Examining the roots of turnover intentions in the Royal Norwegian Navy, the role of embeddedness, work-life conflict and predictability

Job embeddedness refers to the collection of forces that influence employee retention (Mitchell et al. 2001). Embeddedness theory posits that turnover is a product of both work and non-work factors that combine to embed individuals within a group, organisation or workplace (William Lee et al. 2014). The theory describes organisational fit, links and sacrifice as a web of employee retention and posits and that it can account for turnover variance not explained by traditional job satisfaction models (Jiang et al. 2012). Organisational fit refers to the perceived compatibility within the organisation. Organisational links refer to both work and community-based social bonds, and sacrifice is the perceived cost by leaving a job (Mitchell et al. 2001). The current study measures embeddedness using Mitchell’s et al. (2001) theoretical definition of job embeddedness as a multi-factorial measure that includes both formative and reflective components relating to links, fit and sacrifice. Although global reflective measures of job embeddedness have been put forward (Crossley et al. 2007), the current study follows Mackenzie’s (2005) recommendation of utilising a composite formative model of embeddedness. The distinction is important and has practical implications for predicting turnover intentions in the current study because organisational embeddedness and community embeddedness do not share a common latent factor. As such, the current study subscribes to a view of embeddedness where the factors are treated as having separate and distinct consequences. This approach has the advantage of theoretical and conceptual richness but lacks the statistical opportunities of a reflective measure.

The purpose of the study is to investigate the associations between the components of job embeddedness and turnover intentions and their effects on turnover intentions, along with work-life conflict and predictability, but not to evaluate the relationships between latent embeddedness factors. In the current study, the original wording of organisational sacrifice (Mitchell et al. 2001) is referred to as career prospects to ease interpretation. The internal meaning of the construct is unchanged and refers to the sacrifice or loss represented by changing jobs and thereby removing oneself from career prospects within the armed forces. The current study also separates organisational fit and community fit into two distinct factors in line with the on-the-job and off-the-job original framing within the theory (Mitchell et al. 2001). Both career prospects and organisational fit are considered reflective as they tap a set of attitudes and beliefs, while organisational links and community fit are considered formative.

Hypothesis 1 (H1) states that organisational links are negatively associated with turnover intentions. The armed forces are organisations characterised by increasingly complex work tasks and requires cooperation between different disciplines and areas of expertise (Laurence and Matthews 2012). According to job embeddedness theory, such connections indicate an employee’s embeddedness in the organisation (Mitchell et al. 2001) and subsequently reduce turnover intentions. As such, personnel with a larger number of organisational links are hypothesised to experience lower turnover intentions because they are more embedded in the organisation.

Hypothesis 2 states (H2) that organisational fit is negatively associated with turnover intentions. According to job embeddedness theory, organisational fit - perceived compatibility with the workplace - embeds individuals to a workplace and reduces turnover (Mitchell et al. 2001). This perspective is important in the armed forces as work is mostly done in teams, and high personnel turnover hurts teams. Research shows that person-organisation fit predicts both task performance and work engagement (Sørlie et al. 2020). Additionally, according to both the attraction-selection-attrition theory (Schneider et al. 2000) and the self-concordance theory (Parker et al. 2010), people are attracted to organisations that are similar to themselves and are more engaged in achieving work goals if these are in line with their core interests and values (Sheldon and Elliot 1999; Parker et al. 2010). In addition, organisational socialisation processes also likely influence the embeddedness of service members and reduce turnover (Kammeyer-Mueller and Wanberg 2003).

Hypothesis 3 (H3) states that positive perceived career prospects are negatively associated with turnover intentions. Another relevant embeddedness factor is the perceived career prospects within the organisation. Job embeddedness theory highlights how the perceived fit of career plans with the opportunities presented in the armed forces causes turnover intentions if they are not aligned (Mitchell et al. 2001). In addition to theory, recent evidence from the Swedish armed forces places career opportunities as the most important factor to predict turnover (Fors Brandebo and Lundell 2018). In line with job embeddedness theory, the relationship between perceived career prospects and turnover intentions is likely affected by training and education within the armed forces. Personnel with military-specific knowledge and skills may be less likely to leave armed forces than those with skills that are more attractive in the civilian sector.

