You are on page 1of 35

Journal of Vocational Behavior

Volume 67, Issue 2, October 2005, Pages 169198

A meta-analytic review of workfamily conflict and its antecedents


Kristin Byron
,

College of Business, Rochester Institute of Technology, 103 Lomb Memorial Drive, Rochester, NY 14623,
USA

Received 6 November 2003. Available online 2 March 2005. http://dx.doi.org/10.1016/j.jvb.2004.08.009, How to Cite or Link Using DOI Permissions & Reprints

Abstract
This meta-analytic review combines the results of more than 60 studies to help determine the relative effects of work, nonwork, and demographic and individual factors on work interference with family (WIF) and family interference with work (FIW). As expected, work factors related more strongly to WIF, and some nonwork factors were more strongly related to FIW. Demographic factors, such as an employees sex and marital status, tended to relate weakly to WIF and FIW. Overall the analysis supports the notion that WIF and FIW have unique antecedents, and therefore, may require different interventions or solutions to prevent or reduce their occurrence. Lastly, the analysis suggests that demographic variables, such as sex and marital status, are alone poor predictors of workfamily conflict. Researchers are advised to

attend to more finely grained variables that may more fully capture employees likelihood of experiencing workfamily conflict.

Keywords
Work; Family; Conflict; Interference; Meta-analysis; Review

1. Introduction
The increase in dual-career couples and single-parent households and the concomitant decrease in traditional, single-earner families mean that responsibilities for work, housework, and childcare are no longer confined to traditional gender roles. Increasingly, employees find themselves struggling to juggle the competing demands of work and family. The problems and issues encountered by employees taking part in this balancing act has prompted a burgeoning body of research and theory on the intersections of individuals work and family lives (e.g., [Kossek et al., 1999] and [Perrewe and Hochwarter, 2001]). One of the most studied concepts in the workfamily literature is workfamily conflict. Workfamily conflict, also called workfamily interference, is a type of interrole conflict (Kahn, Wolfe, Quinn, Snoek, & Rosenthal, 1964) that occurs when the demands of work and family roles conflict.

Since the construct of workfamily conflict was introduced, a large body of literature has examined its causes and consequences. Recent meta-analyses have examined the relation between workfamily conflict and its consequences, such as job and life satisfaction, burnout, and absenteeism ( [Allen et al., 2000], [Kossek and Ozeki, 1998] and [Kossek and Ozeki, 1999]). These meta-analyses underscore the potentially negative effects of workfamily conflict for individuals and their employing organizations. However, among the published meta-analyses on workfamily conflict, only one has examined a potential antecedent, job/work involvement (Kossek & Ozeki, 1999). No meta-analysis to date has comprehensively considered the myriad causes of work family conflict that have been examined in the literature.

In addition, the concept of workfamily conflict has changed over time. Increasingly, researchers have acknowledged the direction of interference (ODriscoll, Ilgen, & Hildreth, 1992). That is, workfamily conflict is increasingly recognized as consisting of two distinct, though related, concepts, work interference with family (WIF) and family interference with work (FIW). WIF (also termed work-to-family conflict) occurs when

work interferes with family life, and FIW (known also as family-to-work conflict) occurs when family life interferes with work (Frone, Yardley, & Markel, 1997). Support for distinguishing these two concepts comes from several sources. First, in their metaanalysis, Kossek and Ozeki (1998) reported consistent support for distinguishing between the direction of workfamily conflict. Second, recent theory and research on WIF and FIW suggests that these two concepts may have different causes and effects (e.g., [Frone et al., 1992a], [Frone et al., 1992b] and [Kelloway et al., 1999]). In summary, while the potentially harmful effects of workfamily conflict are recognized, we know less about the causes of workfamily conflict and their relative effects on WIF and FIW. Consequently, a systematic review of the literature on workfamily conflict antecedents is needed to explain the experience of workfamily conflict in employees lives. The present study offers such an analysis by providing a quantitative review of potential antecedents and their relation to two types of workfamily conflict, work interference with family (WIF) and family interference with work (FIW).

2. Workfamily conflict and its antecedents


Researchers have considered a number of different variables as possible antecedents of WIF and FIW. Consistent with Eby, Casper, Lockwood, Bordeaux, and Brinley (in press) classification scheme for antecedents of workfamily conflict, the previously researched antecedents can be classified into three categories: work domain variables, nonwork domain variables, and individual and demographic variables. Work domain variables consider the effect of job and workplace factors, such as schedule flexibility and job stress. Nonwork domain variables consider the family demands and other nonwork factors, such as marital conflict, number of hours spent on housework or childcare, and age of youngest child. Demographic or individual variables include personality, behaviors, and other individual differences, such as sex, income, and coping style.

While other theories about the intersection of work and family exist, the constructs of WIF and FIW have their roots in conflict theory. Conflict theory proposes that work and family domains are incompatible due to their different norms and responsibilities (Greenhaus & Beutell, 1985). The differing norms and responsibilities of work and family cause intrusion and negative spillover of one domain on the other. Consistent with research and theory on WIF and FIW, antecedents related to different domains may have differential effects on WIF and FIW. Factors related to an individuals job (work domain variables) are expected to be more related to WIF than to FIW. For example, the more hours individuals spend at work, the more likely it is that their work will interfere with their family life. Similarly, factors related to individuals family and nonwork life (nonwork domain variables) are expected to relate more to FIW than to WIF. For example, individuals with more supportive families may experience less FIW,

yet may not have less WIF. In contrast, individual and demographic variables, such as sex or income, are expected to have equivalent effects on WIF and FIW because they may simultaneously influence both work and nonwork. I propose the following hypotheses: Hypothesis 1. Work-domain variables relate more to WIF than to FIW. Hypothesis 2. Nonwork-domain variables relate more to FIW than to WIF. Hypothesis 3. Demographic and individual variables have equivalent effects on WIF and FIW. Research examining proposed antecedents of WIF and FIW has produced mixed results. Therefore, when warranted, I consider several potential moderators of these relationships. Differences in sampling strategies may explain some of the variation in results between studies. For example, some studies enlisted only parents, whereas others made no such restrictions. In addition, some studies sampled only women or only men, whereas others had a mixed sex sample. I explore whether differences in sample composition, such as the percentage of parents or females in them sample, may moderate the relationship between antecedents and WIF and FIW. In the studies used in the analysis, the percentage of parents ranged from 16 to 100 percent; and the percentage of females in the sample ranged from 0 to 100 percent. Previous research and theory suggests that being female or having children may explain differences in results across studies (e.g., [Eagle et al., 1998] and [Voydanoff, 2002]). Therefore, the present analysis considers whether the percentage of females or the percentage of study participants with children explains between-study variance. In addition, other meta-analyses have reported that variation in how variables are measured account for significant differences in results across studies (e.g., Verquer, Beehr, & Wagner, 2003). Therefore, in the present analysis, I consider whether variation in how antecedents were measured accounts for differences in results between studies. For example, many studies examined whether having more children related to more WIF or FIW. Among the studies considering this relationship, study participants number of children was measured in different ways. Some asked participants how many children they had living at home, others asked participants how many children they had (with no restrictions), and others asked participants whether they had children or not. The different methods of measuring this and other proposed antecedents of WIF and FIW may account for observed differences in findings between studies. Therefore, I explored whether differences in measurement moderate the relationships between proposed antecedents and WIF and FIW. Fig. 1 demonstrates the relationships examined in the present meta-analysis. The solid lines represent relationships that are proposed to be stronger (i.e., of higher magnitude) than those represented by dashed lines. Dotted lines represent relationships of

undetermined magnitude, and curved lines represent the relationship between WIF and FIW. It should be noted that the relationships represented in the figure do not imply that alternate relationships are implausible. Rather, some of these alternate relationships are mathematically equivalent and their plausibility is not being rejected, e.g., family conflict may also be a consequence, rather than a cause, of WIF and FIW. It should also be noted that more complex relationships than those represented in Fig. 1 are similarly plausible. For example, family support may moderate the relationship between spousal employment and WIF or FIW. Unfortunately, the data available to meta-analytic researchers can preclude the investigation of more complex relationships. Therefore, this meta-analysis focuses on the factors that have been identified in the literature as potential antecedents and their relationship to WIF and FIW.

