The present study sought to answer explicit research questions about groups with elevated levels of WLC, important correlates and health-related outcomes of WLC among adult employees in Switzerland as well as potential gender differences in this regard.
Regarding the first research question about identifying groups at risk of experiencing WLC and work-related and other factors associated with WLC, the study clearly showed that high numbers of hours spent at work, regular overtime, long commutes and a high job status (management position) are all associated with elevated levels of WLC. These findings make sense, since high work demands compete with other social roles and private obligations for limited time, energy and psychological resources, thereby leading to multiple burden and increasing role conflict between work and private life [30–33]. Other significant and relevant influencing factors found in the present study were job insecurity and living with dependent children. Interestingly, a differentiated composite score for WLC composed of 2 items yielded more or less the same risk groups with above average prevalence rates as a dichotomous single-item measure (yes/no question) about one's difficulties in reconciling work with private life [28].
Nearly all of the above-mentioned results are fully in line with international research showing that high job involvement and workload and in particular a large time commitment to work as well as job insecurity and childcare responsibilities are antecedents of WFC [2, 7, 10, 20, 34]. On the other hand, three of the findings go against expectations and/or are not consistent with international WFC research. Firstly, contrary to international studies showing an association between flexible work schedules (such as flextime) and lower WFC [10, 34–36], in the present study variable work schedule in terms of alternating or changing working hours from day to day was found to be associated with higher WLC compared to fixed, constant working hours. Variable working hours in this sense may not represent real self-determined time flexibility at work, but poorly predictable and other-directed working hours which of course are not a resource for balancing or reconciling one's work with his/her private life.
Secondly, in international studies, job autonomy--also known as influence at work, decision latitude, operational flexibility or work-time control--was found to be related to positive spillover between work and family and to be protective with regard to WFC [31, 35, 37–39]. But in the present study, job autonomy was not significantly associated with WLC. This may be due to a measurement problem. Participation in decision-making at work as assessed in the SHP survey (see section on "Measures") may not imply real autonomy in how and when the job gets done and is therefore not beneficial for one's WLB.
Thirdly, housekeeping has also been shown to be another antecedent of WFC in international studies [20], but that finding was not replicated in the present study. Since the level of engagement in paid work varies depending on the time spent on housework in our study, more housekeeping hours per week often go along with lower level of employment and a decreased number of weekly work hours which have been proven to be beneficial with regard to WLB.
In spite of the oft-quoted dual burden of working women and despite the fact that women mostly spend more hours in combined work and family activities and have a greater total load than men [8, 40], in the present study women at large were found to be somewhat less affected by WLC than men, as reported in other studies [20, 41]. Every seventh man, but only every ninth woman in the study population showed a (very) high WLC. This finding is in line with gender role theory and perspective and the so-called domain salience hypothesis postulating that role pressures and conflict are itensified when either work or family roles are salient and central to the person's self-concept [5]. In other words: the more important a role is to an individual, the more time and energy that individual will invest in it and the more likely are conflicts with other roles [42]. In light of still persisting traditional gender roles [8] it is assumed that the work domain is a greater source of role pressures and conflict for men while the family domain is a greater source of role pressures and conflict for women [40, 43]. Based on this argumentation and according to the rational view of gender differences [8], it has been concluded that men report more work-to-family/life conflict than women and women report more family/life-to-work conflict than men [44].
But in fact, the gender difference found in this study and going in the assumed direction is caused by an under-representation of women in full-time jobs and management positions in the study population just as well as in the general (working) population. Only 34.2% of women in the study sample work full-time and just 21.1% are in a management position, whereas 86.9% of all men work full-time and 42.8% are in a leading position. If prevalence rates are adjusted for job status and level of employment--i.e., if women and men working full-time and in the same job position are compared (see Table 2) --the gender difference with respect to WLC decreases, disappears entirely or is even reversed.
Findings in the research literatur on WFC referring to gender differences are contradictory [44]. Some research has found no gender differences [40, 42, 45], whereas other studies have reported gender differences in this regard mostly in the sense and direction of women experiencing more WFC than men [8, 36].