Hypothesis 4 (H4) states that community fit is negatively associated with turnover intentions. We predicted that service members who have many social ties within their communities to be more embedded within the armed forces and have fewer turnover intentions. Being embedded within a community has spillover effects on organisational embeddedness (Ng and Feldman 2013). There is reason to believe that this effect is important within a military context as workplaces within the Norwegian Armed Forces are seldomly within large population centres. A sense of community fit embeds service members within the armed forces as the workplace is rooted within a community, and leaving the armed forces likely involves moving away from that community. As such, we hypothesised that community fit would lead personnel to be less likely to leave the armed forces. It should be noted, however, that the same process may cause increased turnover intentions if service members have to move more for work, but want to remain geographically located within a specific community.

In addition to embeddedness, work–life conflict and predictability are also highly relevant as predictors of turnover intentions within the armed forces (Wadsworth and Southwell 2011). We hypothesised that work–life conflict is positively associated with turnover intentions (H5). Secondly, predictability has both a negative direct effect on turnover intentions and a negative indirect effect via worklife conflict (H6). Experiencing mutually incompatible demands from work and non-work is a hallmark of the military profession. In addition, the reliability and predictability of work schedules can also be highly variable. Operational demands and deployment represent unique challenges that may contribute to turnover intentions if they reduce the family and non-work opportunities for the individual service member and create negative spillover effects (Escarda et al. 2020). Having a high degree of predictability and lower levels of work–life conflict are assumed to be associated with a reduction in turnover intentions, but evidence supporting this proposition is mixed. Some studies have found a relationship between work–home conflict and turnover intentions (Heen 2012; Lillekvelland and Strand 2015) some do not (Heilmann et al. 2009). The discrepant results are likely a product of different work–life situations within different branches, organisations and countries. Research nevertheless suggests that work–life conflict is the mechanism that explains why predictability is associated with lower turnover intentions as flexible hours and predictability increase work–life balance (Richman et al. 2008), which, in turn, decreases turnover intentions (Surienty et al. 2014). There is less research within the military context but reports from the Inspector General of the Department of Defence (2021) also suggest the same relationship and highlight the important role of flexible hours. Among female employees in the Norwegian Armed Forces, predictability was found to be important to enable the possibility to coordinate childcare and support from family and friends, and especially if both parents were working in the armed forces (Olsvik and Lauritzen 2014). This is taken to mean that high levels of predictability are associated with lowered work–life conflict, which, in turn, is associated with a decrease in turnover intentions, as an indirect effect. These hypotheses are based on the spillover theory of work–life balance (Staines 1980), which postulates that work predictability enables individuals to integrate work and non-work responsibilities, decreasing work–family conflict and increasing health and work performance (Hill et al. 2003).

Lastly, we evaluate the hypothesis that the option of having flexible, or variable hours, increases predictability, which, in turn, decreases turnover intentions via the relationship stipulated earlier. Hypothesis 7 (H7) states that flexible hours are associated with a decrease in work–life conflict both as a direct effect and as an indirect effect via predictability. Flexible hours are becoming more common and have demonstrable effects on work–life balance (Russell et al. 2009). In the armed forces context, the use of flexible hours is an important tool to increase predictability and reduce some of the inherent work–life conflictsfor military service members. For some, flexible hours may primarily yield an effect on work–life conflict via the ability to plan and organise work tasks ahead of time and increase predictability, as an indirect effect. For others, it may not be related to the predictability at work as an indirect effect, but rather to decrease friction between work and home life, as a direct effect.

