Why work-family conflict can drive your executives away?

This study assessed the relationship between work-family conflict, work engagement, workaholism and turnover intention in business executives. 275 professionals (65.4% men) completed an online survey containing scales to assess work-family conflict, workaholism, work engagement and turnover intention. Descriptive and inferential statistics were conducted, as well as network and pathway analysis. Partial correlations between the workaholism dimension “working excessively” (WE) and work-family conflict (WFC) were r = .21. WE also predicted WFC (χ2 = 26.24, df = 7, p < .001; CFI = .95; RMSEA = .10 (.06 .14), particularly in terms of work interfering in family life, leading to higher turnover

The changing global scenario in recent decades has resulted in new social, political, organizational and family configurations. In addition, organizations have begun to attribute their success and competitive advantage to human factors (Youndt & Snell, 2004). This means that in order for organizations to ensure the efficiency and effectiveness of their business, they must attract competent and innovative professionals (Agapito, Polizzi Filho & Siqueira, 2015). As such, executives become pivotal to organizational operations because of their responsibilities in working alongside people, managing strategic decisions, allocating human resources (Robbins, 2005) and viewing the organization from a strategic perspective, generating tangible results to ensure organizational survival and success (Orlickas, 2012).
Executives with these responsibilities face demands that can lead to conflict between the domains of their professional and family lives (Bakker & Geurts, 2004). When considering people as equipped with limited resources (Ten Brummelhuis & Bakker, 2012), it is evident that an imbalance between work-related and personal demands can result in work-family conflict (Greenhaus & Beutell, 1985). The origin of this conflict lies in the stress caused by the pressures of roles at work and in the family, which are incompatible to the point of harming one of these domains (Greenhaus & Beutell, 1985;Edwards & Rothbard, 2000).
Theoretically, according to Amstad et al. (2011), work-family conflict can be analyzed bidirectionally in terms of work interference with family (WIF) and family interference with work (FIW). In WIF, workers have trouble carrying out their family role in a health manner due to their work demands, while FIW can be characterized as the real or perceived inability to function fully as a worker owing to family demands.
Scientific interest in work-family conflict has grown over the last 30 years (Allen et al., 2012;Allen & Finkelstein, 2014;Michel et al., 2011), but studies with business executives are still scarce. Nevertheless, it can be inferred that this group tends to experience conflict due to the nature and responsibility of their work (Braun et al., 2016). Given the personal and organizational impacts, the nature of business executives' work and the need for further research on this group, the first hypothesis of this study is (H1) that both directions of conflict (WIF and FIW) are central to the research model that explain the interaction between work-family conflict, work engagement, workaholism and turnover intention in executives.
Comparisons of men and women shows that the impacts on family and work are experienced in different ways (Allen & Finkelstein, 2014). Some studies suggest that women are more affected by overlapping between the work and family domains (Sorj et al., 2007). However, studies have shown different results, suggesting that men also experience conflict between their roles at work and at home (Coltrane, Miller, DeHaan & Stewart, 2013;Vandello, 2013). There is also evidence that the presence and intensity of conflict is greater in professional men than women (Knudsen, 2009), indicating a lack of consensus about the effects of the phenomenon and the relevance and need of studies that contribute to this debate. Thus, our second hypothesis (H2) assumes that there will be differences in the explanatory power of the variables analyzed in the model (work engagement, workaholism and turnover intention in executives) when applied to men and women.
The presence of conflict between work and family is related to individual aspects such as health, well-being, job satisfaction, stress and sleep problems (Ng & Feldman, 2014;Goh et al., 2015). It is also associated with reduced organizational performance, absenteeism, turnover, and commitment (Wayne et al., 2013;Nohe & Sonntag, 2014); job and career dissatisfaction, late arrival to workplace, lowered levels of work commitment, and family satisfaction, as well as diminished quality of life and a variety of dysfunctional behaviors (Greenhaus et al., 2001). This leads to the third hypothesis to be tested (H3): that both directions of the variable work-family conflict will be positively related to turnover intention in executives.