<img class="figure large" border="0" alt="Full-size image (20K)" src="http://ars.sciencedirect.com/content/image/1-s2.0-S0001879105000187-gr1.gif" data-thumbsrc="http://ars.sciencedirect.com/content/image/1-s2.0S0001879105000187-gr1.sml" datafullsrc="http://ars.sciencedirect.com/content/image/1-s2.0-S0001879105000187gr1.gif">

Fig. 1. Proposed relationships between variables in meta-analysis. Note. Solid lines represent direct relationships hypothesized to be stronger in magnitude than those represented by dashed lines. The dotted lines represent relationships of undetermined magnitude, and curved lines represent correlation rather than causation.

In the next sections, I describe the method for finding, selecting, and coding studies for the meta-analysis. Then, I specify the method for quantitatively cumulating the results in the studies. In the following section, I present the results of the review for each of the three categories of antecedents of WIF and FIW, and the relationship of WIF and FIW. Lastly, I discuss the implications of the results of the meta-analysis and provide suggestions for future research.

3. Method
3.1. Search strategy
I searched the computer database Psych-Info of the American Psychological Association using the keywords work, family, conflict, and work, family, interference for articles published in academic journals, resulting in more than 500 studies. After eliminating duplicates and studies that were not related to workfamily conflict (e.g., those that were related to family conflict), the 243 remaining studies were reviewed for possible inclusion. In addition, I searched the reference lists of three recently published meta-analyses on workfamily conflict ( [Allen et al., 2000], [Kossek and Ozeki, 1998] and [Kossek and Ozeki, 1999]) and one review article (Swanson, 1992) to locate articles that had not turned up in my computer database search. Lastly, I posted a message on the general on-line forum on the Sloan Work and Family Research Network website, and on the Workfam Newsgroup of the Work/Family Initiative at Pennsylvania State University soliciting research on workfamily conflict. Because so few studies that were not subsequently published were located, the present analysis is restricted to published studies. There are several factors that should mitigate concern about publication bias. First, many of the relationships included in the present analysis were from studies that were not explicitly considering the relationships. For example, most of the studies included in the meta-analysis of sex and workfamily conflict were not explicitly considering this relationship. Second, I included two estimates of the stability of each effect size, (1) the number of studies needed to meaningfully change the estimated effect size and (2) 95% confidence intervals of each effect size.

3.2. Criteria for inclusion


For a study to be considered for inclusion, the study had to meet the following criteria: 1.

WIF and FIW had to be quantitatively measured; qualitative studies of workfamily conflict were eliminated. 2. The study had to report the relationship between a previously proposed antecedent and WIF and FIW or between WIF and FIW in a form that could be converted to a correlation. Studies that merely stated findings without providing details of those results, or did not provide data in a usable form were eliminated. 3. Only studies published by 2002, and written in the English language were included. 4. Only studies that examined both WIF and FIW were included to increase the confidence that observed differences were due to differences in the relationship rather than due to differences in samples. 5. When fewer than five studies could be located for a particular antecedent, the antecedent was not examined. For example, antecedents such as supervisory responsibilities, job type or level, child care satisfaction, and self-efficacy were excluded. Because the formula used in meta-analysis assumes that the studies used are statistically independent, I avoided violating the assumption of independence of studies by eliminating duplicate results from the same dataset. When more than one study used the same sample, only one was included, when a sample was a subset of a larger sample, only the study that used the larger sample was included. However, when studies using the same sample considered different antecedents, each was included in its respective analysis, but only the one with the larger sample was included when they considered the same variable. In all, 61 studies met these criteria and were included in the analysis. Some of these studies had multiple independent samples, which were included as independent outcomes. Table 1 lists the studies by sample included in the meta-analysis, their sample characteristics, and the measure of WIF and FIW used.

Table 1. Summary of studies and their characteristics Author(s) and Publication WIF measure year/sample characteristics Adams and Jex (1999) [Frone et al., [Frone et al., Part-time university students 1992a] and [Frone 1992a] and [Frone 490 US et al., 1992b] et al., 1992b] Adams, King, and King FIW measure N Country

Author(s) and Publication WIF measure year/sample characteristics (1996)

FIW measure

Country

Varied professions, full-time Kopelman, and living with a family Greenhaus, and Burley (1989; 5) member Connolly (1983; 4) Aryee, Fields, et al. (1999) Varied professions Aryee, Luk, Leung, and Lo (1999) Public sector and university Netemeyer, Boles, Netemeyer et al. employees, parents in dual- and McMurrian (1996; 5) earner families (1996; 5) Barling, MacEwen, Kelloway, and Higginbottom (1994) University employees Beutell and Witting-Berman (1999) Varied professions, married, Kopelman et al. and employed (1983; 6) Boles, Howard, and Donofrio (2001) Probation and parole officers Brough and Kelling (2002) Varied professions Bruck, Allen, and Spector (2002) Hospital employees, married Carlson, Kacmar, or living with partner or and Williams Carlson et al. Netemeyer et al. (1996; 5) Netemeyer et al. (1996; 5) Netemeyer et al. (1996; 5) Netemeyer et al. (1996; 5) Kopelman et al. (1983; 4) *Kopelman et al. (1983; 4) *Kopelman et al. (1983; 4)

146 US

Gutek, Searle, and Gutek et al. (1991; 320 Hong Kong Klepa (1991; 4) 4)

243 Hong Kong

141 Canada

177 US

144 US

691

New Zealand

160 US

Author(s) and Publication WIF measure year/sample characteristics child Burke and Greenglass (2001) Registered nurses Carlson and Kacmar (2000) Kopelman et al. (1983; 4) (2000; 9)

FIW measure (2000; 9)

Country

Burley (1989; 4)

527 Canada

Gutek et al. (1991); Gutek et al. (1991); State government [Frone et al., [Frone et al., employees, married or with 314 US 1992a] and [Frone 1992a] and [Frone children et al., 1992b] et al., 1992b] Carlson et al. (2000) Varied; employed full-time Casper, Martin, Buffardi, and Erdwins (2002) Female employees with preschool-aged child Cinamon and Rich (2002) Lawyers and employees in Gutek et al. (1991); Gutek et al. (1991); computer or software field, 213 Israel Self-developed; 7 Self-developed; 7 married Cohen and Kirchmeyer (1995) Nurses Duxbury et al. (1994) Varied professions, parents Bohen and Viveros- Bohen and Viveros1989 Canada Long (1981; 4) Long (1981; 5) of child age 612 Eagle et al. (1998) Varied professions, [Frone et al., employed full-time, married 1992a], [Frone et [Frone et al., 318 US 1992a] and [Frone Shamir (1983; 6) Kirchmeyer (1993; 227 Canada 8) Kopelman et al. (1983; 4) Burley (1989; 4) 143 US Carlson et al. (2000; 9) Carlson et al. (2000; 9) 225 US

