Correspondence should be addressed to Professor Ulla Kinnunen, Department of Psychology, University of Tampere, 33014 Tampere, Finland (e-mail: firstname.lastname@example.org).
This study assessed longitudinal individual and crossover relationships between work-family conflict and well-being in the domains of work (job satisfaction) and family (parental distress) in a sample of 239 dual-earner couples. The results revealed only longitudinal individual effects over a 1-year period. First, high family-to-work conflict (WFC) at Time 1 was related to a high level of work-to-family conflict (WFC) 1 year later in both partners. Second, the wife's high level of FWC was related to her decreased job satisfaction 1 year later. Thus, the longitudinal effects identified supported normal causality, that is, work-family conflict led to poor well-being outcomes or increased perceived work-family conflict later on. Longitudinal crossover effects from one partner to another were not observed within a 1-year perspective.
In today's western society the topic of how to reconcile the competing demands of work and family life continues to be an increasing problem, which often results in work-family conflict for both working men and women (e.g. Aryee, Srinivas, & Tan, 2005; Eagle, Miles, & Icenogle, 1997; Kinnunen & Mauno, 1998). Work-family conflict has been defined as ‘a form of inter-role conflict in which the role pressures from the work and family domains are mutually incompatible in some respect’ (Greenhaus & Beutell, 1985, p. 77). This inter-role conflict functions bidirectionally, that is, work may interfere with family (WFC) and family may interfere with work (FWC). It has been shown that work-family conflict is associated with various negative individual (e.g. strain symptoms), family (e.g. family stress) and organizational level (e.g. decreased organizational commitment) outcomes (for reviews, see Allen, Herst, Bruck, & Sutton, 2000; Kossek & Ozeki, 1998; Kinnunen & Mauno, 2008).
However, less attention has been paid to possible detrimental outcomes of employees' work-family conflict for their partners. According to a recent review, only 9% of work-family research has focused on this issue (Casper, Eby, Bordeaux, Lockwood, & Lambert, 2007), which is highly relevant today when the dual-earner situation is the rule rather than the exception in western families. Thus, in these dual-earner families how both partners manage the interface between their work and personal lives matters, and how partners succeed in this task determines the extent to which they are negatively affected by each other's experiences within the work and family domains. Extending work-family interface research in a direction paying more attention to the family context has been urgently called for (e.g. Barnett, 1998; Casper et al., 2007; Geurts & Demerouti, 2003). Our study aims to respond to this call by focusing on both partners' experiences of work-family conflict and well-being in the domains of work and family.
Our main objectives are threefold. First, we examine at the individual level using an individual as the unit of analysis whether work-family conflict (i.e. WFC and FWC) has effects on domain-specific well-being (i.e. job satisfaction and parental distress) in 1 year. Second, we examine using a couple as the unit of analysis whether one partner's WFC may have longitudinal effects on the other partner's well-being. Finally, because we utilize longitudinal data we are also able to explore reversed causality, that is, whether own impaired well-being predicts own work-family conflict or partner's work-family conflict.
In sum, our study extends the existing work-family research in several ways. First, the most important extension is that we approach the very topical issue of work-family conflict from both longitudinal and crossover perspectives. These perspectives have seldom been combined, and to the best of our knowledge there is only one study in the context of work-family conflict where this has been done using 1-year longitudinal data (see Hammer, Cullen, Neal, Sinclair, & Shafiro, 2005). In the study of Hammer et al. (2005), the focus was on the relation of work-family conflict with depression. Our study extends this perspective into well-being experienced in the domains of work (job satisfaction) and family (parental distress), which are the most important domains when work-family conflict is considered (e.g. Allen et al., 2000). Second, the present study contributes to the existing literature by taking into account the possibility of reversed causality which has recently gained more attention in the context of work-family conflict (e.g. Demerouti, Bakker, & Bulters, 2004; Kinnunen, Geurts, & Mauno, 2004; Rantanen, Kinnunen, Feldt, & Pulkkinen, 2008; Steinmetz, Frese, & Schmidt, 2008), but which has not yet been studied in the context of couples.
Theoretical models of the outcomes of work-family conflict
Frone and his colleagues have presented two interesting models concerning the outcomes of work-family conflict. First, Frone, Russell, and Cooper (1992) tested a ‘Model of Work-Family Interface’ which they later (Frone, Yardley, & Markel, 1997) elaborated as an ‘Integrative Model of the Work-Family Interface’. These models have gained considerable support in cross-sectional studies (see Ford, Heinen, & Langkamer, 2007; Frone, 2003, for reviews). However, there is lack of studies testing the models or their different parts longitudinally and among couples.