Another aspect associated with work-family conflict that is highly relevant for professionals in executive careers is job involvement (healthy and pathological). Research on this topic indicates that having a strong focus on job activities and a high level of dedication to work might result in positive job outcomes, such as increased performance, when combined to positive affect (work engagement). It can also result in negative outcomes, such as work-family conflict, when combined to negative affect (workaholism) (Bakker, Shimazu, Demerouti, Shimada & Kawakami, 2014).
Work engagement is defined as a persistent and generalized positive affective-cognitive state characterized by vigor, dedication and absorption . It describes an intense and effective connection with work activities, as well as professional and personal confidence in the ability to deal with work demands (Schaufeli et al., 2002). Workers who report a high level of engagement show greater affective connection to their work and high-energy dedication to their tasks (Schaufeli et al., 2006). They also tend to show more family satisfaction and a better work performance because work engagement increases the likelihood of work-family facilitation. According to Greenhaus and Powell (2006), interrole facilitation can occur through an instrumental pathway when resources gained from one role directly improve functioning in another role or through an affective pathway, when experiences in one domain produce positive affect within that domain, which, in turn, improves individual functioning in another domain.
In contrast to engagement, workaholism (work addiction) is viewed as a pathology, characterized by a compulsive need for continued, excessive work. This need arises from internal pressures and recurring thoughts about work when not working (Clark et al., 2014). Workaholics tend to feel guilty when not working and to work under any conditions, even when they feel no pleasure in their activities or show physical symptoms of fatigue (Schaufelli et al., 2008). This phenomenon impacts negatively individual's family satisfaction, wellbeing, and satisfaction with life because professionals who have a compulsive drive to work excessively hard tend to invest time and effort in their work, leaving them with fewer resources to devote to their family (Bakker, et al., 2014;Wojdylo et al., 2014). Thus, the fourth hypothesis of this study (H4) is that workaholism affects negatively work engagement.
A recent study found that executives and other professionals who use smartphones for work tend to be connected to it for around 13.5 hours every workday (Deal, 2015). In line with this result, research on job stress has pointed that psychological detachment from work during off-job time allows professionals to regulate their affect over the course of the work week. Such measure showed to be even more necessary when employees invest heavily in their work activities or when they are leaders. Evidence shows that leaders' workaholism can negatively influence not only their own well-being, but also their followers', through interindividual crossover of affective, cognitive, and behavioral components of workaholism (Clark, Stevens, Michel, Zimmerman, 2016). Therefore, considering that uninterrupted involvement with work can lead to negative outcomes, and that workaholism, and especially overwork, can become part of the professional life of executives, we present (H5), which suggests that workaholism contributes to increasing turnover intention.
Empirical studies on work engagement demonstrate it is positively related desirable organizational results, such as job and life satisfaction, as well as better work performance. By contrast, workaholism is associated with negative results, including difficulties in interpersonal relationships with colleagues, feelings of physical exhaustion and metal strain ( Wojdylo et al., 2014), psychosomatic symptoms, mental and physical health problems, greater work-family conflict and lower levels of life satisfaction (Wojdylo et al., 2014). Workaholism has a significant impact on the lives of workers. However, the literature does not clarify whether the correlation between work-family conflict and workaholism among business executives is similar to that observed in other groups of professionals. Thus, we hypothesize (H6) that both forms of work-family conflict (WIF and FIW) will be positively associated with workaholism.

