users tending to be younger, have more year s of education, and a higher household income than non-users of the Internet In this study, cu rrent employment and white collar occupations were also associated with Internet use. All of these factors support the idea that Internet us e is associated with higher SE S. Despite a broadening of Internet use (Katz & Aspden, 1997), the results of this study show evidence of a digital divide. These results suggest th at cost and availability of th e Internet may be a factor in any program that considers usi ng the Internet as a tool.
35 The first null hypothesis examined states th at for U.S. adults 51 years and older, there is no difference between Internet users and non-users with rega rd to their quality social ties. One theory that looks at Internet use and so cial networks contends that Internet use replaces high qua lity face-to-face ties with w eaker online ties (Coget et al., 2002; Kraut et al., 1998). Wher eas this study found no differen ces in the number of close relationships or confidants between Internet users and non-users, a st atistically significant difference in frequency of contacts with friends and family was found even when considering the control variable s used in this study. This study rejects the first null hypothesis in favor of the alternative that for U.S. adults 51 years of age and older, the quality of social ties for inte rnet users differ from that of non-users of the Internet. Berkman and Glass (2000) report that frequenc y of contact is one measure of the quality of social ties. This study shows a positive association between Internet use and frequency of contact with friends and famil y. Thus, the Internet may be a tool which could be used to strengthen this meas ure of the quality of social ties. The second null hypothesis examined is that for U.S. adults 51 years and older, there is no difference between Internet users and non-users with rega rd to frequency of in-person social contact. Tw o of the five measures of in -person social contact were found to be statistically signifi cantly different for Internet us ers and non-users. Internet users met in-person fewer times with close ot her family members than did Internet nonusers. It is not clear, however, if this finding supports the displacement theory (Kraut et al., 1998; Nie & Hillygus, 2002), or if res pondents meet less often with this group because of other factors such as lack of mobility or physical distance. Internet users are more likely to participate in organizations or clubs (excluding relig ious services) than
36 Internet non-users. This finding is consistent with other studi es that suggest the Internet is a positive influence on community involvement (Bargh & McKenna, 2004). Though Internet users in this sample were more healthy and younger than Internet non-users, those factors were considered in the regres sion model suggesting Internet use increases community participation even when those fact ors are considered. Th is study rejects the second null hypothesis in favor of the alternativ e that, for U.S. adults 51 years and older, there are differences between In ternet users and non-users with regard to frequency of inperson social contact. The differenc e suggests a positive impact on community involvement, but is inconclusive on in-person family contacts. The hypothesis that there is no difference in the association be tween Internet use and social networks based on age for adults 51 years and older in the U.S. is supported by the results of this study, as no tests for mode ration were statistically significant. Though older adults do have more discretionary time than younger adults (Moss & Lawton, 1982), it does not appear to impact any association between Intern et use and social networks. Results of this thesis support the idea th at Internet use may improve frequency of contact with family and friends. Results also show a positive association between Internet use and in-person community i nvolvement for middle-aged and older adults in the United States. With the findings from the literatu re review that social networks play an important part in human well being, it is importa nt to utilize available tools to counter the weakening of social networks that face olde r adults. As the Internet evolves, it is becoming more capable of supporting human soci al interactivity, as shown by the growth of web sites dedicated to social interaction. This study finds that the Internet could be used as a tool to strengthen soci al networks of older adults.