According to the earlier ‘Model of Work-Family Interface’, WFC and FWC have unique role-related outcomes. The outcomes of WFC reside in the family domain (e.g. parental distress) and the outcomes of FWC reside in the work domain (e.g. job satisfaction). In other words, each direction of conflict should be related to outcomes in the domain receiving the conflict. Frone and his colleagues (1992) also claimed that WFC and FWC have both a direct and a reciprocal relation to each other in such a way that when WFC (or FWC) is experienced it increases the likelihood that family (work) responsibilities will remain unfulfilled, and these unfulfilled family (work) duties, in turn, begin to impede work (family) performance, leading to experience of FWC (WFC).
The later ‘Integrative Model of Work-Family Interface’ proposed that the reciprocal relation between WFC and FWC might alternatively be indirect, so that the two directions of work-family conflict are related to each other through work and family distress. In addition to being respectively domain-specific outcomes of FWC and WFC, work distress and family distress are also domain-specific antecedents of these experiences. More specifically, work distress is proposed to be an antecedent of WFC and family distress an antecedent of FWC, that is, each direction of conflict should be related to antecedents in the domain from which the conflict has originated. Thus, the aim of the present study is to shed further light on the views described above using longitudinal data collected among Finnish working couples.
Although the normal causal approach has received support, its premises can be questioned. It is possible that if an individual suffers from psychological strain, he or she has consequently fewer resources to cope with simultaneous work- and family-related responsibilities. Therefore, psychological strain accompanied by impaired daily functioning may increase the incidence of work-family conflict. This picture is in line with the view that resources are finite and in the long run they may diminish because of changes in well-being and health (Edwards & Rothbard, 2000). Accordingly, five longitudinal studies support this reversed causal approach (i.e. strain causes work-family conflict). First, among business travellers pre- and mid-trip burnout preceded work-family conflict (both directions were included but combined into one measure) during and after the trip (the three measurement points were 1 week before, during and a few days after the trip; Westman, Etzion, & Gortler, 2004). Second, among married soldiers with children job dissatisfaction preceded WFC across 3 or 4 months (Britt & Dawson, 2005). Third, in men, but not in women, marital dissatisfaction, parental distress, and psychological as well as physical stress symptoms were antecedents of WFC 1 year later (Kinnunen et al., 2004). Fourth, family dissatisfaction predicted FWC in men, but not in women, and job dissatisfaction predicted WFC among both genders 1 year later (Huang, Hammer, Neal, & Parrin, 2004). In the most recent longitudinal study with a 1-year time lag by Steinmetz et al. (2008) reversed causality also gained support; namely, depression served as an antecedent of WFC.
In addition to these two approaches, there is the reciprocity approach. According to this, work-family conflict leads to increased strain, which in turn gives rise to increased work-family conflict. Thus, adopting the terms of the conservation of resources theory (Hobfoll, 1989; see also Grandey & Cropanzano, 1999) the question is of loss spirals, in which negative experiences trigger other negative experiences over time. Such reciprocal associations between work-family conflict and psychological strain have been found in three longitudinal studies. Over a 6- and 12-week period WFC was both an antecedent and an outcome of job exhaustion (Demerouti, Bakker, & Schaufeli, 2005). Over 3 months WFC preceded job exhaustion, dysphoric mood, and marital dissatisfaction, and simultaneously WFC was an outcome of job exhaustion and a high amount of conflict with family members, and FWC was an outcome of marital dissatisfaction (Leiter & Durup, 1996). In addition, FWC was an antecedent, and WFC an outcome of psychological stress symptoms 6 months later (Kelloway, Gottlieb, & Barham, 1999).
Taken as a whole, these longitudinal findings seem to give some support for the two models of Frone and his colleagues (Frone et al., 1992; Frone, Yardley et al., 1997). First, WFC was an outcome of job exhaustion and antecedent of marital dissatisfaction over 3 months (Leiter & Durup, 1996). Second, WFC predicted parental distress 1 year later in women (Kinnunen et al., 2004), and third, family dissatisfaction predicted FWC in men (Huang et al., 2004). Fourth, job dissatisfaction predicted WFC both across 3 or 4 months (Britt & Dawson, 2005) and 1 year (Huang et al., 2004). Although these findings are in line with the domain-specific antecedents and outcomes presented in the Frone models, there are simultaneously several findings against the models. However, it is worth noticing that the studies were not originally planned to test the Frone models.