Participants
Inclusion criteria to participate in the study were: being a coordinator or higher; responsible for managing a team; currently employed; in an executive position for at least one year; in the job market for at least five years. Participants were 275 Brazilian professionals, men (65.4%) and women, with an average age of 40 years (SD= 8.60 years), in different executive positions (coordinator, supervisor, manager and director). The subjects had occupied executive positions for an average of 11 years (SD= 7.90 years). With respect to area of expertise, most worked in the service sector (55.6%). Monthly salaries ranged between BRL 2,000 -5,000 (14.3%), BRL 5,000 to 10,000 (37.2%) and over BRL 10,000 (48.5%). All subjects reported they were in a committed relationship, most were living with their partners (89.5%) and had children (61.7%) who were financially dependent on them (61.7%).

Measures
Data were collected using an online survey containing questions aimed at characterizing the sample as well as different psychological scales. The variables analyzed, and the instruments used are described below: Work-family conflict. This variable was measured using the Work-Family Conflict Scale -WAFCS, which was adapted and validated for use in Brazil by Bastos and Aguiar (2014). It consists of 10 items assessed on a Likert scale, ranging from 1 (Strongly disagree) to 5 (Strongly agree). The scale assesses two dimensions: work interference with family (WIF; α= .90) and family interference with work (FIW; α=.85).
Work Engagement. Was measured using the Brazilian version of the Utrecht Work Engagement Scale (UWES; Vazquez et al., 2015). It contains 17 items scored on a Likert scale ranging from 1 (never) to 7 (always). The items evaluated are: vigor (α = .86), dedication (α = .87) and concentration (α = .85); Work Addiction. This variable was measured using The Work Addiction Scale, which was adapted and validated for use in Brazil by Carlotto and Del Líbano (2010). The instrument assesses work addiction in its two main dimensions: Compulsive Work (α=.70) and Overwork (α=.74). It consists of 10 items in the form of statements scored from 0 (never) to 3 (every day) on a Likert scale.
Turnover Intention. Was measured using the Turnover Intention Scale -TIS, which is an adaptation of Amstad et al.'s scale (2011) made by Siqueira et al. (2014). It consists of three items on the form of statements aimed at evaluating the degree to which employees plan to leave their jobs (α = .95). Answers are scored on a Likert scale from 1 (never) to 5 (always).

Procedures
The study was carried out in line with Resolutions 466/2012 and 510/2016 of the National Health Council and approved by under protocol number CAAE 46277015.1.0000.5336. Data were collected from June 2015 to April 2016 using an online questionnaire. Participants were recruited in three stages: 1) invitations extended through social media; 2) recruitment by recommendation and personal invitations from members of the research group; and 3) recommended by other participants.

Statistical Analysis
Data were submitted to descriptive and exploratory analysis to assess missing data, item distribution and sample description. This was followed by confirmatory factor analysis of the scales and subscales with Mplus software version 5.2 (Muthén & Muthén, 2008), using the factor scores in subsequent analyses.
Network analysis was performed to investigate the relationships between the study variables. The network analysis platform uses graphs to represent a system of variables (nodes) and possible relations between them (edges or lines) in a two-step estimation. First, condi-tional relationships (partial correlations) were estimated through a node-wise regularized multiple regression with the application of Least Absolute Shrinkage and Selection Operator (LASSO; Friedman et al., 2008) algorithm. To control for sparsity (i.e., near-zero associations were fixed to zero) many levels of tuning parameter were estimated (n = 100) producing a continuum from more liberal (many edges) to conservative models (few edges remaining). At this phase an information index, named Extended Bayesian Information Criterion (EBIC, Chen & Chen, 2008), offered the best balance of true/false negatives (i.e. omitted edges).
In the second stage, conditional relations were expressed by means of graphical representation. Relationships between nodes were weighted and showed differences in magnitude and direction (positive or negative), depicted by different color intensities and line styles (Machado et al., 2015). For better understanding, the variables were arranged in a relative space and attracted or repelled according to the magnitude of their relationships, using the Fruchterman and Reingold algorithm (1991).
This technique is often used in work and organizational studies, like in the investigation of information flow and decision making (Merrill et al., 2008), inter-organization communication and bridging between sectorial activities (Wilson & MacDonald, 2018), and to investigate sustainable change in innovation and entrepreneurship programs in organizations (Matthew, Monroe-White & Engelman, 2017). Other applications concern the estimation of complex and multicausal inductive models in areas like health and well-being (Deserno, Borsboom, Begeer & Geurts, 2016). In the present study we applied the network analysis paradigm to investigate organizational health-related behaviors.
For the purposes of the present study, measures of centrality and shortest paths were analyzed based on the LASSO estimation. These measures produce information about the impact of variables in the system as well the more effective route of influence between them (Opsahl, Agneessens & Skvoretz, 2010). Network analyses were carried out using the qgraph package. Based on the results of graphical lasso analysis, path models were constructed (path analysis, Blunch, 2008) to illustrate, in a causality diagram, the main relationships observed by the nondirectional inductive approach. After model estimation of the total sample, the models for men and women were compared to fix the intercepts and coefficients. Path analysis was performed using Mplus software version 5.2 (Muthén & Muthén, 1998). Figure 1 shows the results of correlation and partial correlation network analyses, respectively. The nodes represent the variables analyzed and the thickness of the lines depicts the magnitude of relationships between the variables. Additionally, the solid lines indicate positive correlations while dotted lines are negative correlations. Figure 1 shows that all the variables are correlated and the variable WIF is central to the system investigated. Correlations are also stronger between nodes that represent a same construct (ex. WC and WE, which comprise Work Addiction). Additionally: WC (working compulsively) and WE (working excessively) were weakly correlated with FIW, ENG (engagement) and TI (turnover intention); there were moderate correlations between WC and WE and WIF, between FIW and WIF, and WIF and TI; with a strong correlation between ENG and TI.