In addition to the hypotheses, the study also attempt so answer the research question (R1): Do embeddedness, work–life conflict and predictability affect turnover intentions differently for commissioned and non-commissioned personnel? A serious challenge to the generalisability of research on turnover intentions within the Norwegian Armed Forces is unobserved heterogeneity. Individual service members likely differ by measured and non-measured demographic factors that group personnel into sub-populations, which may differ in turnover intentions and other characteristics. The most likely source of differential turnover mechanisms is the categorisation between commissioned (CO) and non-commissioned (NCO) personnel. COs are employed until retirement, while NCOs are either contractually employed until 35 years old or retirement. The mechanisms that produce turnover intentions for NCOs may be different from COs turnover mechanisms because they likely have differing levels of commitment, embeddedness, skills and opportunities within and outside of the armed forces. NCOs may not be as embedded within the armed forces because they have less education and training within the armed forces than COs who have several years of leadership training. There are also likely differences between commissioned and non-commissioned personnel in civilian career opportunities. Specifically, NCOs may be more heterogeneous in their job tasks, and have specific knowledge and experience. The rate of turnover is also higher among NCOs, and they are recruited at a younger age, making them less established in their careers than their CO counterparts (Lillekvelland and Strand 2015). Taken together, these differences between NCOs and COs imply that different mechanisms produce turnover intentions. Existing evidence does not lend itself to formulating specific hypothesis but does warrant the use of multi-group analysis to answer the research question. The study also includes stratified descriptive information regarding education and service length, living arrangements, children and weekly commute that is used to compare and contrast the commissioned and non-commissioned personnel.

A cross-sectional design with stratified sampling was applied for this study. Potential respondents in the Royal Norwegian Navy were divided into groups of personnel category, age, gender, commuters and non-commuters, marital status, educational level, parental status and affiliations. A total of 1,736 NCOs and officers from the Norwegian Fleet and Norwegian Coast Guard were invited to participate in this study. Conscripts, civilians, hired consultants, retired personnel on short-term contracts and apprentices were excluded from this study due to their irrelevant employment status. The survey was administered electronically, and 465 completed the questionnaire, resulting in a 27% response rate. All pre-defined stratums achieved a satisfactory response rate. As such the results are deemed to be representative of the Royal Norwegian Navy. To ensure privacy, data on organisational affiliation, gender and age were excluded from further studies after checking representativeness. The sample characteristics are presented in Table 1. As expected, COs have longer service length and more education. They are also more likely to be married, have children and commute. These differences highlight how divergent turnover mechanisms may operate for the two personnel groups.

Turnover intentions were measured with the turnover intention scale (TIS-6) (Bothma and Roodt 2013) but were adapted to reflect turnover intentions to leave the armed forces. Example items include ‘How often do you consider leaving the Armed Forces’ and ‘How often do you daydream about another job that better suits your needs’. Items were scored on a 5-point Likert scale ranging from 1 (never) to 5 (all the time) and indicate whether the employee intends to quit. The scale achieved a high omega reliability of 0.85.

The work-life conflict was measured with six items from the work-family conflict scale (Mansour and Mohanna 2018) and the work-leisure conflict scale (Wong and Lin 2007). Preliminary factor analysis revealed that the indicators were best represented by a single factor as work-family and work-leisure were highly correlated. Items were scored on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Example items include ‘I don’t have enough spare time after work for leisure activities’ and ‘My work interferes with my duties in the home’. The scale achieved a high omega reliability of 0.91.

Predictability items were developed from scales inspired by Golden et al. (2006) and measured with the following four indicators: 1: ‘My unit work plans are reliable and predictable’, 2: ‘My work plans are predictable to the extent that I can plan weeks ahead’, 3: ‘My units work schedule is predictable enough for me to plan ahead for months’ and 4: ‘I often have to come to work on my days off (reversed)’. The scale achieved a high omega reliability of 0.88. Flexible hours were measured with a single item: ‘My workplace allows for the use of flexible hours’. Both flexible hours and predictability items were scored on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).

The job embeddedness measure was based on the scale developed by Mitchell et al. (2001). For the purposes of the study, the authors translated the items to Norwegian and adapted 16 items to fit in an armed forces context. Example items include ‘I am well compensated for my performance (Career opportunities)’, ‘I consider the area I live, to be my home (Community fit)’, ‘I am a good fit in the culture of the Navy (organsational fit)’ and ‘How many work groups do you belong to? (organisational links)’. Answers for embeddedness items, except organisation links, were measured on a 5-point Likert scale ranging from 1 (to a very small extent) to 5 (to a very large extent). Answers for organisational links items were recorded on a five-point ordinal scale with two-point increments ranging from 1 (zero) to 5 (more than seven). The complete scale was subjected to a principal component analysis to determine sub-factors, using parallel analysis and promax oblique rotation. The results conformed to the theoretically specified four-component structure that included career prospects (Omega: 0.88), organisational fit (Omega: 0.76), community fit and organisational links. Career prospects and organisational fit denote reflective latent factors and are represented as latent factors, while community fit and organisational links are composite scale scores.