In addition, the observed longitudinal results are inconsistent in several regards. First of all, not all studies have taken the possibility of reversed causality into account (e.g. Frone, Russell et al., 1997; Grant-Vallone & Donaldson, 2001). Second, such issues as different follow-up periods, different measures, or different study designs may play a role. Of these issues, the choice of the follow-up period is perhaps the most crucial question. Theoretically, it is very difficult to define an optimal follow-up period within which, for example, the negative outcomes of work-family conflict should emerge (see De Lange, Taris, Kompier, Houtman, & Bongers, 2004). Based on the previous findings we know that cross-lagged effects between work-family conflict and well-being have been found across 6 weeks (Demerouti et al., 2004) and up to 4 years (Frone, Russell et al., 1997).
The follow-up period is 1 year in our study, since the possible outcomes - job satisfaction (e.g. Dormann & Zapf, 2001) and parental distress (e.g. Seginer, Vermulst, & Gerris, 2002) - have been shown to be fairly stable. Also, both WFC and FWC has turned out to be relatively stable across different time periods (e.g. Kelloway et al., 1999; Kinnunen et al., 2004; Rantanen et al., 2008). Therefore, it is necessary to have an extended period of time to allow for change to occur by critical incidents within work and family domains. However, satisfaction and distress belong to short-term outcomes in the categorization of outcomes related to job insecurity (Sverke, Hellgren, & Näswall, 2002) compared with outcomes related to health and performance, which were under the label of long-term consequences. Consequently, we consider the 1-year time lag sufficient to reveal meaningful variation in the outcomes studied; longer time lags may add the risk to underestimate the true causal effect. Altogether, we were not interested either in very short-term changes (e.g. in emotions) which are captured better by the daily diary studies or in very long-term changes (e.g. in health) which would need a still longer follow-up period.
Crossover between partners
To differentiate the partner effects from the individual level effects, the term crossover has been introduced (Westman, 2001). The crossover process occurs when a stressor or psychological strain experienced by one person affects the level of strain of another person. This process has also been termed as emotional transmission (e.g. Jones & Fletcher, 1993) or contagion (Bolger, DeLongis, Kessler, & Wethington, 1989). Partner crossover may be either direct or indirect (Westman, 2001). A direct empathic crossover is said to occur when stressors/strains experienced by one partner directly affect the stressors/strains experienced by the other partner, whereas indirect crossover occurs via mediating processes, for example, via interpersonal conflict (Westman & Etzion, 1995). In the present study, we focus on work-family conflict in terms of direct partner crossover effects from a longitudinal perspective.
Work-family conflict has been examined from the perspective of crossover effects only in a few and mainly cross-sectional studies. Among the first studies, based on hierarchical linear regression analyses using samples matched by the partner, that conducted by Hammer, Allen, and Grigsby (1997) showed that one partner's experienced work-family conflict (a global measure without directionality) had a crossover effect on the experience of work-family conflict in the other partner. Hammer, Bauer, and Grandey (2003) have also shown that work-family conflict had a crossover effect on work-related withdrawal behaviours. Husband's FWC was related to lateness for work reported by wife, and wife's FWC was related to the number of interruptions while at work and absences from work (due to family/personal-related issues) reported by husband.
In a more recent study, which concentrated on work-to-relationship conflict (measured with a single-item concerning the frequency of how often the demands of one's work interfere with his/her partner relationship), Matthews, Priore, Acitelli, and Barnes-Farrell (2006) showed using structural equation modelling (SEM) that wife's work-to-relationship conflict was positively related to husband's reports of relationship tension. However, if the husband reported higher levels of work-to-relationship conflict, the wife reported lower levels of relationship tension. In addition, one partner's relationship tension had a direct negative crossover effect on the other partner's relationship satisfaction.
The only longitudinal study concerning the crossover effects of work-family conflict by Hammer et al. (2005) showed that work-family crossover effects were linked to depression. They found - using hierarchical regression analysis in a sample matched by the partner - that only positive spillover (i.e. functioning in one domain having a positive effect on functioning in another domain) had a longitudinal crossover effect on depression. Namely, husband's positive work-family spillover had a longitudinal (across 1 year) crossover effect on wife's decreased depression, and wife's positive family-work spillover had a longitudinal crossover effect on husband's decreased depression. In addition, one cross-sectional crossover effect from husband's WFC to wife's depression was documented. However, the study did not reveal any significant longitudinal individual effects of either work-family conflict or positive spillover on individuals' depression. The authors consider this finding noteworthy, because one would expect longitudinal crossover effects to be more difficult to detect than would be longitudinal individual effects.