Results
Analysis of the second model of Figure 1 reveals that eliminating spurious correlations (partial correlation network) reduces the strength of relationships between variables. However, the positive relation between WC and WE and between ENG and TI remains strong. Though less intense, relations between WE and WIF and WIF and TI remain. The same occurs between WE and TI. Specifically, among the variables studied, there was a significant and high magnitude correlation between WC and WE (r= .73), WIF and FIW (r= .32), WE and WIF (r= .21), WIF and TI (r= .20). A correlation was also observed between WC and TI (r= .10) and WE and FIW (r= .30). Negative correlations occurred between ENG and TI (r= -.40), FIW and ENG (r= -.07), and FIW and TI (r= -.12). Important information regarding the shortest path between the nodes can also be extracted by correlation network analysis (Table 1). The first column in the table shows the source node and the remaining columns the destination nodes, making it possible to identify direct and mediated relations in the system. It can be observed that the shortest paths to turnover intention (node 6) start at Work Interference with Family (node 2) and work Engagement (node 3), indicating closeness between these variables and the absence of mediation. It is also noteworthy that the shortest path to Work Interference with Family (node 2) begins at Excessive Working (node 4).  Figure 2 shows centrality, that is, the descriptive measures of the roles of the variables (nodes) in the estimated model. Strength indicates the extent to which a node is directly linked to other nodes. Closeness demonstrates how well a node is indirectly linked to other nodes and betweenness shows how important a node is in the path between two other nodes, or the degree to which it serves as a 'bridge' between other nodes. The results indicate that the node with the highest betweenness is Work Interference with Family (WIF) and those with the lowest betweenness are work Engagement (ENG) and Family Interference with Work (FIW), for both the correlation and graphical lasso networks. Similarly, the node with the highest closeness is WIF and the lowest ENG. Finally, the nodes with the greatest strength in the correlation and graphical lasso networks, respectively, are WIF and WE, while those with the lowest strength are ENG and FIW. The results of path analysis considered consecutive methods, consisting of initial network analysis estimated based on path analysis. Values were considered from χ² = 26.24, df = 7, p < .001; CFI = .95; RMSEA = .10 (.06-.14). After this adjustment, a comparative model was performed between the groups (men and women), with free parameters to enable intergroup variation and values of χ² = 40.68, df = 14, p <.001, CFI = .94, RMSEA = .12 (.08 -.16). Testing was also conducted using a model with constrained parameters, considering the intercept and coefficient with values of χ² = 56.81, df = 22, p<.001; CFI = .92; RMSEA = .11 (.07 -.15). Finally, the difference between the models was tested, with reported values of Δχ² = 16.13, Δdf = 8, p = .04. It was concluded that models with free and constrained parameters between the groups differed significantly in favor of the free-parameter model, which assumes different intergroup values. Table 2 shows which variables contributed significantly to model adjustments for the entire group (men and women), as well as the male and female groups, respectively. The term "by" indicates influence and directionality; for example, TI by FIW is how much FIW influences TI. Estimators separated by bars indicate the regression coefficients for the overall group, men, and women, respectively.  Table 3 shows the results of path analysis for the male and female groups, respectively. It can be observed that the models maintain their structure for both groups, while the weighted estimators between groups differ.