Beyond this, some interesting gender differences were found. While in men regular overtime, variable work schedule, and high job status (executive or management position) were most strongly associated with a higher level of WLC, in women the time committed to work or rather the number of contracted hours spent at work was by far the strongest explanatory variable. Job insecurity, long commutes, and living with dependent children in contrast turned out to be additional factors of equal strength among both men and women. These findings need to be replicated by others before they can be generalised, since they are inconsistent with findings from other studies like Jansen et al.'s cohort study [20] that found job insecurity to be an antecedent of WFC especially for men, and overtime work, long commuting time and having dependent children to be greater risk factors for women than for men.
Altogether the variance explained by all explanatory and controlling variables included in the linear regression model was only 11.1% for men and 13.5% for women. This large 'unexplained variance' suggests that there are other relevant, but unconsidered factors especially in the nonwork domain such as family obligations beyond childcare, leisure activities not covered in the SHP, and personality traits or individual preferences concerning the need for WLB, work ethos, commitment to family etc.
As far as the second research question regarding health correlates of WLC is concerned, findings of the present study confirmed that WLC is quite strongly associated with impaired general well-being, reduced physical health and limited mental health. These findings concur with international studies. Allen and Armstrong [46], Grandey and Cropanzano [33] and Frone et al. [24] found that WFC was associated with poor physical health. Others showed that WFC had an adverse affect on mental health [19, 47] or was strongly related to depression [24], fatigue [20] and psychiatric disorders [3].
Results show reasonably strong associations with a clear gradient (odds ratios increase in tandem with WLC) almost throughout, and associations are not diminished when adjusting for age, education and different work and private life characteristics. Those men and women who report a moderate or (very) high WLC show almost consistently (although not always significantly) an increased relative risk for general ill-being and different mental and physical health problems (e.g., poor self-rated health, serious headaches, negative emotions and depression, sleep disorders, fatigue) in comparison with those with inexistent or very low WLC (reference group). Multiple adjusted odds ratios for the most exposed group with high or very high WLC range from 1.5 to 4.7 depending on the health outcome. Only sick leave or rather being absent from work for 20 days or more in the past 12 months constitutes an exception to this pattern and is not associated with WLC at all.
Interestingly, when looking at mental health problems as outcome variables, women show slightly higher odds ratios--i.e., somewhat stronger associations between WLC and mental health impairments--than men. This finding is supported by a previous longitudinal study of Kinnunen et al. [45] who found work-to-family conflict to be more detrimental to women's satisfaction and well-being than that of men. An explanation for this gender difference might be that negative spillover or interference from work to family/private life is more stressful and problematic mentally for women because the family role and private life domain is more important to the woman's self-concept and social identity [5, 8, 40]. Men in turn obtain their personal and social resources (e.g., self-esteem, social status, social identity, social support) more from work and are therefore less affected mentally by such role conflicts from work to family or private life and by negative sanctions as a result of noncompliance with family role demands [5, 8, 40]. So work-to-life conflict as measured in the present study may have more adverse effects on women's (mental) health and well-being, whereas life-to-work conflict which was not assessed in the study may impair men's health.
Strengths and limitations of the present study
One of the goals of the present cross-sectional study was to overcome some of the major limitations of the international research on WFC. With their generally homogenous, non-representative samples, findings between studies in this field of research usually cannot be compared with each other or transferred to other groups, much less the general population or the entire labor force. With very few exceptions, there are practically no representative population-based studies. By broadening the limited scope and traditionally narrow focus on WFC, by using nationally representative survey data, and by having a study population that includes not only white-collar employees or specific subpopulations or occupational categories but also blue-collar workers, findings from the present study can be generalised to the entire employed population in Switzerland, thereby overcoming the widespread middle-class bias in this field of research and partly compensating for the lack of evidence in this country.