Author(s) and Publication WIF measure year/sample characteristics or living with child

FIW measure

Country

al., 1992b], [Gutek et al., 1992b]; et al., 1991] and [Gutek et al., 1991] [Wiley, 1987]; 11 and [Wiley, 1987]; 13

Eagle, Miles, and Icenogle (1997) [Frone et al., University employees 1992a], [Frone et (married or with children at al., 1992b], [Gutek et al., 1991] and home) [Wiley, 1987]; 11 Fox and Dwyer (1999) Registered nurses Frone and Yardley (1996) [Frone et al., 1992a], [Frone et Financial services company al., 1992b] and employees [Gutek et al., 1991]; 6 Frone (2000) National Commorbidity Survey, employed and married or parent of child under 18 Frone, Russell, and Barnes (1996) [Frone et al., [Frone et al., Longitudinal follow-up (Erie 1992a] and [Frone 1992a] and [Frone 496 US County, NY), Wave 2 et al., 1992b] et al., 1992b] Longitudinal follow-up (Buffalo, NY), Wave 3 [Frone et al., [Frone et al., 1992a] and [Frone 1992a] and [Frone 605 US et al., 1992b] et al., 1992b] [Frone et al., 1992a], [Frone et al., 1992b] and [Gutek et al., 1991]; 6 Self-developed; 2 Self-developed; 2 113 US [Frone et al., 1992a], [Frone et al., 1992b], [Gutek 493 US et al., 1991] and [Wiley, 1987]; 13

252 Canada

Self-developed; 2

Self-developed; 2

2700 US

Author(s) and Publication WIF measure year/sample characteristics Frone et al. (1992a)

FIW measure

Country

Longitudinal follow-up (Erie [Frone et al., [Frone et al., County, NY), employed and 1992a] and [Frone 1992a] and [Frone 631 US married or with child at et al., 1992b] et al., 1992b] home Frone et al. (1997) [Frone et al., Varied professions, married 1992a], [Frone et or with children living at al., 1992b] and [Gutek et al., home 1991]; 6 Fu and Shaffer (2001) University employees Gignac, Kelloway, and Gottlieb (1996) Work and family survey, Gutek et al. (1994; Gutek et al. (1994; employees caring for elderly 659 Canada 4) 4) relative Grandey and Cropanzano (1999) University faculty Greenhaus, Parasuraman, and Collins (2001) Public accountants, married Greenhaus and and with one or more Beutell (1985; 9) children Grzywacz and Marks (2000) National Survey of Midlife Development Self-developed; 4 Self-developed; 4 1986 US Kopelman et al. (1983; 6) Self-developed; 5 132 US Carlson et al. (2000; 9) Carlson et al. (2000; 9) 267 Hong Kong [Frone et al., 1992a], [Frone et al., 1992b] and [Gutek et al., 1991]; 6

372 Canada

DNR; 4

199 US

Author(s) and Publication WIF measure year/sample characteristics Gutek et al. (1991) Psychologists Senior managers in executive education Hepburn and Barling (1996) University employees Higgins, Duxbury, and Lee (1994) Federal and private-sector employees, married with children Hughes and Galinsky (1994) Married employees of large Self-developed; 8 corporation Hughes, Galinsky, and Morris (1992) Pharmaceutical company employees Jex and Elacqua (1999) Varied professions, employed students Judge, Boudreau, and Bretz (1994) Self-developed; 8 *Kopelman et al. (1983; 4) Kopelman et al. (1983; 4) Kopelman et al. (1983; 4)

FIW measure

Country

Burley (1989; 4)

423 US

Burley (1989; 4)

176 US

*Kopelman et al. (1983; 4)

17

Canada

Bohen and Viveros- Bohen and Viveros3616 Canada Long (1981; 4) Long (1981; 8)

Self-developed; 6

429 US

Self-developed; 6

522 US

[Frone et al., [Frone et al., 1992a] and [Frone 1992a] and [Frone 515 US et al., 1992b] et al., 1992b]

Male executives from search Gutek et al. (1991; Gutek et al. (1991; 1062 US firm database 4) 4) Kelloway et al. (1999)

Author(s) and Publication WIF measure year/sample characteristics Healthcare and grocery organizations employees Kinnunen and Mauno (1998)

FIW measure

Country

Self-developed; 11 Self-developed; 11 236 Canada

Public sector, manufacturing, and home market employees

[Frone et al., [Frone et al., 1992a] and [Frone 1992a], [Frone et et al., 1992b]; Brett al., 1992b] and and Yogev (1985); [Brett and Yogev, 3 1985]; 3

501 Finland

Kirchmeyer and Cohen (1999) School teachers, administrators, and staff Klitzman, House, Israel, and Mero (1990) Manufacturing company employees Kossek, Colquitt, and Noe (2001) University employees Leiter and Durup (1996) Hospital employees with families Loerch, Russell, and Rush (1989) [94] and [Wiley, University support staff and 1987]; Selfadministrators developed; 18 MacEwen and Barling (1994) Police, employed full-time, married, and with child at Kopelman et al. (1983; 8) Kopelman et al. (1983; 8) 40 Canada Kopelman et al. (1983; 4) Burley (1989; 4) 151 Canada Gutek et al. (1991; Gutek et al. (1991; 490 US 4) 4) Self-developed; 3 Self-developed; 3 573 US Shamir (1983; 6) Kirchmeyer (1993; 200 Canada 8)

Self-developed; 12 156 US

Author(s) and Publication WIF measure year/sample characteristics home Mallard and Lance (1998) Federal employees with children living at home Marks (1998) Wisconsin Longitudinal Study, varied; full- or parttime employees Matsui, Ohsawa, and Onglatco (1995) Employees at refinery, advertising and HR consulting companies McManus, Korabik, Rosin, and Kelloway (2002)

FIW measure

Country

Self-developed; 17 Self-developed; 15 143 US

Self-developed; 3

Self-developed; 4

5782 US

Self-developed; 5

Self-developed; 5

131 Japan

Healthcare and grocery Gottlieb, Kelloway, Gottlieb et al. employees, married or single and Barham (1998; (1998; 11) mothers 11)

178 Canada

Accountants, engineers, and banking and Gutek et al. (1991; Gutek et al. (1991; telecommunications 72 4) 4) employees, married or single mothers Netemeyer et al. (1996) Teachers Small business owners Real estate agents Nielson, Carlson, and Lankau (2001)

Canada

Self-developed; 22 Self-developed; 21 182 US Self-developed; 22 Self-developed; 21 162 US Self-developed; 22 Self-developed; 21 186 US

Undergraduate business [Gutek et al., [Gutek et al., 272 US graduates, married or living 1991], [Frone et al., 1991], [Frone et al.,