37 Limitations This study uses a cross-sectional desi gn, and as such, cannot determine if differences in characteristics of social networ ks are determined by Inte rnet use, or if the use of the Internet is caused by differences in characteristics of social networks. Internet use is a dichotomous measure in this st udy and therefore the am ount of time spent on Internet use is not known. Nie and Hillygus (2002) suggest that possible effects of Internet use will be concealed if time spent on th e Internet is not considered in analysis of Internet use. In addi tion, the use of the phrase "regular use of the World Wide Web" in the survey question is a subjective measure of internet use, so the reported use of the Internet may not be equa l between respondents. SAS 9.1 is not capable of adjusting the we ights of cases based on the reduction of sample size (due to missing data). Weights were designed for the full 2004 wave of the Health and Retirement Survey. Though they were normalized according to size of the sample used in this study, the weights them selves were not adjusted. This fact may reduce the generalizability to the population from which the sample was taken. There were many statistically significant differences between cases retained for this study and cases dropped due to missing da ta. The sample is large so it is not surprising to find statistical significance. Most of the differences have small effect sizes ( V <=0.1, r <=0.1). However, significantly different measures with the largest effect sizes show retained cases were more likely to be Internet user, 2(1, N =3155) = 60.34, p <.0001, V = 0.1427, be younger, rpb (56)=.1178, p <.0001, have had a white collar occupation, 2(2, N =3273) = 80.40, p <.0001, V = 0.1693, and have more years of education, rpb (56)=.1740, p <.0001, than dropped cases. Thus, there is a possible bias towards respondents that are
38 younger and have a higher SES, indicating the findings may not be generalizable to the population of all middle-aged and ol der adults in the United States. Implications for Health E ducation and Public Health The positive association this study shows In ternet use has with social networks can be useful to health educators and other public health authoritie s in planning social network interventions. Some strategies used in social network interventions include enhancing existing social network linkages and developing new social network linkages (Heaney & Israel, 2002). These strategies coul d make use of the Inte rnet to facilitate strengthening of social networks The Internet could be of special use where the selected population lacks mobility, or has close ties at a distance. In addition, this study suggests that Internet use by older adults will increas e as the current middleaged population ages, and that those entering retirem ent are more likely to have computer knowledge than those who retired previously. The more computer literate aging populati on will make Internetbased interventions more re alizable and less expensiv e as training costs drop. This study also suggests that changes to policies that affect availability of the Internet should be considered. There has been legislation to limit taxes on the Internet as an incentive for commercial trans actions (Kraut et al., 1998). It may be possible to create similar policies that would foster social ne tworking via the Internet or increase the availability of the Internet. Such policies should be considered to reduce the digital divide which result suggest exist between those with low SES and those with a higher SES. The Internet could also be tappe d by communities to increase community involvement. The findings of this study that In ternet use is associated with higher levels
39 of participation in meetings (other than religious services) suggests that communities' interventions may include Internet tools withou t the fear that Internet use will diminish in-person participation. Future Directions The results of this study, along with ot her studies reported in the literature, suggest a need for a detailed study of the dida ctic ties that make up social networks and the effect Internet use has on those ties. Such a study would help in understanding how ties based on the Internet fair to off-line ties and the spectrum between. Only through this kind of study can one start to understand the impact Inte rnet use has on the strength of ties. More encompassing details of personality also need to be included in such studies, including level of extroversion for example. With the use of the Internet growing, it is critical that one understa nd its impact on society.
40 References 2004 HRS final core code book (2006, October 2006). Retrieved December 3, 2006, from http://hrsonline.isr.umich.e du/meta/2004/core/codebook/h04_00.html Bargh, J. A., & McKenna, K. Y. A. (2004). The Internet and social life. Annual Review of Psychology, 55 573-590. Berkman, L. F. (1985). The relationship of social networks and social support to morbidity and mortality. In S. Cohen & S. L. Syme (Eds.), Social Support and Health (pp. 241-262). Orlando, FL: Academic Press. Berkman, L. F., & Glass, T. (2000). Social integration, social networks, social support, and health. In L. F. Berkman & I. Kawachi (Eds.), Social Epidemiology (pp. 147-173). New York: Oxford University Press. Bovbjerg, B. D. (1998). Social Security re form: Raising retirement ages improves program solvency but may cause hardship fo r some. In U. S. G. A. Office (Ed.). Coget, J., Yamauchi, Y., & Suman, M. (2002). The Internet, social networks, and loneliness. IT & Society, 1 (1), 180-201. Cohen, S., & Syme, S. L. (1985). Issues in the study and app lication of social support. In S. Cohen & S. L. Syme (Eds.), Social Support and Health (pp. 3-22). Orlando, FL: Academic Press. Crawford, G. (1987). Support networks and health-related change in the elderly: Theory-based nursing strategies. Family and Community Health, 10 39-48.