Of course, the study also has some methodological limitations. WFC research has been criticized for an overreliance on cross-sectional study designs [25], and this criticism applies just as well to the present study which does not permit causal inferences. Since WLC and health outcomes have been measured simultaneously, causality is uncertain and doubtful. Statistical associations found in observational studies can never prove causal relationships [48]. Cross-sectional data in particular cannot respond to the question if exposure precedes the outcome - a key criterion on Austin Bradford Hill's widely-cited list of criteria to be considered before inferring causation when observing a statistical association [49]. And although longitudinal data by contrast comply with this criterion of temporality, longitudinal data and evidence are not sufficient to fully allow the assumption of causality [45] and to conclude from association to causation either. None of Hill's criteria are sufficient and none, perhaps with the exception of temporality, are absolute conditions and sine qua non for causation [50].
Yet strong associations with a clear gradient (linearly rising relative health risks with increasing WLC) found consistently between WLC and diverse (mental) health outcomes suggest a potential cause-effect relationship according to two additional criteria of causation, namely strength and linearity (gradient) [49]. And in addition, considering alternative explanations by stratification of analysis and/or controlling statistically for potentially confounding factors is a useful strategy to distinguish effects of exposure from those of confounding and another way to "study association before we cry causation" [49].
However not only the question of causality remains open-ended in this study, but even the direction of causation is unclear since in recent years researchers found evidence and support for reverse causation between WFC and health. Several longitudinal studies on antecedents and consequences of WFC have shown lagged effects indicating the hypothesised causal relationship [51, 52] as well as reversed effects and bidirectional or reciprocal relations [51, 53–56] between work-family interaction or work-home interference on the one hand and different work stressors or health outcomes on the other.
Another point of criticism in most WFC research is poor measurement [25]. This is also a major limitation of the present study. By relying on secondary analysis of existing data, we were strongly limited in the measurement of WLC as a multidimensional construct. WLC was assessed with a 2-item scale containing just two out of three distinguished forms (time- and strain-based) aligned to one causal direction (work-to-life conflict) only and showing a low reliability coefficient (alpha = .53). The third form (behavior-based) and the other type or direction of conflict (life-to-work conflict) were not measured at all in the SHP and therefore could not been used for this study. In other words, assessment of WLC in the present study is far away from the well-established and best-validated multiple-item measures such as the 18-item scale of Carlson et al. [4] or the 12-item scale of Netemeyer et al. [57].
Since health and social sciences increasingly use multiple-item scales to measure complex multidimensional constructs such as self-esteem, health status, stress, job satisfaction, and many others, there is an ongoing debate on the validity and reliability of single-item measures compared to multiple-item measures [58, 59]. However, by replacing a dichotomous single-item measure of WLC [28] with a score composed of two 11-point Likert-scaled items, there are fewer concerns at least about validity and potential measurement problems in this study. Furthermore, the construct validity of the scale used is bolstered by finding similar groups with elevated levels of WLC and mostly consistent associations with different health outcomes in other research on WFC.
However, poor measures (especially when measured by questionnaire) can produce information bias and as a result misclassification bias. Misclassification of exposure which is independent of other (measurement) errors and non-differential with respect to the outcome results in estimation bias towards the null value (which is 1 when using common measures of association such as the odds ratio) [60, 61]. When exposure is uncommon or when misclassification of exposure and outcome is non-differential but not independent of one another, bias away from the null can result [62, 63].
In the present study, exposure is not rare, but using single-source self-report survey data carries a potential risk of non-differential non-independent misclassification of exposure and outcome which occurs when misclassification errors are correlated. This is the case when both work-life conflict (exposure) and physical and mental health disorders (outcomes) are either overreported or underreported by the same subjects leading to an overestimation or an underestimation of the true association. 'Finding' a non-existing association due to (dependent) misclassification would be a major problem.
Using different sources of information or data for exposure and outcome would have strongly reduced the risk of dependent misclassification [61, 63] and of common method variance or bias. In the present study, using a differentiated ordinal scale instead of a dichotomous variable as a measure of exposure (which is by nature gradual and dose-dependent and not a binary state) may have reduced the risk of information and misclassification bias. Similar associations found consistently for all health outcomes may indicate that the probability of dependent misclassification is rather low. It is not plausible otherwise that the same measurement or misclassification error could have been observed for all health outcomes.