Author(s) and Publication WIF measure year/sample characteristics with child

FIW measure

Country

1992a] and [Frone 1992a] and [Frone et al., 1992b]; 5 et al., 1992b]; 5

ODriscoll et al. (1992) Varied professions, working Self-developed; 7 at least 20 h/week Parasuraman, Purohit, Godshalk, and Beutell (1996) Business owners Perrewe, Hochwarter, and Kiewtiz (1999) Hotel managers Shaffer, Harrison, Gilley, and Luk (2001) Married international Gutek et al. (1991; Gutek et al. (1991; 324 Varied assignees, varied professions 3) 3) Shannon et al. (2001) Hospital employees Stoeva, Chiu, and Greenhaus (2002) Senior civil servants Thompson and Blau (1993) *Burke, Weir, and *Burke et al. (1979; DuWor (1979; 234 US DNR) DNR) Kopelman et al. (1983; 4) Burley (1991; 4) 147 Hong Kong DNR; 3 DNR; 4 327 Canada Gutek et al. (1991; Gutek et al. (1991; 263 US 4) 4) Kopelman, et al. (1983; 6) Kopelman et al. (1983; 4) 111 US Self-developed; 7 121 US

Varied professions

Van der Hulst and Guerts (2001)

Author(s) and Publication WIF measure year/sample characteristics Postal service employees Vinokur, Pierce, and Buck (1999) Females who served in air force Wiley (1987) Geurts (2000; 9)

FIW measure Geurts (2000; 6)

Country

751 Netherlands

[Frone et al., [Frone et al., 1992a] and [Frone 1992a] and [Frone 525 US et al., 1992b] et al., 1992b]

MBA students and graduate students enrolled in evening Wiley (1987; DNR) Wiley (1987; DNR) 191 US classes Williams and Alliger (1994) Full-time employees Gutek et al. (1991; Gutek et al. (1991; 41 3) 3) US

Note. Measures of WIF and FIW indicate source of scale and number of items; *indicates that scale used in study was an adaptation. DNR, did not report. 3.3. Coding of variables
When a study gave a range for the sample size, I recorded the lowest number. Apparent typographical errors were clarified with the studys lead author. To observe the need for independence among studies, I averaged multiple time periods, and multiple measures of FIW or WIF or the antecedent, which is consistent with other meta-analyses of workfamily conflict (e.g., Allen et al., 2000). When provided, the reliability (internal consistency) for each measure was recorded. When it was not provided, the average reliability for that variable was inputted (except for measures that were assumed to be perfectly reliable, e.g., sex and number of children). Table 2 details how each of the antecedents were coded or measured in the studies used in the present analysis. In addition, it includes the range of reliability (internal consistency) of the measure used for each antecedent and the average reliability weighted by sample size.

Table 2. Summary of measures used for antecedent variables

Reliability Antecedent Measures Items (weighted Categories reliability)

Work variables
Job involvement Work involvement; Job involvement Number of hours worked; hours worked outside the normal work Hours spent at work week (e.g., working on weekends or at night or traveling for business) Supervisor support; Work support organizational support; coworker support; mentor support Schedule flexibility Schedule (in)flexibility Job stress; role stress; role Job stress conflict; role ambiguity; role overload; Psychological demands 352 .67.89 (.79) 17 335 .73.94 (.83) .52.68 (.58) 1 N/A 25; # 410 .63.90 (.81)

Nonwork variables
Family/nonwork involvement Family involvement; child care involvement; household involvement Hours spent on family; hours Hours of nonwork spent on housework; percentage of time spent parenting Family support; spouse support Family support (instrumental and emotional support) Family stress Role stress; role conflict; role ambiguity; role overload Marital conflict; parental conflict; Family conflict Number of children marital tension; relationship agreement; marital anger Number of children; number of 1 519 49 244 .75.95 (.87) .66.83 (.76) .77.87 (.80) N/A 1 N/A 4, # 211 .60.92 (.82)

24; #

Reliability Antecedent Measures family dependents; number of children living at home; number of children under a particular age; have children or not Age of youngest child Marital status Spousal employment Age of youngest child Marital status (married coded higher) Spousal employment; number of hours spouse works 1 1 1 N/A N/A N/A 35; # 2 23; # Items (weighted Categories reliability)

Demographic/individual variables
Sex Income Coping style and skills Sex (female coded higher) 1 N/A N/A .73.83 (.79) 2 322; # Personal income; family income 1 Active coping style; time management behaviors; personal coping style 533

To test for accuracy in coding, a quarter of the studies were randomly selected to be coded twice. The interrater agreement between the two coders was calculated by dividing the number of data points in agreement by the total number of data points coded. The interrater agreement between the two coders was .99. The high level of agreement between the two coders was likely due to the fact that most of the coding in the present meta-analysis involved merely recording data and possibly applying simple decision rules, and that both coders had prior experience coding studies for meta-analysis.

3.4. Moderators
Several potential moderators were investigated to determine if they explained variation in the effect sizes between studies. Two continuous moderators were considered, the percentage of females in the sample and the percentage of parents in the sample. In addition, one categorical moderator was considered, the use of different coding schemes or measurement of antecedent variables. For example, most studies coded spousal employment as a dichotomous variable (i.e., a spouse was either employed or not); whereas three studies each used a different coding scheme (e.g., trichotomized or number of hours spouse works). When different coding schemes were used, studies that used similar coding schemes were grouped and analyzed. I analyzed

sub-group analyses with five or more studies only. In addition, some antecedents were operationalized differently across studies (see Table 2). For example, some studies used measures of overall job stress, which often combined role ambiguity and role overload, whereas other studies measured only role overload. When a particular antecedent was measured using different constructs, I conducted sub-group analyses for each different construct when there were more than five studies for a given construct. All moderators were coded by recording the sample characteristics of each study (i.e., percentage of females in sample and percentage of parents in sample) and information about the types of measures used in each study (e.g., whether the measure was continuous or categorical). Nearly all studies were conducted in the US or Canada (as shown in Table 1), therefore, the country in which the study was conducted could not be considered as a potential moderator.

3.5. Analysis strategy


For the meta-analysis, I calculated effect sizes with the Hunter and Schmidt (1995) metaanalytic method, which provides the most accurate method of estimating population effect sizes from heterogeneous effect sizes (Field, 2001). To estimate the population effect size, I calculated a frequency weighted mean effect size (corrected for measurement error, when applicable). Before proceeding, I determined whether there were outliers in each analysis by calculating the sample-adjusted meta-analytic deviancy statistic (SAMD) and inspecting the plot of SAMDs for each relationship analyzed (Huffcutt & Arthur, 1995). A study was removed from the analysis for a particular antecedent when a study was identified as an outlier for both WIF and FIW. One antecedent, job insecurity, was not further analyzed because there were too few studies (k < 5) remaining after removing outlier studies. For the remaining relationships considered, I continued the analysis after removing any outliers by assessing the stability of the estimated population effect size by inspecting its modified fail-safe N (MFN) (Huffcutt, Roberts, & Steel, 2004) and by inspecting its 95% confidence interval. The MFN for each effect size indicates the number of additional studies (not additional subjects) with results differing by two standard deviations that would need to be discovered to effect a meaningful change (defined as plus or minus .10 based on Cohens (1998) framework for evaluating effect sizes) in the estimated effect size. In addition, I subjected each analysis to two tests of homogeneity to determine how different the results for each analysis are. First, I examined the percentage of variance explained by artifacts. Possible moderators were investigated if the percentage of variance explained was less than 60%. The use of a lower threshold when effect sizes are only corrected for reliability is consistent with Meyer, Stanley, Herscovitch, and Topolnytsky (2002). Second, I inspected the range of the 90% credibility interval to determine if zero was included, which may indicate that moderators are present (Whitener, 1990).