Discussion
This section is organized according to the order in which analyses were performed and highlights the results obtained. The hypotheses confirmed and refuted by each analysis are presented and discussed under the subtitles: 1) Correlation and graphical lasso networks, 2) Shortest path and Centrality, and 3) Path analyses.

Correlation and graphical LASSO networks
Hypothesis 1, which contends that both forms (WIF and FIW) of the variable work-family conflict are positively related to turnover intention (TI), was partially confirmed. Additionally, although WIF and FIW showed weak correlations with TI in the first analysis, only WIF maintained a positive, albeit weak, correlation with TI. According to the literature, work-family conflict leads to reduced organizational commitment and greater turnover intention. Furthermore, difficulty reconciling the work and family dimensions decreases levels of wellbeing, job satisfaction and affective commitment to the organization, resulting in turnover intention (Poelmans & Caligiuri, 2008).
Hypothesis (2) states that both forms of work-family conflict (WIF and FIW) are positively associated with workaholism. This was partially confirmed, since both dimensions of conflict were positively correlated with aspects of work addiction (moderate for WIF with WE and WC, and weak for FIW with WE and WC). This result was not evident in graphical lasso analysis, where only WIF was associated with the WE dimension of workaholism. This corroborates the findings of Piotrowski and Vodanovich (2006), who reported an increase in the number of working hours associated with work and non-work activities (resulting from technological advances) as motivators of excessive working. Similarly, Spurk et al. (2015) found that people who work long hours (compulsively or excessively) do so because they identify with their career or occupation and feel a need to reinforce this, particularly in cases of perceived job insecurity.
Hypothesis 3 stipulates that both forms of work-family conflict are negatively correlated with work engagement (ENG). The results partially confirm this, given that the dimensions of conflict showed negative but low correlations with work engagement. This corroborates the findings of Innstrand et al. (2008), who observed that work-family conflict may lead to increased exhaustion and a decline in work engagement. Similarly, Moen et al. (2015) reported that perceived conflict affects the engagement, productivity and mental health of both individual workers and teams within organizations.

Shortest Path and Centrality
Hypothesis 4 of the present study proposes that both forms of conflict are central to the model investigated. The results of shortest path and centrality analyses demonstrated that only WIF was central in the model as a function of connectivity and closeness with other variables. The shortest path to turnover intention was perceived WIF. The results partially confirmed H4, since FIW was not central to the model, the strongest variable in the model being working excessively (WE). The fact that WE was the central variable in the model corroborates the idea that, in our sample, work extends to non-work activities (Salanova et al., 2008), that is, personal and family life. WE was the most decisive variable in terms of work interference with family (WIF) and exhibited the highest explanatory power in relation to conflict (Byron, 2005), indicating that high work demands, which lead people to work longer hours, are the source of conflicting demands and perceived interference between the work and family domains (Michel et al., 2011;Spurk et al., 2015).
However, it is important to consider that WE may be the result of situational factors, such as the threat of losing your job, accumulated tasks, developing career objectives and high organizational commitment, which can heighten work efforts and dedication, leading to an increase in WE. Aguilera (2010) highlights that some companies or executives consider working overtime a standard practice due to the pursuit of results and central role of workers. In this regard, Oltramari and Grisci (2014) reported that everyday aspects of executives' lives also affect family dimensions, since time that should be dedicated to family is often used to solve problems they work unable to resolve during normal working hours.