When looking at the study findings from the perspective of the Frone models (Frone et al., 1992; Frone, Yardley et al., 1997), the studies seem to give some support for the models. The cross-sectional findings showed, first, that one partner's work-family conflict (without directionality) was linked to his or her partner's work-family conflict (Hammer et al., 1997), which can be considered supporting the view that different types of work-family conflict (WFC and FWC) are related. Second, one partner's FWC was related to other partner's work-related outcomes (Hammer et al., 2003) as well as WFC was related to other partner's family-related outcomes (Matthews et al., 2006). All these relationships are in line with the Frone models. In the only longitudinal study (Hammer et al., 2005) work- and family-related outcomes were not studied.
In addition, crossover studies in the context of work and family focusing on other experiences than work-family conflict have been published. Of these studies, those focusing on the issue of whether well-being or distress crosses over from one partner to another are relevant from the perspective of the present study. That is, because we examine both partners' well-being in the domains of work and family, and therefore also their crossover from one partner to another. Westman (2001), in her review of 29 studies examining crossover effects, noted many studies that have found one person's distress to be positively related to his or her spouse's distress. Among these reviewed studies there were also longitudinal ones, but only few had a long-term perspective like that in our study. Of these few studies, that by Barnett, Raudenbush, Brennan, Pleck, and Marshall (1995) showed direct longitudinal crossover of distress between partners, that is, an increase in distress (i.e. anxiety and depression) of one partner was mirrored in the changes in distress of the other over a 1-year time period. This study was also the only one using multi-level modelling, that is, taking into account the hierarchical structure of the data.
The main purpose of this study is to assess the longitudinal relations of work-family conflict (i.e. WFC and FWC) and the experiences of job satisfaction and parental distress among dual earner couples. On the basis of the theoretical views presented by Frone and his colleagues (Frone et al., 1992; Frone, Yardley et al., 1997), we test the following six hypotheses:
Hypothesis 1 (stability)
WFC, FWC, job satisfaction, and parental distress show considerable stability across the 1-year period.
Hypothesis 2 (normal causality)
WFC at Time 1 is related to parental distress and FWC at Time 2, and FWC at Time 1 is related to job dissatisfaction and WFC at Time 2.
Hypothesis 3 (reversed causality)
Job dissatisfaction at Time 1 is related to WFC at Time 2, and parental distress at Time 1 is related to FWC at Time 2.
Hypothesis 4 (normal causality)
One partner's WFC at Time 1 is related to the other partner's parental distress and FWC at Time 2, and one partner's FWC at Time 1 is related to the other partner's job dissatisfaction and WFC at Time 2.
Hypothesis 5 (reversed causality)
One partner's job dissatisfaction at Time 1 is related to the other partner's WFC at Time 2, and one partner's parental distress at Time 1 is related to the other partner's FWC at Time 2.
Hypothesis 6 (domain-specific well-being effects)
One partner's job satisfaction at Time 1 is related to the other partner's job satisfaction at Time 2, and one partner's parental distress at Time 1 is related to the other partner's parental distress at Time 2.
The data were gathered as a part of the research project ‘Economic Crisis, Job Insecurity, and the Household’, in which 608 couples participated in the first phase of the study (1999; Time 1). One year later (2000; Time 2), these couples were asked to participate in the second phase of the study. Of the couples, 468 complied with the request, yielding a response rate of 77%. Originally, the participants were drawn randomly from the database of the Population Register Centre of Finland (see Kinnunen et al., 2004; Kinnunen & Feldt, 2004). The participants have been shown to represent reasonably well both the original sample and Finnish working aged people in demographic background factors available (gender, age, marital status, and geographical location).
The present sample was restricted to those couples in which both partners were employed at both measurement times. Of the original 608 couples, 387 were working couples (63.6%), and from them 239 (61.8%) participated at Time 2. The demographic characteristics of the participants did not change significantly between Times 1 and 2, thus only Time 1 demographics are described. About 78% of the couples were married and 22% cohabiting. The marital or cohabitational relationship had on average lasted 17 years (SD = 10.8). The average age of the men was 43.7 years (SD = 8.7) and of the women 42.2 years (SD = 8.6) (p <.001). The great majority of the partners were between 35- and 54-years-old. Of the couples, 21% (N = 50) did not have children, and among those with children living at home, the number of children was typically one or two and the median age of the youngest child was 9 years. Most had a vocational school or a college education, men more often vocational school education (33% vs. 25%) and women college education (37% vs. 23%; p <.001). Consequently, men worked more often than women as blue-collar (35% vs. 28%) workers, whereas women worked more often than men as white-collar workers (67% vs. 50%; p <.001). The majority of men (95%) and women (88%) worked full-time (>34 h/week), and the average number of weekly working hours was 44.0 (SD = 10.5) for men and 38.5 (SD = 7.3) for women (p <.001).