To test the moderating effects of continuous variables, I fit a series of weighted least-square regression models for each effect size and moderator, weighted by the inverse variance of each effect size ( [Callendar and Osburn, 1988], [Hedges and Olkin, 1985] and [Sanchez-Meca and Marin-Martinez, 1998]), which provides the most accurate results for testing continuous moderators in meta-analysis (Steel & Kammeyer-Mueller, 2002). The correlations were transformed to Fishers Z, and the inverse sampling error variance was estimated by the sample size for each study minus three ( [Callendar and Osburn, 1988] and [Steel and KammeyerMueller, 2002]). To test the significance of moderators, I calculated Z HO, the unstandardized regression coefficient divided by the corrected standard error ( [Hedges, 1994] and [SanchezMeca and Marin-Martinez, 1998]), and determined whether it exceeded the critical value. I included studies that were previously identified as potential outliers because it was possible that the aberrant results were due to differences in sample composition. To test the moderating effects of categorical variables, I split the group into subgroups based on the categorical moderator and conducted a separate meta-analysis on each subgroup. However, in some of the analyses, there was no comparison sub-group because the coding schemes in the remaining studies were all different or because the sub-group contained too few studies. The percent of variance explained by artifacts for the subgroup(s) was compared to the percent of variance explained by artifacts for the overall group analysis to determine if the moderator explained some between-study variation.

4. Results
Meta-analytic results of correlations between work interference with family and family interference with work and the proposed antecedent variables are presented in Table 3. As anticipated, all work variables had a greater impact on WIF than on FIW in the expected direction. For all of the six work variables, the relationship between WIF and the work-related antecedent was of greater magnitude than the relationship between the antecedent and FIW, and the 95% confidence (not credibility) intervals did not significantly overlap. Employees who have higher job involvement or job stress, or spend more time at work have more WIF than FIW, and employees who have less supportive co-workers or supervisors or less flexible schedules have more WIF than FIW. Among work variables, job stress ( = .48) and schedule flexibility ( = .30) were most strongly correlated with WIF. Employees who have more job stress have more WIF, and those with more flexible job schedules have less WIF.

Table FIW

3.

Work,

nonwork,

and

demographic antecedents of WIF and

Work interference with family

Family interference with work (FIW)

(WIF)

Work variables
Job involvement Hours spent at work Continuous only Work support 10 2766 22 9527 18 8092 17 4165 .14

SDO

SD

95% CI MFN SDO

SD

95% CI MFN

.15 .14 +.11/+.18 5

.07 .11 .09 +.03/+.10 9 .01 .07 .05 .01/+.03 75 .01 .07 .05 .02/+.03 60 .12b .17 .11 .09 .15/.09 15 .14 .13 .21/.13 4

.26b .12 .11 +.24/+.27 17 .27b .10 .09 +.25/+.29 18 .19b .09 .07 .22/.16 21 .30 .26 .26 .34/.27 2 .48b .13 .12 +.46/+.49 12 .48b .07 .07 +.46/+.51 15 .65b .16 .16 +.63/+.67 5

Schedule flexibility 8 2620 Job stress 19 7034

.29b .12 .11 +.27/+.31 15 .29b .10 .09 +.26/+.32 8 .40b .18 .18 +.38/+.42 4

Overall stress only 7 3183 Role overload only 10 4402

Nonwork variables
Family/nonwork involvement 9 2741 .02 .08 .05 .07/+.00 18 .02 .09 .07 .06/+.02 13 .01a .07 .04 .04/+.03 26 .11b .10 .07 .14/.07 14 .30b .13 .12 +.27/+.33 5 .32 .12 .11 +.28/+.36 4 .02 .15 .14 .05/+.02 5

Hours of nonwork 10 2875 Continuous only Family support Family stress 9 2764 14 2886 8 2937

.21b .14 .12 +.17/+.24 6 .21b .14 .13 +.17/+.24 5 .17b .13 .11 .21/.14 10

.47b .18 .18 +.44/+.50 3 .49 .20 .19 +.46/+.52 2 .32b .16 .15 +.28/+.36 4 .16b .12 .11 +.14/+.17 19 .17b .12 .11 +.13/+.21 6 .15b .12 .11 +.13/+.18 11 .22b .08 .07 .24/.19 16

Overall stress only 5 2008 Family conflict Number of children Number of children Number living at homec Age of youngest child 8 1674

.35b .10 .07 +.31/+.40 9 .12 .11 +.07/+.11 21 .12 .11 +.01/+.09 6

27 10,467 .09 8 2557 15 6700 9 7303 .05

.08b .08 .06 +.06/+.11 28 .17a,b .04 .02 .20/.15 N/Ad

Work interference with family

Family interference with work (FIW)

(WIF)

Marital status Spousal employment Dichotomous coding 14 9378 9 4358 6 3413 .03

SDO

SD

95% CI MFN SDO .05

SD

95% CI MFN

.10 .09 +.01/+.05 14

.08 .07 .07/.03 25

.01a,b .05 .00 .02/+.04 189 .03 .10 .09 +.00/+.06 10 .01a .05 .03 .04/+.03 299 .02 .09 .08 .02/+.05 7

Demographic variables
Sex Income Coping style and skills 27 18,125 .03 .11 .10 .04/.01 24 13 7046 6 2002 .10b .08 .07 +.08/+.12 22 .12a,b .03 .00 .16/.08 14 .06 .12 .11 +.04/+.07 21 .00 .08 .07 .03/+.02 21 .15b .11 .10 .19/.10 5

Note. WIF, Work interference with


family; FIW, Family interference with work;

k, number of independent

samples included in each analysis; N, combined sample sizes of studies included in each analysis; , weighted average corrected correlation; SDO, observed standard deviation of corrected correlation; SD, estimated true/population standard deviation of corrected correlation; 95% CI, 95% confidence interval, MFN, modified fail-safe N (indicates the number of additional studies needed to cause a meaningful change in the estimated population effect size). a

More than 60% of the observed variance is accounted for by sampling error. b Zero is not included in the 90% credibility interval. c Number children living at home

includes studies that asked individuals to report how many children were living at home, how many children were living at home under a particular age, or how many family dependents they had. In contrast, when studies asked participants to report how many children they had, the number could include older children, adult children, or other nondependents. d The Modified Fail-Safe

was

negative, indicating that no number of studies could exact a meaningful change in the estimated population correlation.

Contrary to expectations, the correlations between nonwork variables and FIW (as compared to WIF) did not have consistently stronger relationships in the expected direction. Several of the nonwork variables showed similar relationships to WIF and FIW. The 95% confidence intervals for the estimated population effect sizes for family/nonwork involvement, family support, family conflict, age of youngest child, and spousal employment overlapped in their relationships to WIF and FIW. The remaining four nonwork variables, hours of nonwork, family stress, number of children, and marital status, demonstrated a pattern that was consistent with expectations. The more hours spent on family, housework, childcare or other nonwork-related activities, the more FIW experienced ( = .21), but not significantly more WIF, as indicated by 95% confidence interval that included zero ( = .02, 95% CI: .06/+.02). In addition, employees who