41 Cummings, J., Lee, J., & Kraut, R. E. (2006). Communication technology and friendship: The transition fr om high school to college. New York, NY: Oxford University Press. Fiori, K. L., Toni, A. C., & Cortina, K. S. (2006). Social network typologies and mental health among older adults. Journal of Gerontology: Psychological Sciences, 61B (1), P25-P32. Fox, S. (2004). Older Americans and the Internet Retrieved January 5, 2007, from http://www.pewinternet.org/pd fs/PIP_Seniors_Online_2004.pdf Fox, S. (2006). Internet penetration and impact April 2006 Retrieved January 5, 2007, from http://www.pewinternet.org/ pdfs/PIP_Internet_Impact.pdf Glass, T. A., F., M. D. L. C., Seema n, T. E., & Berkman, L. F. (1997). Beyond single indicators of social netw orks: A lisrel analysis of so cial ties among the elderly. Social Science and Medicine, 44 (10), 1503-1517. Hall, A., & Wellman, B. (1985). Social netw orks and social support. In S. Cohen & S. L. Syme (Eds.), Social Support and Health (pp. 23-41). Orlando, Fl.: Academic Press, Inc. Havens, B., Hall, M., Sylvestre, G., & Jivan, T. (2004). Social isolation and loneliness: Differences between ol der rural and urban Manitobans. Canadian Journal on Aging, 23 (2), 129-140. Heaney, C. A., & Israel, B. A. (2002). Soci al networks and social support. In K. Glanz, B. K. Rimer & F. M. Lewis (Eds.), Health Behavior and Health Education (pp. 185-209). San Francisco, CA: Jossey-Bass.
42 Heeringa, S. G., & Connor, J. H. (1996, August 9, 2002). Technical Description of the Health and Retirement Survey Sample Design Retrieved December 12, 2006, from http://hrsonline.isr.umich.edu/docs/ sample/sho_samp.php?hfyle=ref023b&xtyp=1 HRS data description and usage (2006). from http://hrsonline.isr.umich.e du/meta/2004/core/desc/h04dd.pdf Irizarry, C., Downing, A., & West, D. ( 2002). Promoting modern technology and Internet access for under-represented older populations. Journal of Technology in Human Services, 19 (4), 13-30. Katz, J. E., & Aspden, P. (1997). Motivations for and barriers to Internet usage: Results of a national public opinion survey. Internet Research, 7 (3), 170-188. Katz, J. E., & Rice, R. E. (2002). In Social consequences of Internet use: Access, involvement, and interaction. (pp. 1-14). Cambridge, MA: The MIT Press. Katz, J. E., Rice, R. E., & Aspden, P. (2001). The Internet, 1995-2000. American Behavioral Scientist, 45 405-419. Kraut, R., Kiesler, S., Boneva, B., Cu mmings, J., Helgeson, V., & Crawford, A. (2002). Internet Paradox Resvisited. Journal of Social Issues, 58 (1), 49-74. Kraut, R., Patterson, M., Lundmark, V., Kiesler, S., Mukopadhyay, T., & Scherlis, W. (1998). Internet paradox. A social technol ogy that reduces social involvement and psychological well-being? American Psychologist, 53 (9), 1017-1031. Lawhon, T., Ennis, D., & Lawhon, D. C. ( 1996). Senior adults and computers in the 1990's. Educational Gerontology, 22 (2), 193-201.
43 Levy, K., McConnaughey, J., Lader, W., Brodsky A., Laousis, S., Price, L., et al. (2000). Fall through the Net: Toward digital inclusion Washington, DC: U.S. Department of Commerce. Litwin, H. (1996). A cross-national social network analysis. In H. Litwin (Ed.), The Social Networks of Older Pe ople; A Cross-National Analysis (pp. 205-221). London: Praeger Publishers. Lubben, J., & Gironda, M. (1996). Asse ssing social support networks among older people in the United States. In H. Litwin (Ed.), The Social Networks of Older People: A Cross-National Analysis (pp. 143-161). London: Praeger Publishers. Malcolm, M., Mann, W., Tomita, M., Fraas L., Stanton, K., & Gitlin, L. (2001). Computer and Internet use in physically frail elders. Physical & Occupational Therapy in Geriatrics, 19 (3), 15-32. Mercer, Z. B., Chiriboga, D., & Sw eeney, M. A. (1997). Using computer technology with older adults: a pi lot study on advanced directives. Gerontology & Geriatrics Education, 18 (1), 61-76. Moss, M., & Lawton, M. P. (1982). Time budgets of older people: A window on four lifestyles. Journal of Gerontology, 37 (1), 115-123. Mullins., L. C., Elston, C. H., & Gutkowski, S. M. (1996). Social determinants of loneliness among older Americans. Genetic, Social, and General Psychology Monographs, 122 (4), 453-473. Neugarten, B. L. (1977). Personality and ag ing. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the Psychology of Aging (pp. 626-649). New York: Van Nostrand Reinhold Co.