Path Analysis
Hypothesis 5 states that work addiction and conflict (both forms) contribute to increasing turnover intention. This was partially confirmed, given that only WIF explained TI. This corroborates the findings of Nohe and Sonntag (2014), who conducted a longitudinal study with 665 German professionals and observed that increased WIF raised TI over time, which did not occur for FIW. The literature indicates that work-family conflict interferes with life satisfaction (Fiksenbaum, 2014) and compromises family life due to work demands (Oltramari & Grisci, 2014). Moreover, turnover intention is closely linked to job satisfaction, level of perceived stress and job control (Hayes et al., 2012). Thus, since work-family conflict is a source of stress (Lipp, 2005;Nohe & Sontag, 2014) and many executives experience time management difficulties (Oltramari & Grisci, 2011), this creates the ideal scenario for the emergence or increase in turnover intention. In other words, executives whose jobs promote conflict between work and family domains and exhibit high addiction potential may be more likely to considering leaving the organization in pursuit of better work-life balance and overall satisfaction.
Hypothesis 6 stipulates that work addiction and conflict (WIF and FIW) negatively affected work engagement. The results reject this hypothesis, since these variables did not explain work engagement. It was believed that workaholism in executives could lead to a decline in positive mental state and be an antagonist of engagement . However, the rejection of this hypothesis may indicate a need to understand other mediator and moderator variables for the model, such as resilience or work-family enrichment (Greenhaus & Powell, 2006).
Finally, Hypothesis 7 was confirmed, signaling differences in the explanatory power of variables for men and women. Nevertheless, the structure of the models was similar, albeit with different regression coefficients and explanatory power. The explanatory power of WIF for turnover intention was three times higher among women (.36) than men (.12) and WE explained greater variation in WIF among women (.40) when compared to men (.29). Females executives studied here are more likely to exhibit turnover intention when they perceive their work as interfering with their family role. Alam et al. (2011) and Leaptrott and McDonald (2011) found that woman found it more difficult to balance work and family demands, experiencing greater work-family conflict. There is evidence in the literature that women tend to be largely responsible for domestic chores and nonwork activities even when they occupy executive positions (Machado et al., 2008 andStrobino &Teixeira, 2014). These aspects may partially explain their lower psychological investment in their career, as well as perceived difficulty handling both work and family or the realization that time dedicated to work affects their work-family balance, compromising family life and prompting the emergence of turnover intention.

Strengths and Limitations
It is important to underscore the methodological limitations of this study. It is a cross-sectional study with a sample from a specific cultural context, which relied exclusively on self-report scales. Furthermore, although the sample size and composition (more men than women) reflects the population of Brazilian executives, it precluded the testing of specific models that would allow a more detailed understanding of the behavior of variables for men and women. Further qualitative studies are suggested to understand the beliefs, attitudes and subjective aspects that correspond to a space of deeper relationships, especially with groups of women. There is also a need for studies that include data triangulation to correlate the facts, verbal observations and actions of individuals that result in a broader understanding of socially constructed meanings.
Despite these limitations, this study reinforced the idea that although the investigated population is difficult to access and therefore deserves greater scientific attention, it is vital to investigate how the consequences of conflict are experienced by executive professionals and their families. This is because the results obtained here highlight the need for companies to revise their people management policies, particularly those focused on work demands and distribution among executives. This could potentially reduce the negative effects of working excessively (overwork) for executives of both sexes, thereby promoting health in the workplace and family setting, in addition to reducing turnover intention.
Another practical implication is the finding of differences between men and women in terms of work-family conflict, working excessively and turnover intention. This allows companies to devise different organization initiatives tailored to men and women. According to the findings of a meta-analysis carried out by Butts et al. (2013), more than organization strategies that help administer work-family conflict, availability contributes to more positive work-related attitudes and, consequently, reduces conflict. Also, regarding the difference between men and women, it is important to note that women need greater organization support in the face of family demands (Matias et al., 2012).
Finally, we found that, in general, conflict has a negative effect on individuals, organizations and society as a whole. As such, this topic is highly relevant and further study is needed on the mental health of workers and its impacting factors, such as work-family conflict, in populations of business executives of the same sex in different cultural contexts.