experienced more family-related stress experienced more FIW ( = .47, 95% CI: +.44/+.50) than WIF ( = .30, 95% CI: +.27/+.33). Employees with more children or who were single had more FIW ( = .16, 95% CI: +.14/+.17, and = .05, 95% CI: .07/.03; respectively) than WIF ( = .09, 95% CI: +.07/+.11, and = .03, 95% CI: +.01/+.05; respectively). Among all nonwork variables, family stress ( = .47) and family conflict ( = .32) were most strongly related to FIW. Employees who had more stress and more conflict at home had more family interference with work. Demographic and individual antecedent variables were expected to have equivalent effects on WIF and FIW. To determine the extent to which the variables have similar relationships to WIF and FIW, I inspected the 95% confidence intervals of the estimated population effect sizes for each demographic variable and its relationship to WIF and FIW. Of the demographic and individual variables, only one, coping style and skills, tended to have a similar relationship to both WIF and FIW. Having a positive coping style or having better coping skills seems to provide some protection from WIF and FIW ( = .12 and = .15, respectively). The other two demographic and individual variables, sex and income, tended to vary in their relationship to WIF and FIW, as indicated by nonoverlapping 95% confidence intervals. Male employees tended to have slightly more WIF ( = .03, 95% CI: .04/.01) and female employees tended to have more FIW ( = .06, 95% CI: +.04/+.07), although the differences between sexes and the difference between sexs relationship to WIF and FIW were small. There were also significant differences between the relationship of income to WIF and FIW. Employees with higher incomes had more WIF ( = .10, 95% CI: +.08/+.12), whereas income was not significantly related to FIW ( = .00, 95% CI: .03/+.02). Next, I considered whether a search for potential moderators was warranted. For nearly all antecedents, there seemed to be significant variation between studies used in the metaanalysis. In fact, the two tests of homogeneity used in this analysis indicated homogeneity in their relationship to WIF or FIW for only two antecedents, age of youngest child and coping style/skills. However, the 90% credibility intervals suggested homogeneity for most of the relationships. Because the results for the homogeneity tests failed to be consistent for the majority of analyses, I proceeded to conduct the proposed moderator analyses. First, I considered the categorical moderator, differences in coding scheme or measurement (as shown in Table 3). Six variables that did not meet the two tests of homogeneity had differences in coding schemes between studies, hours spent at work, job stress, hours spent on nonwork, family stress, number of children, and spousal employment. The estimated population effect sizes for the overall group analysis and the sub-group analysis of those that measured the variable continuously do not differ greatly, however, for nearly all analyses, the sub-group analyses explain more of the variance by artifacts or have more stable effect size estimates.

When considering only studies that measured time at work continuously, employees who spend more time at work experience slightly more WIF but the same amount of FIW ( = .27 and = .01, respectively). On the other hand, employees who spend more time in family or household duties and activities experience less WIF although the same amount of FIW ( = .02 and = .21, respectively). Studies that examined the relationship between job and family stress and WIF and FIW differed in their measurement of job and family stress. Some studies used overall measures of job or family stress, some used more specific measures of job stress, such as role overload or role ambiguity. For job stress, the sub-group analysis of those studies that used overall measures of job stress explained more variation than did the overall analysis, suggesting the sub-group analysis may reflect more accurate estimates of the relationship between job stress and WIF and FIW ( = .48 and = .29, respectively). When considering only studies that examined role overload, the estimated effect sizes in regard to WIF and FIW tended to be greater but less stable and homogenous ( = .65 and = .40, respectively). For family stress, the only subgroup analysis that could be conducted was on those studies that used overall measures of family stress. The estimated effect sizes for this sub-group analysis did not significantly differ from or improve upon those of the overall analysis. Lastly, I considered whether differences in coding might account for between-study variation for two other family domain variables, number of children and spousal employment. Some studies asked employees to indicate how many children (with no restrictions) they had; whereas other asked employees to indicate how many children they had living at home or under a particular age. The coding scheme for number of children explained significant between-study variation, although the 95% confidence intervals for the estimated effect sizes tended to overlap across each coding scheme for WIF ( = .05, 95% CI: +.01/+.09, and = .08, 95% CI: +.06/+.11; respectively) or FIW ( = .17, 95% CI: +.13/+.21, and = .15, 95% CI: +.13/+.18). Most studies dichotomized spousal employment (i.e., either the spouse works or not); the remaining three studies each used a different coding scheme, and therefore were not considered in a sub-group analysis. Compared to the overall analysis, the sub-group analysis for studies that dichotomized spousal employment accounted for less between-study variation and had larger 90% credibility intervals. This suggests that the overall analysis provides a more accurate estimate of the relationship between spousal employment and WIF and FIW despite the fact that there were differences in coding schemes. In summary, for most of the relationships considered, differences in coding schemes tended to explain some of the variance between studies or provide more stable estimated effect sizes over the overall analyses. Next, each of the two proposed continuous moderators were fitted into a separate weighted least square equation for each relationship considered. The antecedent, coping style and skills,

was excluded for the proposed moderator, percent of sample with children, due to too few studies (k < 4) that provided data. I excluded other antecedents because their interpretation lacked conceptual meaning (i.e., for percent of parents in sample, number of children and age of youngest child, and, for percent female in sample, sex was excluded). The results of the regression models for percent of sample with children are included in Table 4, and the results of the regression models for percent female in sample are included in Table 5.

Table 4. Percent of study sample with children as a moderator of effect size WIF

FIW

r Work variables
Job involvement Hours spent at work Work support Schedule flexibility Job stress

ZHO

ZHO

10 .21 .20 1.20 .00 .23 0.95 21 .12 .08 1.29 .05 .10* 2.13 14 .35 .31 1.14 .39 .30 1.79 5 .72 .72* 4.73 .18 .55 1.30 13 .01 .29* 3.21 .13 .43* 2.88

Nonwork variables
Family/nonwork involvement 9 .19 .10 0.00 .35 .24 0.95 Hours of nonwork Family support Family stress Family conflict Number of childrena Age of youngest child a Marital status Spousal employment 4 .02 .34 0.41 .12 .22 0.98 11 .26 .26 1.16 .23 .33 1.70 7 .31 .07 0.00 .24 .01 0.06 7 .08 .01 0.04 .48 .72* 3.89

13 .17 .45* 2.54 .35 .78* 3.50 10 .59 .52 0.88 .02 .36 1.58

Demographic variables
Sex Income Coping style and skillsa 23 .06 .65* 4.85 .24 .46* 19.52 14 .05 .09 0.00 .02 .03 0.00

Note.

r,

Pearson

correlation;

standardized

regression

coefficient

WLS regression; ZHO, test of null hypothesis that = 0; * indicates 95% confidence that does not equal zero. a Both antecedents concerned with

children, number of children and age of youngest child, were excluded from analysis because considering the percentage of parents as a moderator of their relationships to WIF and FIW lacked conceptual meaning. Only three studies that considered coping style and skills provided data on the percent with children in the study; therefore, excluded. Table 5. Percent female in study sample as a moderator of effect size WIF these analyses were

FIW

r Work variables
Job involvement Hours spent at work Work support Schedule flexibility Job stress

ZHO

ZHO

13 .66 .71* 6.47 .33 .35* 3.24 27 .10 .03 0.28 .14 .12 0.00 18 .20 .10 0.00 .06 .24 1.39 8 .10 .39* 2.83 .63 .78* 5.19 20 .04 .08 0.00 .32 .22* 2.80

Nonwork variables
Family/nonwork involvement 11 .33 .53 1.91 .35 .23* 2.02 Hours of nonwork Family support 10 .12 .39 1.48 .02 .21* 2.42 16 .15 .30 1.35 .56 .40* 3.03

WIF

FIW

r
Family stress Family conflict Number of children Age of youngest child Marital status Spousal employment