44 Nie, N., & Hillygus, D. S. (2002). The imp act of Internet use on sociability: Time diary findings. IT & Society, 1 (1), 1-20. Oxman, T. E., & Berkman, L. F. (1990). A ssessment of social relationships in elderly patients. International Journal of Psychiatry in Medicine, 20 (1), 65-84. Pillemer, K., & Glasgow, N. (2000). Soci al integreation and aging: Background and trends. In K. Pillemer, P. moe n, E. Wethington & N. Glasgow (Eds.), Social integreation in the second half of life (pp. 19-47). Baltimore: The John Hopkins University Press. RAND Contributions: RAND HRS Data File (v.F) (2006). Retrieved December 28, 2006, from https://ssl.isr.um ich.edu/hrs/file s.php?versid=34 Seeman, T. E. (1996). Social ties and hea lth: The benefits of social integration. Annals of Epidemiology, 6 (5), 442-451. Seeman, T. E., & Berkman, L. F. (1988). Structural characteristics of social networks and their relationship with social support in the el derly: Who provides support. Social Science and Medicine, 26 (7), 737-749. Stevens, J. (1996). Applied multivariate st atistics for the social sciences. 108-109. Stokes, J. P. (1983). Pred icting satisfaction with so cial support from social network structure. American Journal of Community Psychology, 11 (2), 141-153. Wu, B., & Prorell, F. (2000). Job charact eristics and leisure physical activity. Journal of Aging and Health, 12 (4), 538-559.
46 Appendix A: Data Descriptions The following is a description of all modifi cations made to variables in this study. The first line describes the variable, the sec ond line gives the origin al variable name and dataset. Datasets are noted as follows: PLB is the Psychosocial Leave-Behind dataset; HRS Core is the 2004 Core dataset; Tracker is the 2004 Tracker file released in January 2007; RAND is the 2004 RAND HRS dataset vers ion F. If the variable is renamed, the new name follows the original name/dataset. If there is any additional information about the variable or details of changes made, they are listed on a third line. Miscellaneous Variables Have any immediate family (example, any br others or sisters, parents, cousins or grandchildren) o JLB514 PLB Have any friends o JLB518 PLB Have any children o JLB510 PLB have_chd o If missing and h7child=0 then =no; if mi ssing and h7child>0 then =yes (change 5 to 0). Number of living children o H7CHILD PLB Meet up with children o JLB512A PLB mt_child o If no children (jlb510) and is missing da ta, set to 6 (never), reverse code. Speak on phone with children o JLB512B PLB ph_child o If no children (jlb510) and is missing da ta, set to 6 (never), reverse code. Write or Email children o JLB512C PLB wr_child o If no children (jlb510) and is missing da ta, set to 6 (never), reverse code. Meet up with other family members o JLB516A PLB mt_ofm o If no other family members(jlb514) and is missing data, set to 6 (never), reverse code.