Lastly, the participants answered demographic questions regarding service length (How many years have you been in the Royal Norwegian Navy?) and education length (What is you level of education?). Years in the navy were recorded on a five-point scale with 4-year increments ranging from 1 (zero to four) to 5 (more than twenty). We also asked three yes/no questions: Do you live with a partner? Do you have children? and Do you weekly commute to and from the workplace during the workweek?

Firstly, a multi-group latent factor measurement model was used to determine the measurement model fit and evaluate measurement equivalence between COs and NCOs. The latent factors included in the model were organisational fit, career prospects, turnover intentions, work-life conflict and predictability. Secondly, a baseline model where factor loadings (metric) were allowed to vary freely across groups was used. The model included all the study’s latent factors and indicators, and covariance parameters between all latent factors, as well as a covariance parameter between two indicators of work-life conflict, based on similar wording (Time spent at work makes it difficult to fulfil my duties at home/The demands of my work make me unable to complete tasks at home). We fit the same model with factor loading constrained to be similar across groups and evaluated the loss of model fit using a χ² Wald test (Ward and Ahlquist 2018) to determine equivalence between the two groups. The χ² test assesses constraints on statistical parameters based on the weighted distance between the unrestricted loading and the constrained loading under the null hypothesis of no differences between parameters (Ward and Ahlquist 2018). The test can combine restrictions into a combined joint test for each parameter class.

The baseline measurement model showed acceptable fit to the data (Hu and Bentler 1999): a comparative fit index (CFI) of 0.950, a root mean square error of approximation (RMSEA) of 0.052, a standardised root mean square residual (SRMR) of 0.058 and a Bayesian information criterion (BIC) of 27051. Constraining factor loadings to be equal across commissioned and non-commissioned personnel resulted in a small reduction in model fit (CFI: 0.948, RMSEA: 0.053, SRMR: 0.065, BIC: 27,090), resulting in statistically significant differences between the constrained and the unconstrained model (χ² (19) = 30.66, p < 0.044). Using modification indices, we determined that one item on the latent turnover intentions factor (This job satisfies my personal needs (reverse scored)) contributed most to the lack of metric equivalence between the groups. Removing this parameter resulted in a non-significant difference to the baseline model (χ² (18) = 17.69, p < 0.476). Thus, partial metric equivalence was reached. We decided to ignore the item from the scale to ensure content validity and because the magnitude of the difference was small. The difference is likely due to the different personal needs of COs and NCOs.

The multi-group structural equation modeling (SEM) framework was also used to test the hypotheses. The SEM has the advantage of assessing relationships of latent error-free constructs, and the multi-group function allows the analyses to be conducted groupwise in one parsimonious model (Bollen and Hoyle 2012). As such, evidence for and against a hypothesis can be evaluated for each personnel group separately, in addition to differences between the groups in structural paths, covariance paths and measurements. All models included covariance parameters between exogeneous variables, and there were no covariance parameters between error terms of endogenous structural variables. We used a specify–test–respecify procedure, where the theoretical embeddedness model with work–life conflict and predictability was tested first and then evaluated. We used modification indices on the original model to identify missing parameters and respecified with an alternative model. The original model regresses turnover intentions on the four embeddedness factors as well as predictability and work–life conflict as direct effects. In addition, the model includes indirect effects of predictability on turnover intentions via work–life conflict as well as an indirect effect of flexible hours on work–life conflict via predictability. The models were estimated using maximum likelihood. The indirect effects were calculated using the medsem package (Mehmetoglu 2018) to obtain confidence intervals. The results are based on the Zhao et al. (2010) approach to mediation using Monte Carlo simulation and decompose the indirect effect as a percentage of the total effect. The estimation of indirect effects was done with separate samples. We used the multi-group function to determine differential effects between the two groups and path coefficients between groups were tested with χ² Wald tests to determine differential effects. Thus, the current study contains both confirmatory elements with hypothesis testing, but also two exploratory elements regarding the research question of differential effects and the relative impact of factors on turnover intentions. Initial data inspection revealed that many indicators were did not show normality. Because of this, all SEM models used the Satorra and Bentler (1988) estimator. The dataset was complete as all participants completed the entire survey. All analyses were conducted using Stata v.16 (StataCorp 2019).