Two identical postal questionnaires bearing the same code number were posted at Time 1 to each member of the randomly-selected sample. One questionnaire was intended for the target individual and the other, where relevant, for his or her partner. At Time 2 the questionnaires were sent only to those completing the questionnaires at Time 1. The respondents were instructed to complete the questionnaires independently and to post them in separate sealed envelopes to the researchers. The partners were matched by the code number.
WFC was measured with two-items (e.g. ‘My work keeps me from my family more than I would like’) from the scale by Stephens and Sommer (1996), and FWC with two-items (e.g. ‘I have to put off doing things at work because of the demands on my time at home’) from the scale developed by Netemeyer, Boles, and McMurrian (1996). The items were rated on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). The correlations (all significant at the level of p <.001) of the two WFC items were .72 (Time 1) and .65 (Time 2) for men and .65 (Time 1) and .74 (Time 2) for women, and the corresponding correlations for the two FWC items were .53 and .53 for men and .54 and .66 for women, respectively.
Job satisfaction was assessed globally with three-items (e.g. ‘In general, I am satisfied with my job) using a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). The items are from the job diagnostic survey (Hackman & Oldham, 1980), which has been validated in Finland (Vartiainen, 1989). The Cronbach's alphas for the scale were .85 (Time 1) and .82 (Time 2) for men and .84 (Time 1) and .80 (Time 2) for women.
Parental distress was measured with six-items from the parental stress index developed by Abidin (1990). The measure used indicates the degree to which the parent reports experiencing the parent role as stressful (three-items, e.g. ‘Being a parent is harder than I thought it would be’) and restricting (three-items, e.g. ‘I feel trapped by my responsibilities as a parent’). All items used a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). The Cronbach's alphas for the parental distress scale were .81 (Time 1) and .85 (Time 2) for men and .83 (Time 1) and .86 (Time 2) for women.
The hypotheses of the present study were tested within the SEM framework using the Mplus 5.0 programme (Muthén & Muthén, 1998–2006). The responses of each participant and those of his/her partner were matched and the models were estimated simultaneously for both genders, first, because it is not possible to investigate the crossover process between the partners using separate models for men and women and, second, because the married or cohabiting couples have to be considered as a dependent sample. A latent variables approach was chosen as it offers potential advantages enabling measurement errors to be taken into account. Each latent variable (i.e. WFC, FWC, job satisfaction, and parental distress) was therefore constructed by multiple observed items or scales. Scales were used in the case of parental distress, as using the high number of original items would have resulted in an unacceptably high ratio of estimated parameters in relation to sample size.
The analytical procedure included three major phases. In the first phase, a stability model based on the all latent constructs was tested in order to establish the invariance of the factor loadings of the latent variables across time and across partners. Demonstration of structural invariance of the latent variables allows the further investigation of the longitudinal individual and crossover effects with no need to worry about whether observed effects are due to gender-related factors or to structural change in the latent constructs over time. In the second phase, the hypotheses (2–3) of the individual effects were tested by adding all the individual-effect paths to the stability model. After that, the significant paths (t-values 1.96 or above) were retained and the non-significant paths (t-values less than 1.96) were deleted from the model. In the third and final phase of the analytical procedure, the hypothesized crossover paths (4–6) were added to the model, and again, the non-significant paths were dropped from the model.
As the data included both categorical (items indicating WFC, FWC, and job satisfaction) and continuous variables (two mean scores of parental distress) we used weighted least squares means and variance adjusted (WLSMV) estimation with full information data treatment (i.e. weighted least square parameter estimates using a diagonal weight matrix with standard errors and mean- and variance-adjusted chi-squared test statistic that use a full weight matrix). In order to test the adequacy of the hypothesized SEM model, a number of key model fit indices were examined as part of the analysis. First, the chi-squared and degrees of freedom are presented. A non-significant chi-squared value is considered to be indicative of an acceptable model.
Second, when the competing models were compared in testing the invariance of the stability model across time and partners, we performed chi-squared difference testing. It is notable that the chi-squared value for the WLSMV method of estimation cannot be used for chi-squared difference tests as such. Chi-squared difference testing, that is to be used in conjunction with the WLSMV method of estimation is described in detail in the Mplus technical appendices (see www.statmodel.com) and in the Mplus User's Guide by Muthén and Muthén (1998–2006). If the chi-squared difference test produces a non-significant loss-of-fit, the invariance assumptions are supported.