ZHO

ZHO

9 .45 .56* 3.67 .62 .78* 11.44 8 .22 .30 1.28 .15 .13 1.12 31 .22 .34* 2.09 .24 .27* 2.37 12 .17 .14 0.00 .03 .22* 2.19 15 .02 .32* 2.49 .18 .57* 3.38 11 .16 .20 0.00 .26 .15 0.00

Demographic variables
Sexa Income Coping style and skills 14 .17 .16 0.00 .30 .43 1.57 6 .11 .12 0.00 .66 .60* 2.36

Note.

r,

Pearson

correlation;

standardized

regression

coefficient

WLS regression; ZHO, test of null hypothesis that = 0; * indicates 95% confidence that does not equal zero. a Sex was excluded from analysis because considering the percentage of females as a moderator of its relationships to WIF and FIW lacked conceptual meaning. The percent of parents in the sample related significantly to the study effect size for over 32% of the relationships considered, suggesting that differences in the composition of the sample explains between-study variation for some relationships. In particular, the percent of parents in the sample seems to affect the relationship between job stress and workfamily interference. The more parents in the sample, the stronger the positive relationship between job stress and WIF and FIW. While sex had a very small direct effect on WIF or FIW, the percent of parents in the sample does moderate this relationship. Namely, when there are more parents in the sample, there is a greater sex difference in the experience of WIF and FIW, such that mothers experience more WIF and FIW than fathers. When there are fewer parents in the sample, men

tend to experience more WIF and FIW. Lastly, while there tended to be no difference between married and single employees in their experience of WIF and FIW overall, marital status is negatively related to WIF and FIW as the number of parents in the sample increases. This suggests that single parents have more WIF and FIW than parents who are married; whereas married and single employees without children tend to have similar levels of WIF and FIW. The percent of female employees in the sample related significantly to the study effect size for more than half of the relationships considered, suggesting that differences in the sex composition of the sample explains between-study variation for some relationships. For example, having a higher percentage of women in a sample associated with a weaker positive relationship between job involvement and WIF and FIW. Job involvement seems to relate more positively to WIF and FIW for men than for women. Conversely, family involvement related more positively to WIF and FIW for women than for men (although only approaching statistical significance at the level of .05 for WIF). For men (as compared to women), being more highly involved in their jobs is linked to more interference whereas, for women (as compared to men), being more highly involved in their family lives is linked to more interference. In addition, a higher percentage of females in a sample is negatively related to the study effect size for schedule flexibility and WIF and FIW. Flexible schedules appear to provide more of a protective benefit for women than for men. However, family stress and family conflict are more positively related to WIF and FIW for men than for women. Stress and conflict in the family domain is linked to more interference for men as compared to women. When more women are represented in the sample, the employees number of children is less positively related to WIF and FIW. For example, the weighted mean average correlation between number of children and WIF is .15 for all male samples and .02 for all female samples; and, for FIW, is .21 and .08, respectively. The percentage of female employees in the sample also moderates the relationship between marital status and WIF and FIW. For men, more so than for women, being married is associated with more WIF and FIW. For women, marital status had a near-zero relationship to WIF and FIW, suggesting that being married or single has little effect on female employees experience of WIF and FIW. In particular, the percentage of females in the sample affected the relationship between the antecedents and FIW (as compared to WIF), as the percentage of females in the sample was a significant moderator in 11 of the 16 (69%) relationships considered. Lastly, I considered the relation between WIF and FIW. The weighted average corrected correlation is .48 (SDO = .11; SD = . 10; 95% CI: +.46/+.49), which is the result of cumulating the results of 47 studies with a total sample size of 13,384 (after eliminating 9 studies identified as outliers). In all studies, WIF and FIW related positively. More interference in one domain tends to associate with more interference in the other. In fact, only one antecedent, job stress,

related nearly as strongly to WIF ( = .48), and only one antecedent, family stress, is nearly as strongly related to FIW ( = .47) as WIF and FIW related to each other.

5. Discussion
The results of this meta-analytic review support the differentiation between work interference with family and family interference with work. Employees seem to differentiate between the source, or direction, of interference, and the two types of interference appear to have different antecedents. The results of the analysis partially support the pattern of relationships expected: work-related antecedents tend to associate with more work-related interference than nonwork interference. Nonwork-related antecedents tend to relate to more family interference with work than work interference with family, although the differences were not always statistically significant. However, of all of the antecedents, job stress, family stress, and family conflict have among the strongest associations with both WIF and FIW, suggesting that while there is differentiation, some work and family factors can have simultaneously disruptive effects on employees work and family lives. Surprisingly, the two demographic variables, sex and income, which have often been proposed in the literature as antecedents of WIF and FIW, had relatively low relationships to WIF and FIW. For example, sex, which has been proposed as an antecedent in dozens of studies, had a near zero relationship to WIF and a weak, positive relationship to FIW. Contrary to hypotheses in many studies, the present analysis suggests that overall men and women have similar levels of WIF and FIW. This finding coincides with other research that has reported no sex difference in the experience or perception of occupational stress (Martocchio & OLeary, 1989). The only individual variable considered in the analysis, coping style and skills, seemed to offer some benefit to employees. Those with better time management skills or a better coping style tended to have less WIF and FIW. While demographic variables tended to be weak predictors of WIF and FIW, they did tend to have indirect effects on WIF and FIW. The percentage of women or parents in the sample explained between-study variance in more relationships than would be expected by chance. This coincides with recent theory that supports the use of social categories as moderators in the workfamily literature (Voydanoff, 2002). In general, being male appears to exacerbate any negative effects of family domain antecedents, such as family stress, family conflict, number of children, and marital status, related to workfamily conflict. Paradoxically, females tend to enjoy greater protective benefits from those antecedents, such as flexible work schedules, and, to some extent, supportive families, that lessen the experience of interference. While not as consistently as the percentage of females in the sample, the percentage of parents in the sample also explained some differences in results across studies. For employees with children

as compared to those without children, having more job stress, being single, and being is related to more workfamily conflict. Overall, the results provide partial support for the hypotheses of the study. In view of that, some exceptions and other surprising results deserve note. First, for WIF and FIW, only one antecedent each was as correlated with WIF and FIW as they were with each other. While the differential effects of antecedents provides support for discriminating between the two constructs, the strong positive relationship between them deserves further study. Perhaps the perception of interference between domains can be explained by a common third variable, such as being high in negative affect or having expectations of separate domains. Second, contrary to expectations, nonwork domain variables did not have a consistently stronger relationship to FIW than to WIF. Nonwork domain variables that have been referred to as family demands (i.e., number of children, age of youngest child, marital status, and spousal employment) were nearly as related to FIW as to WIF. Perhaps this speaks to the asymmetric permeability of domains, such that family demands cause family life to interfere with work and for work to interfere with the relatively greater family demands. Lastly, family involvement had a near-zero correlation with FIW (and WIF), rather than being positively related to FIW as expected. Employees who had higher family involvement experience the same amount of FIW (and WIF) as those who were less involved with their families.