The descriptive statistics stratified based on officer commission can be viewed in Table 2. The descriptive statistics confirmed that commissioned and NCOs differ in embeddedness, turnover intentions and flexible hours. These differences are consistent with the notion that COs are more embedded within the armed forces. However, both groups experience about the same level of work–life conflict. The most notable differences are in career prospects and turnover intentions, where COs enjoy higher career prospects and lower turnover intentions. The descriptive statistics also revealed that NCOs vary considerably in their career prospects and community fit, showing that the group is more heterogeneous than the commissioned personnel group. The overall sample means turnover intention reveals that Royal Norwegian Navy personnel are, on average, neutral to the prospect of leaving the armed forces. Evidence of considerable turnover intentions was also observed in answer to the question ‘How often do you consider leaving the armed forces’, where 14% answered ‘all the time’, 29% answered ‘often’, 24% answered ‘neither often or rarely’, 29% answered ‘rarely’ and only 3% answered ‘never’. The bivariate associations provide preliminary support to the hypotheses as turnover intentions were closely correlated with embeddedness variables as well as work–life conflict and predictability.

Before evaluating the parameters of the model and testing hypotheses, we evaluated the global model fit of two structural models in both sub-samples. We fit the original a priori model and a respecified model with modifications to the original model. Modification indices revealed that the original model was lacking a structural parameter linking career prospects to predictability. We did not originally include an effect of career prospects on predictabilitybased on theory, but the strength of the relationship and improved model fit strongly suggested its inclusion. The theoretical implication is that service members with higher career prospects also show more predictable work schedules. This may be due to a number of reasons, and the relationship is likely bidirectional in nature. The addition of the parameter of the model contributes to a new indirect effect of career prospects on turnover intentions via predictability. The respecified model thus posits that having positive career prospects gives more predictability, which, in turn, reduces turnover intentions.

The model fit of the original and the respecified model can be viewed in Table 3. The results showed that the original model did not achieve adequate model fit. By contrast, the respecified model achieved satisfactory model fit for COs on all fit indicators. For NCOs and the total sample, the results showed acceptable fit in all indicators, except CFI, which was slightly lower than the recommended cut-off (Hu and Bentler 1999). In all, we regard the respecified model as reasonably consistent with the data and adequate to test the hypotheses.

The standardised path coefficients from the multi-group SEM can be seen in Figure 1. Measurement coefficients can be viewed in Table 4. The results support the view that COs and NCOs share most, but not all, turnover mechanisms, and the multi-group SEM model showed acceptable fit. Overall, the predicted mechanisms explained approximately 70% of the variation in turnover intentions for both groups. The results showed a small, but significant, association between organisational links and turnover intentions for COs, but not for NCOs, partly confirming Hypothesis 1. The same pattern of results was found for organisational fit on turnover intentions, partly confirming Hypothesis 2. Hypothesis 3 states that career prospects predict turnover intentions. This hypothesis applies for both COs and NCOs, and an association of notable effect size was found in both groups. The link was larger for NCOs and accounted for most of the explained variance in turnover intentions in this group. The results did not support any relationship with community fit in either group, disconfirming Hypothesis 4. The χ² tests of group differences in embeddedness coefficients did not support significant overall differences in structural paths between COs and NCOs, indicating that the same turnover mechanisms exist for both commissioned and non-commissioned personnel.

Fig. 1.

The covariance coefficients are presented in Table 4. The two groups showed different covariance parameters (χ² (13) = 31.02, p < 0.01). The difference between the two groups was mainly due to non-significant covariance parameters between organisational links and career prospects for NCOs.