Third, in conjunction with the chi-squared statistics, a combination of the Tucker Lewis index (TLI) and the comparative fit index (CFI) was used. In addition, the root mean square error of approximation (RMSEA), which takes model complexity into account and is generally considered to be one of the most informative indices, was in use. As adjusting the index cutoff values on the basis of the model characteristics (e.g. the number of participants and the observed variables included in the model) has been recommended (Hair, Black, Babin, Anderson, & Tatham, 2006), we followed these guidelines. In our case with 239 couples and 36 observed variables included in the SEM model, the criteria for a good model fit are .90 and above for TLI and CFI values, and .07 or less for RMSEA values with an acceptable CFI value. In addition, a significant chi-squared value can be expected (Hair et al., 2006).
The means, standard deviations and correlations of the study variables are presented separately for the husbands and the wives in Table 1. Based on paired sample t tests, gender differences emerged for six of the eight model variables. Husbands had a higher level of WFC at Time 1 (p <.01) and at Time 2 (p <.05) compared to wives. In a similar vein, husbands reported a higher level of FWC than wives at both Times 1 (p <.001) and 2 (p <.05). However, wives were more likely than husbands to report parental distress at both Times 1 (p <.01) and 2 (p <.05). Job satisfaction was at a similar level in both partners at both measurement times. No mean level changes in any of the variables occurred between Time 1 and Time 2.
Table 1. Pearson correlations of the study variables among husbands and wives (N = 190–239)
As can be seen in Table 1, WFC and FWC at Time 1 correlated significantly with parental distress at Time 2 both among the husbands and the wives. In addition, parental distress at Time 1 correlated significantly with WFC and FWC at Time 2 among both partners. However, of the longitudinal relations between WFC and job satisfaction, only two turned out to be significant: WFC at Time 1 was related to decreased job satisfaction at Time 2, and job satisfaction at Time 1 was related to decreased WFC at Time 2 for the wives. Thus, the individual longitudinal correlations seem to support the reciprocity model.
The inter-correlations of the study variables between the partners are presented in Table 2. As can be seen in the table, both partners’ WFC and FWC as well as job satisfaction and parental distress correlated significantly with each other, providing a first indication of covariation between the partners' responses. Furthermore, one partner's WFC and FWC at Time 1 showed a significant correlation with the other partner's parental distress at Time 2, whereas only husband's parental distress at Time 1 was related to his wife's WFC at Time 2. In addition, wife's job satisfaction at Time 1 correlated with her husband's decreased WFC at Time 2. Thus, in the light of the longitudinal crossover correlations the normal causality seems to receive more support than the reversed causality.
Table 2. Pearson correlations of the study variables between husbands and wives (N = 190–239)
Testing the hypotheses
In the first phase of our SEM analyses, we tested the adequacy of the stability model of the latent constructs investigated. To do this, each latent variable was constructed either by multiple items (i.e. all items of WFC, FWC and job satisfaction were set on the corresponding latent variable) or scales (i.e. the mean scores of stress and restrictions of parental role were set on the latent variable of parental distress) in husbands and wives at both measurement times. We first tested a freely estimated stability model where each latent construct was set to predict itself and no invariance constraints were imposed in factor loadings. In this freely estimated stability model, the associations between the latent variables at Time 1 were estimated. At Time 2, none of the residual error covariances between latent variables were significant, and were therefore not included in the model. The fit statistics of this freely estimated model were well within good fitting limits (χ2 = 165.152, df = 95, p <.001, CFI = .97, TLI = .98, RMSEA = .056). The significant chi-squared value did not support this very complex model (16 latent variables, 36 observed variables), as expected, but the other fit indices easily met the threshold of a good model fit.
Next, the invariance of the factor loadings across time was tested by setting the corresponding factor loadings equal over time (χ2 = 167.348, df = 97, p <.001, CFI = .97, TLI = .98, RMSEA = .055). These equality constraints did not weaken the model fit as indicated by the chi-squared difference -test (Δχ2 = 12.39, Δdf = 9, ns). Consequently, the structure of the latent variables was concluded to be invariant across time. To detect the invariance across partners, the corresponding factor loadings were finally set equal across husbands and wives (χ2 = 164.803, df = 97, p <.001, CFI = .97, TLI = .98, RMSEA = .054). As the deterioration in model fit was not statistically significant (Δχ2 = 3.427, Δdf = 4, ns), the model was concluded also to be invariant across partners. Thus, this stability model with equality constraints in factor loadings across time and across partners was chosen as the basic model for our subsequent analyses of longitudinal individual and crossover effects.