5.1. Theoretical implications


Several theoretical models can glean support from these findings. Overall, the results provide support for conflict theory (Greenhaus & Beutell, 1985). They highlight the potential incompatibility of work and family roles and ensuing conflict from having multiple roles, at least for some people. For example, the present analysis found that employees who experience more stress on the job are more likely to experience interference from their work into their family lives. Likewise, employees who experience stress at home are more likely to experience interference from their family lives into their work day. Furthermore, the results suggest that both spillover and congruence are apparent linking mechanisms between the work and family domains (Edwards & Rothbard, 2000). Spillover as a linking mechanism occurs when stress or strain from one domain surface in another domain. Congruence as a linking mechanism between work and family domains occurs when a third variable links the domains of work and family domains by having a congruent effect on both (Edwards & Rothbard, 2000). Support for spillover as a linking mechanism can be seen in the positive relationship between job stress and WIF and between family stress and conflict and FIW. Stress from one domain is interfering with the other domain. While the results suggest that negative spillover can occur from one domain to another, the results also support the notion that

positive spillover can occur. For example, employees who are employed in more supportive workplaces or who have more supportive families tend to experience less workfamily conflict. Support for congruence as a linking mechanism is found in the similar relationship employees coping style and skills to both WIF and FIW. Employees who have better time management skills and coping behaviors experience less WIF and FIW. The results of the meta-analysis provide some support for the rational view, which predicts that the more time one spends in a role, or the more one specializes or is involved in a role, the more he or she will perceive interference in the secondary role from the participation in the primary role (Pleck, 1977). Consistent with this view, the number of hours spent on work was more positively related to WIF than to FIW, and the number of hours spent on nonwork was more positively related to FIW than to WIF. Similarly, employees with higher job involvement had more WIF than FIW. However, inconsistent with the predictions of this view, employees with more family or nonwork involvement did not tend to have more WIF or FIW. Alternatively, the results do offer some support for the sex-role hypothesis, which proposes that sex or sex roles moderate the relationship between role involvement and psychological distress (Voydanoff, 2002). As mentioned above, the present analysis found an inconsistent relationship between role involvement in a given role and WIF and FIW, which is surprising given the frequency with which these relationships have been explored in the literature. Employees sex does seem to moderate the relationship between job and family involvement and WIF and FIW. For three out of the four effect sizes, having more females in the sample related to the strength of the relationship between role involvement and workfamily interference. Namely, job involvement seems to relate more positively to WIF and FIW for men than for women. In addition, when more of a studys participants were parents, there was a greater sex difference in the experience of WIF and FIW, such that mothers experience more WIF and FIW than fathers. When there were fewer parents in the sample, men tended to experience more WIF and FIW. Perhaps because women tended to take on greater responsibilities for childcare, mothers experience more distress from the greater workload but only when they are also highly involved in their work. In summary, the results provide support for multiple theoretical models. This suggests that no single model can fully explain how employees experience the intersection between their work and nonwork domains. Future theorizing should work toward creating an integrative model that more fully explains the complexity suggested by the results presented here.

5.2. Future research

The present study also offers some suggestions for future research. First, the continued use of bidirectional measures is supported. The present results provide support for the discriminant validity of these constructs. Second, the relative importance of these antecedents may guide future research aimed at better understanding the causes and prevention of workfamily conflict. Factors such as job stress and family conflict, which were strong predictors of both WIF and FIW, are important topics for future research. Lastly, differences in the composition of the study sample (i.e., percentage of females and percentage of parents) and the lack of homogeneity in many of the analyses suggest that researchers should be thoughtful about choosing their sample. In cumulating these studies, differences between sampling strategies became apparent. For example, some studies only considered parents (e.g., Duxbury, Higgins, & Lee, 1994), some only included married participants (e.g., Beutell & Witting-Berman, 1999), and some did not employ restrictions (e.g., Aryee, Fields, & Luk, 1999). Future research should further investigate whether differences in sampling strategies explain differences in results across studies. In addition, in reviewing the primary research that considers antecedents of workfamily interference, several gaps in the literature became apparent. Few studies explicitly considered the difference between caring for adults and children. This gap in the literature has been noted by others: [H]uman resource and organizational behavior scholars often overlook how caring for a parent differs from caring for a child (Kossek et al., 1999, p. 114). In addition, most studies began with the assumption that having multiple roles would necessarily lead to distress. However, research in other areas supports the idea that individuals who hold multiple roles reap benefits from doing so. In the management literature, there should be more recognition of the benefits (rather than detriments) that arise from the participation in multiple roles (e.g., Thoits, 1992). Research should discard the overly simplistic notion that distress must be found at the intersection of work and family, and instead focus on determining the conditions that distinguish when multiple roles leads to distress and when multiple roles leads to increased fulfillment ( [Barnett and Hyde, 2001] and [Greenhaus and Parasuraman, 1999]). Lastly, few studies considered individual variables, such as personality or skill level. It seems as though work family conflict research, for the most part, has not focused on individual differences in success at handling workfamily conflict (Baltes & Dickson, 2001, p. 53). Future research should consider whether other individual variables are useful at explaining differences in outcomes in the workfamily literature.

5.3. Practical implications


In addition to providing guidance to workfamily researchers, the results of the analysis offer some practical implications. Namely, employers who seek to reduce their employees perceived stress from competing demands should focus on reducing job-related stress. Job-related stress,

such as role conflict, ambiguity, and overload, seems to be spilling over into employees lives away from work. Furthermore, while outside of the work domain, employers may consider offering guidance to their employees on improving their relationships with their spouses and children to reduce family conflict. Employees who reported more marital strife or more conflicts with their children had more interference between their work and family lives. Perhaps more employers should offer training to their employees on managing family conflict, although the benefits of this type of training to employers are not well established. Clearly, employees are not checking their family concerns at the workplace door, suggesting that employers may have an interest in helping employees with these concerns. The analysis also suggests that employers can effectively reduce the experience of workfamily interference among their employees. Namely, employees who had more flexible schedules or who had more supportive coworkers or supervisors reported less WIF and FIW. This suggests that some employer interventions may be beneficial at reducing distress for employees. While the present analysis does provide guidance for future research and has some practical implications, the analysis is not without its limitations. One of the primary limitations is that metaanalysis cannot partial out the effects of other variables on the relationships considered. For example, spousal employment relate more strongly to WIF and FIW when the presence of young children in the home is simultaneously considered. In addition, some of the effect sizes were cumulated from a small number of studies. Field (2001) warned that estimates and significance tests from meta-analytic studies containing less than 30 samples should be interpreted very cautiously (p. 179). However, the two estimates of effect size stability generally provide support for the relative stability of the effect sizes. As with all meta-analyses, more stable and accurate estimates may be obtained with the addition of more primary research. Lastly, as mentioned previously, the effect size estimates have significant between-study variation. The present analyses suggest that sample composition is one source of betweenstudy variation, and future research should seek to identify other sources of variation between studies of workfamily interference. This study provides support for the bidirectional nature of workfamily conflict, and it suggests that researchers should employ measures that distinguish between WIF and FIW. Furthermore, it supports the notion that WIF and FIW have unique antecedents, and therefore, may require different interventions or solutions to prevent or reduce their experience. Lastly, the analysis suggests that demographic variables, such as sex and marital status, are alone poor predictors of workfamily conflict. Researchers are advised to examine more finely-grained variables that may more fully capture employees likelihood of experiencing workfamily conflict.

This study provides support for the bidirectional nature of workfamily conflict, and it suggests that researchers should employ measures that distinguish between WIF and FIW. Furthermore, it supports the notion that WIF and FIW have unique antecedents, and therefore, may require different interventions or solutions to prevent or reduce their experience. Lastly, the analysis suggests that demographic variables, such as sex and marital status, are alone poor predictors of workfamily conflict. Researchers are advised to examine more finely-grained variables that may more fully capture employees likelihood of experiencing workfamily conflict.

You might also like