47 Appendix A: (Continued) Speak on phone with other family members o JLB516B PLB ph_ofm o If no other family members(jlb514) and is missing data, set to 6 (never), reverse code. Write or Email other family members o JLB516C PLB wr_ofm o If no other family members(jlb514) and is missing data, set to 6 (never), reverse code. Meet up with friends o JLB520A PLB mt_frd o If no friends (jlb514) and is missing da ta, set to 6 (never), reverse code. Speak on phone with friends o JLB520B PLB ph_frd o If no friends (jlb514) and is missing da ta, set to 6 (never), reverse code. Write or Email friends o JLB520C PLB wr_frd o If no friends (jlb514) and is missing da ta, set to 6 (never), reverse code. Open up to Spouse o JLB508C PLB open_sp o If no spouse(jcouple=5) and is missing data, set to 4 (not at all), reverse code. Open up to children about worries o JLB511C PLB open_chd o If no children (jlb510) and is missing data, set to 4 (not at all)., reverse code. Open up to other family about worries o JLB515C PLB open_ofm o If no other family members(jlb514) and is missing data, set to 4 (not at all). Open up to friends about worries o JLB519C PLB open_frd o If no friends (jlb514) and is missing data, set to 4 (not at all)., reverse Code. Number of children in close relationship o JLB513 PLB num_chd o If no children (jlb510) and is missing data, set to zero. Number of other family in close relationship o JLB517 PLB num_ofm o If no other family members(jlb514) and is missing data, set to zero. Number of friends in close relationship o JLB521 PLB num_frd o If no friends (jlb514) and is missing data, set to zero Left Behind Survey Type o JLBTYPE PLB o (1=financial, 2=psycosocial, 3=both)
48 Appendix A: (Continued) Weight o JWGTR Tracker rel_wgt o Raw weight used to account for oversamp ling (divide by mean wgt of sample to get relative weight). SECU o SECU Tracker o Used to account for complex sample design (SAS Cluster variable). Strata o STRATUM Tracker o Used to account for complex sample design (SAS Strata variable). Number of missing variables o num_miss o Calculated by checking for missing data on all variables to be used in bivariate and regression analyses. Dependent Variables Contacts with Friends o con_friend o Mean(mt_frd, ph_frd, wr_frd). Contacts with Family o con_family o Mean (mt_child, ph_child, wr_ch ild, mt_ofm, ph_ofm, wr_ofm). In-person meetings with friends o mt_frnd ipm_friend In-person meetings with children o mt_child ipm_child In-person meetings with other family members o mt_ofm ipm_ofm Child confidant o open_chd cfd_child Other family confidant o open_ofm cfd_ofm Friend Confidant o open_frd cfd_friend Number of close Family o num_chd nc_child Number of close other family o nc_ofm o num_ofm, if>12 then set to 12 (95th percentile) Number of close friends o nc_friend o num_frd, if >14 then set to 14 (95th percentile)
49 Appendix A: (Continued) Attend meetings or programs o JLB502 PLB att_meet o Reverse code. Attend religious Services o JB082 PLB att_rel o Codes 8&9 = missing, reverse Code. Independent & Control Variables Internet Use o JW303 HRS Core int_use o Recode 5 to 0 (do not use Internet ), 1= use Internet, over 5=missing. Race of respondent o RACE Tracker race2 o 1=W(1), 2=AA(2), 3=OTH(7), missing (0,.) African American dummy code o race2 race_aa o African American/ white as re ference (dummy code). Other race dummy code o race2 race_oth o Other / white as reference (dummy code). Age at start of 2004 interview o JAGE Tracker Education in years o SCHLYRS Tracker schooly o Recode over 17 missing (97=other). Hispanic o HISPANIC Tracker hisp2 o From Mexican American, Other Hisp anic, Not Hispanic, to Yes/No. Gender or respondent at time of interview o GENDER Tracker o Change from 1,2(female) to 0,1(female). Married or Partnered o JCOUPLE Tracker married o Recode 5=0. Health stops me from doing things o JLB503H PLB hlth_bar o Reverse code. Self Reported Heatlh Status o JC001 HRS Core health o recode over 5 is missing (8&9=missing codes). Reverse Code. Household Residents o H7HHRES RAND hh_residents o If >4 set to 4 (95th percentile).
50 Appendix A: (Continued) Job Status o JJ005M1 HRS Core work_now o If =1 set to 1(Working); else if 2-7 se t to 0 (Not working); else set to missing. Longest Job Occupation Code o R7JLOCC RAND occ_code o [1-4]=1 (white collar), [5-16]=0 (blue collar), [17-military, or missing]=2 (unknown). Blue collar dummy code o occ_code occ_blue o Blue collar, white collar as reference (dummy code). Unknown occupation dummy code o occ_code occ_other o Unknown occupation with white collar as reference (dummy code). Total household income (respondent & Spouse) o H7ITOT RAND hh_income o If > 187500 then it is set to 187500 (95th percentile). Divide by 1000.