In the second phase, six hypothesized individual-effects paths in both partners (hypotheses 2 and 3) were added to the above described stability model with equality constraints (χ2 = 154.018, df =97, p <.001, CFI = .98, TLI = .99, RMSEA = .050). A closer examination of the t-values indicated that, of the 12 individual paths, nine were non-significant. After dropping the nine non-significant paths (starting from the smallest t-value), the model fit was adequate (χ2 = 150.746, df = 97, p <.001, CFI = .98, TLI = .99, RMSEA = .048). Of the individual paths, the following three turned out to be significant (see Figure 1): First, husbands' high FWC at Time 1 was linked to their high WFC at Time 2. Second, wives' high FWC at Time 1 was linked to their high WFC at Time 2. Third, wives' high FWC at Time 1 was linked to their decreased job satisfaction at Time 2.
In the third phase of the analysis, all the hypothesized crossover effects (eight in both partners; see hypotheses 4–6) were added to the model (χ2 = 158.876, df = 96, p <.001, CFI = .98, TLI = .98, RMSEA = .052). Of these, none turned out be significant, and therefore the predecessor model, shown graphically in Figure 1, with three longitudinal individual-effects paths was the final model. For the sake of clarity, the concurrent ψ-associations between the latent variables at Time 1 are presented in Table 3.
Table 3. Concurrent ψ-associations between the latent variables at Time 1
As can be seen in Figure 1, the stability of the latent constructs investigated was consistently relatively high, stability coefficients ranging from .55 to 1.00. As already stated, one longitudinal individual effect was detected in husbands; high FWC at Time 1 was related to husbands' high level of WFC 1 year later. Two longitudinal individual effects were detected in wives. Similarly to husbands, wives' high FWC was linked to their high WFC 1 year later. In addition, wives' high FWC was linked to decreased job satisfaction 1 year later.
Our results revealed that only FWC showed long-term effects. First, high FWC at Time 1 was related to high WFC 1 year later in both partners, and second, wife's FWC was linked to her decreased job satisfaction over 1 year. These findings were in line with our Hypothesis 2, which was based on the earlier model of Frone and colleagues (1992). According to this model, the outcomes of FWC reside in the domain receiving the conflict (i.e. WFC and job dissatisfaction). The present findings also support Frone and colleagues' cross-sectional results showing that only FWC (and not WFC) was related to job distress and depression (Frone et al., 1992), and longitudinal results showing that only FWC had a detrimental effect on an individual's health over 4 years (Frone, Russell et al., 1997). However, the theoretical elaboration presented by Frone, Yardley et al. (1997b) concerning the indirect reciprocal nature of the relationship between WFC and FWC was not supported (Hypothesis 3), that is, neither job dissatisfaction nor parental distress increased WFC and FWC later on.
Consequently, the significant longitudinal findings at the individual level gave support for normal causality which has so far also received most support in other studies (e.g. Frone, Russell et al., 1997; van Hooff et al., 2005). Thus, the theoretical explanation, according to which FWC reflects a state in which an individual's resources are threatened and depleted leading to an increase in WFC and a decrease in well-being received support (e.g. Eby et al., 2005; Voydanoff, 2002). One possible statistical reason making it difficult to find reciprocal relationships may lie in the moderately high stability of the experiences examined during the 1-year follow-up among both husbands and wives. However, the observed moderate stability of the experiences was in line with our stability hypothesis (1).
The longitudinal relationship established from FWC to WFC in both partners is consistent with the views of gender similarities in work-family interface experiences (Gutek, Searle, & Klepa, 1991; Voydanoff, 2002). According to the social-role hypothesis perspective (Voydanoff, 2002), the key issue is a comparable role involvement, not gender: if both men and women invest the same amount of time, attention, and energy in their work and family roles, these involvements should have similar effects on the experiences of men and women. In the Nordic countries, men and women are equally active in working life, and in this respect their investments could be considered to be identical, rendering our results interpretable. By contrast, the longitudinal effect observed from FWC to job dissatisfaction among women is against the social-role hypothesis. This finding suggests that perhaps family is still a more salient and resource consuming role for women, and therefore FWC has a more marked effect on women's psychological well-being (Voydanoff, 2002).