The results from the multi-group SEM supported Hypothesis 5 and showed that work-life conflict predicted turnover intentions for both COs and NCOs. The size of the effect was approximately equal between groups, lower than career prospects and smaller than organisational fit for COs. The results also supported the hypothesis that predictability is negatively associated with turnover intentions. Furthermore, the results from the test of indirect effect of predictability on turnover intentions via work-life conflict was significant for both NCOs (β = -0.14, SE = 0.04, Z = -03.34, p = 0.001) and COs (β = -0.13, SE = 0.04, Z = -3.07, p = 0.002). The mediation was complete for NCOs and partial for COs, confirming Hypothesis 6. For COs, the indirect effect comprised 38% of the total effect. For NCOs, the indirect effect comprised 56% of the total effect. Hypothesis 7 states that flexible hours would be associated with a decrease in work-life conflict by increasing predictability. The results supported a direct effect in both groups, and a partial indirect effect for COs (β = -0.09, SE = 0.04, Z =-2.23, p = 0.026), lending partial support for Hypothesis 7. Approximately half of the total effect of flexible hours on work-life conflict was mediated by predictability. Thus, the results indicate that flexible hours decrease work–life conflict because of both work and home factors for COs and only home factors for NCOs.

The results point to differential targets for interventions aimed at reducing turnover for COs and NCOs and are summarised in Table 5. For NCOs, interventions to reduce turnover intentions should be aimed at increasing career prospects within the armed forces. The strength of the relationship observed here shows the largest potential for turnover intention reductions. Secondary objectives are attempts to increase predictability to reduce work-life conflict. Reducing work-life conflict would likely be directly beneficial for all service members but may be impractical due to operational and strategic factors. Efforts could instead be aimed at increasing predictability. For NCOs, the results of the current study suggest that this will, in turn, reduce work-life conflict.

The results for COs also point to career prospects as the most important factor, in predicting turnover intentions, but not to the same extent as NCOs. For COs, organisational fit represents another notable predictor of turnover intentions and could be a practical target for interventions. This could entail efforts to increase workplace cohesion and consider organisational fit in placements. The association between organisational links and turnover intentions means that this also could be the target of interventions for COs, but the small nature of the relationship shows that other factors may be better candidates for these efforts. Officers are also negatively impacted by work-life conflict, and increasing predictability may reduce both work-life conflict and turnover intentions. For COs, the overall relationship between predictability and turnover intentions is smaller than career prospects. The significant direct effect indicates that there are unaccounted mechanisms under which predictability affects turnover intentions, in addition to the indirect effect via a work–life conflict. Because of this, predictability is an important target for possible interventions for COs. Flexible hours are a candidate for this, but the effect is small, and other options to increase predictability should be explored. It should be noted that the list of targets for interventions to reduce turnover is not an exhaustive list. A notable portion of variance in turnover intentions was not explained by the study’s predictors.

It should be noted that these findings are likely influenced by survivor bias (Simundic 2013), where unsatisfied personnel leave the armed forces before becoming established in their career and social life. However, the partial overlap between the findings of the current study and the results of studies that ask personnel why they left the armed forces (Fors Brandebo and Lundell 2018), implying some of the same mechanisms of turnover. Nevertheless, the study utilises a cross-sectional dataset. As such, there are several limitations to the inferences that can be made. Without temporal control, we cannot rule out alternative models where turnover intentions cause embeddedness; work–life conflict and low predictability cannot be ruled out. Future research would do well to remedy this by measuring predictors at one time point and measuring changes in turnover intentions from one time point to the next. An autoregressive cross-lagged design would be particularly well-suited to ruling out competing common method models as well as the temporally reverse effect of turnover intentions on embeddedness and work–life conflict. In addition, there are likely some other factors that cause or interact with turnover intentions that are not accounted for. This is especially likely for the NCO results where model fit was worse. Future research should attempt to uncover a better model for predicting NCO turnover intentions. As previously mentioned, the results are also likely partly the product of survivor bias, and a complete understanding of why service members choose to quit cannot be obtained without a complete sample of both personnel who have not ended their career in the armed forces, and personnel who has. Lastly, the study is limited by the lack of actual turnover data. Although there is some debate regarding the relationship between turnover intentions and actual turnover (Cohen et al. 2016), recent evidence by Wong et al. (2021) suggests that the two are closely related. Because of this, turnover intentions are the natural starting point as they predict who is likely to end their career in the armed forces. Nevertheless, some research indicates that the turnover intentions to actual turnover link may be different for COs and NCOs (Strand 2019). As such, subsequent research would benefit from follow-up of actual turnover rates.

The results have implications for other branches of the armed forces and operational personnel in general. Many of the same organisational processes and mechanisms exist beyond the sample of this study, and the results are likely generalisable within this context. It should be noted, however, that differences in organisational cultures and conditions may lead to different results.