51 Appendix B: Tables of Association Table B-1: Chi-square, P-value, and Cramers V for categorical variables 1 2 3 4 5 6 7 1.Internet Use 2.Race 2 25.35 p > 2 0.0002 V 0.1004 3.Gender 2 0.70 0.82 p > 2 0.4285 0.6466 V 0.0167 0.0180 4.Hispanic 2 21.46 439.49 1.39 p > 2 0.0001 0.0000 0.2761 V 0.0923 0.4179 0.0235 5.Married 2 67.25 57.69 60.80 0.37 p > 2 0.0000 0.0000 0.0000 0.5103 V 0.1635 0.1514 0.1554 0.0121 6.Working Now 2 190.18 0.13 19.08 0.71 23.97 p > 2 0.0000 0.9338 0.0001 0.5126 0.0000 V 0.2749 0.0071 0.0871 0. 0168 0.0976 7.Occupation 2 205.19 19.28 62.81 10. 47 23.27 170.66 p > 2 0.0000 0.0062 0.0000 0. 0025 0.0003 0.0000 V 0.2855 0.06188 0.1580 0. 0645 0.0962 0.2604 8.Have Children 2 9.17 1.57 4.43 0. 01 85.17 19.92 2.94 p > 2 0.0257 0.6499 0.1040 0.944 9 0.0000 0.0008 0.4018 V 0.0604 0.0250 0.0420 0.001 7 0.1839 0.0890 0.0342 Note: A p-value of zero indicates p<.0001.
52 Appendix B: (Continued) Table B-2: Correlation matrix 1 2 3 4 5 6 7 8 9 10 1. Age 1 2. HH Residents -0.2587 1 3. HH Income -0.2590 0.1928 1 4. Education -0.1744 0.0000 0.4140 1 5. Health Barriers 0.1739 -0.0378 -0.2515 -0.1713 1 6. Health Status -0.0708 0.0246 0.3288 0.3103 -0.5719 1 7. Contacts w/Family 0.0559 0.1191 0.0656 -0.0263 -0.0178 0.0839 1 8. Contacts w/Friends 0.0344 -0.0749 0.1345 0.1817 -0.0843 0.1510 0.3467 1 9. In-person w/Children 0.0834 0.1177 0.0443 -0.0960 0.0000 0.0253 0.7056 0.1002 1 10. In-person w/Other Family 0.0000 0.0000 -0.0577 -0.1319 0.0000 0.0000 0.6213 0.1698 0.3563 1 11. In-person w/Friends 0.0636 -0.0891 0.1058 0.1133 -0.0790 0.1122 0.2293 0.8023 0.1495 0.2281 12. Children Confidants 0.1906 0.1336 0.0000 -0.0754 -0.0361 0.0794 0.5111 0.1098 0.5322 0.1645 13. Other Family Confidants 0.0073 0.0000 -0.0515 -0.0425 -0.0327 0.0589 0.3527 0.1751 0.1335 0.3437 14. Friend Confidants 0.0000 -0.0884 0.0289 0.1039 -0.0579 0.1209 0.1493 0.5241 0.0399 0.1040 15. Number Close Children 0.1942 0.2202 -0.0293 -0.1417 0.0000 0.0000 0.3821 0.0000 0.4480 0.1392 16. Number Close Other Family 0.1158 0.0609 -0.0355 -0.1094 0.0000 0.0000 0.3217 0.1127 0.1895 0.3426 17. Number Close Friends 0.1674 0.0688 0.0000 0.0550 -0.0490 0.0774 0.1144 0.3886 0.0491 0.0870 18. Attend Meetings 0.0948 -0.0432 0.1070 0.1927 -0.0757 0.1229 0.1592 0.3130 0.0420 0.0676 19. Attend Religious Services 0.1218 -0.0261 0.0000 0.0776 -0.0493 0.1340 0.1999 0.1711 0.1355 0.1070 20. Internet Use + -0.3225 0.1121 0.3581 0.3986 -0.1458 0.2013 0.0889 0.2367 -0.0759 -0.1167 21. Race 0.1121 0.0666 0.1052 0.0940 0.0301 0.1367 0.0558 0.0876 0.0269 0.0243 22. Gender + 0.0277 -0.0678 -0.1000 -0.1007 0.0346 0.0000 0.2095 0.1885 0.0895 0.1004 23. Hispanic + -0.0986 0.1106 -0.1084 -0.1931 0.0133 -0.1017 0.0000 -0.1078 0.0000 0.0000 24. Married + -0.1215 0.4659 0.3607 0.0497 -0.0797 0.1309 0.1111 -0.0230 0.1343 0.0000 25. Working Now + -0.5241 0.1537 0.3769 0.2584 -0.2530 0.2372 -0.0221 0.0000 -0.0347 -0.0246 26. Occupation 0.3686 0.1144 0.3102 0.4366 0.1745 0.2320 0.0351 0.1562 0.0505 0.0888 27. Have Children + 0.1362 0.1965 0.0000 -0.1209 0.0000 0.0000 0.5072 -0.0304 0.6064 0.0925 Values are Spearman rank correlations except where marked. = Eta values. + = Point-biserial All correlations can be interpreted as Pierson Product Moment correlations. Critical Values for 2284 cases: at r =.0410 =.05, at r =.0539 =.01, at r =.069 =.001, at r =.0795 =.0001. For categorical to categorical associations, see table B-1.
53 Appendix B: (Continued) Table B-2. Correlation Matrix (Continued) 11 12 13 14 15 16 17 18 19 1. Age 2. HH Residents 3. HH Income 4. Education 5. Health Barriers 6. Health Status 7. Contacts w/Family 8. Contacts w/Friends 9. In-person w/Children 10. In-person w/Other Family 11. In-person w/Friends 1 12. Children Confidants 0.1049 1 13. Other Family Confidants 0.1327 0.2812 1 14. Friend Confidants 0.4330 0.2034 0.2900 1 15. Number Close Children 0.0172 0.5073 0.1079 0.0000 1 16. Number Close Other Family 0.1089 0.2373 0.3570 0.1107 0.3133 1 17. Number Close Friends 0.3706 0.1478 0.1862 0.3293 0.0805 0.3124 1 18. Attend Meetings 0.2655 0.0649 0.0931 0.1183 0.0386 0.0738 0.1843 1 19. Attend Religious Srvs 0.1487 0.1429 0.1445 0.1221 0.1238 0.1473 0.2100 0.2892 1 20. Internet Use 0.0522 -0.0609 -0.0422 0.0507 -0.0898 -0.0925 -0.0392 0.1073 0.0000 21. Race 0.0991 0.0383 0.0000 0.0605 0.0164 0.0619 0.0604 0.0466 0.0692 22. Gender + 0.0697 0.1382 0.1323 0.2095 0.0509 0.0577 0.0000 0.0433 0.0841 23. Hispanic + -0.1102 0.0000 0.0457 -0.0598 0.0000 0.0000 -0.0301 -0.0111 0.0084 24. Married + -0.0159 0.1255 -0.0217 -0.0553 0.1094 0.0560 0.0218 0.0000 0.0340 25. Working Now + -0.0249 -0.0941 -0.0170 0.0000 -0.1034 -0.0978 -0.0868 -0.0535 -0.0706 26. Occupation 0.0605 0.0553 0.0000 0.0887 0.0997 0.0309 0.0529 0.1457 0.0865 27. Have Children + 0.0000 0.6505 0.0427 -0.0154 0.5009 0.0941 0.0000 0.0158 0.1011 Values are Spearman rank correlations except where marked. = Eta values. + = Point-Biserial All correlations can be interpreted as Pierson Produce Moment correlations. Critical Values for 2284 cases: at r =.0410 =.05, at r =.0539 =.01, at r =.069 =.001, at r =.0795 =.0001. For categorical to categorical associations, see table B-1.