It is worth noticing that these effects occurred in spite of men's higher levels of WFC and FWC compared to those of women. The finding of men's higher level of WFC is consistent with earlier studies, for example, across twenty-seven studies (Byron, 2005) male employees tended to have slightly higher WFC. However, in this same review, female employees tended to have higher FWC than men. Thus, our findings suggest that Finnish women are more resistant than their partners to the perceived conflicts. One reason for these gender differences may be the fact that in practice women have been more active (e.g. utilizing care and flexibility benefits) in juggling with the demands of work and family. This means that women have developed strategies for handling both work and family responsibilities over the years (see e.g. Hammer et al., 2005).
In the present study, no longitudinal crossover effects were found. This was contrary to our Hypotheses 4–6. However, we based our hypotheses mostly on cross-sectional studies, because there is a lack of longitudinal crossover research. The only existing longitudinal study addressing the crossover effects of work-family conflict showed no crossover effects of work-family conflict on depression 1 year later (Hammer et al., 2005). In that study only positive work-family spillover showed longitudinal crossover effects on decreased depression among partners. Thus, one reason for the absence of crossover effects may be due to our focus on work-family conflict. It could be that positive work-family experiences are more likely to cross over in close relationships over time. This issue is an important area for further research.
Another possible reason for not finding longitudinal crossover effects is that, in order to find crossover effects of work-family conflict, a different time lag than the 1 year used in the present study, would be needed. Although we regarded 1 year as an adequate time period in regard to our outcomes, in the absence of theoretical views, it is hard to say what the optimal time-lag would be to establish longitudinal crossover effects. Therefore, we encourage researchers to adopt different time lags in order to learn more about lagged crossover effects. Previous studies suggest that rather a shorter than a longer time period than 1 year would be needed in order to be able to demonstrate, for example, reciprocal crossover relationships between work-family conflict and domain-specific well-being (see Demerouti et al., 2004; Leiter & Durup, 1996, for reciprocal effects at an individual level). However, De Lange et al. (2004) indicated that a 1-year time lag was the most appropriate for demonstrating causal relationships between demands, control and social support and indicators of mental health.
A third reason for not finding longitudinal crossover effects may be the fact that we focused on time-based conflict. This could be regarded as a limitation of the present study, although in this way (i.e. not using strain-based items), an overlap in item contents was avoided between work-family conflict and well-being variables, which could have caused overestimated associations, as pointed out by Hurrell, Nelson, and Simmons (1998). On the other hand, time-based work-family conflict, and especially time-based FWC, might have captured women's real worries (e.g. delays and interruptions during working day due to family commitments such as taking children to daycare and picking them up from daycare in fixed times) better than those of men. Perhaps, time-based conflict is perceived as an external restriction by women, that is, they feel that it is beyond their control, and therefore time-based WFC has detrimental effects for women (Steinmetz et al., 2008).
The present study is not without limitations. First, although each partner was requested to complete the questionnaires separately, we cannot be entirely sure if this really was the case. However, if partners completed the questionnaires jointly, this may have contributed to greater crossover; therefore, this may not be a problem in our study. A second limitation concerns the fact that we only used self-report questionnaires. Common method bias may thus have strengthened the relationships. Third, it has been proposed that crossover research should start looking for mechanisms (e.g. mediating roles of support and empathy or moderating role of personality) via which crossover might occur (e.g. Westman, 2001, 2006). Unravelling such processes, however, was beyond the scope of this study. Focussing on mediators and moderators in subsequent longitudinal crossover research could be a fruitful avenue. Fourth, our results might be best generalizable to countries where sociocultural expectations and practices concerning working, household duties, and child rearing among men and women are characterized as egalitarian rather than traditional (see Westman, 2005).
Of the types of work-family conflict, only FWC had long-term effects, suggesting that it is more important than WFC. Therefore, the prevention of FWC could be given priority when long-term preventive and intervention strategies are planned for working couples with children. This suggests that in the Nordic countries, too, where the state has an active role in reconciling work and family demands, additional efforts are still needed in order to avoid negative long-term consequences of FWC. In this respect it is important to develop daycare system as well as children's care after school. In Finland, the daycare system is functioning well, but there are not enough after school activities available for school-aged children. This is a problem because Finnish parents are often employed full-time. Also, extra domestic help (e.g. cleaning help) would prevent feelings of exhaustion and save parents' energy for their children. Equal division of domestic duties between partners would especially be beneficial for women, because Finnish women spend still more time in doing domestic tasks than men. All these measures would be helpful in preventing the experience of FWC.
The research project ‘Economic Crisis, Job Insecurity and the Household’ was financially supported by the Academy of Finland (grant No. 43553).