Gender differentiates the predictors of an intention to leave the workplace: a meta-analysis of the effort-reward imbalance workplace stress studies

ABSTRACT

Workplace stress is related to job turnover and intention to leave (ITL) the job and/or profession. The specific mechanisms that drive this association have received less attention, however substantive research using the effort-reward imbalance (ERI) workplace stress model provides an avenue to address this issue and inform workplace interventions to improve employee retention. Our meta-analysis of the associations of components of the ERI model with ITL (k = 23, N = 73,671) demonstrated that low rewards (r = -.41, CI: -.46, -.37) were more associated with an ITL than ERI (r = .33 CI: .27, .38), overcommitment (r = .26, CI: .21, .31), and effort (r = .25, CI: .20, .30). Moderation analysis showed that females were more likely than males to report ITL when effort was high and that males were more likely to report an ITL the profession when rewards were low. Although all components of the ERI model were related cross-sectionally with ITL, prospective designs that target male-dominated professions are required to assess if the findings are robust.

KEYWORDS:

• ERI
• turnover
• employee
• overcommitment
• retention

Introduction

The deleterious effects of work stress are many and include physical (Siegrist & Li, 2016) and mental health (Harvey et al., 2017), dysregulated physiology (Eddy et al., 2016, 2018, 2023), decreased productivity (Hassard et al., 2018) and unsafe work practices (Dodoo & Al-Samarraie, 2019). Therefore, it is no surprise, that work stress is associated with job turnover (Söderberg et al., 2014) and intention to leave (ITL) the job or profession (Kim & Kao, 2014; Park & Min, 2020). Job turnover is estimated to cost an organisation up to 200% of an annual salary for recruiting, hiring, and training replacement staff (Allen et al., 2010), with additional indirect costs such as a loss of tacit knowledge (Dess & Shaw, 2001) and reduced customer satisfaction (Heavey et al., 2013). A strong proxy measure of job turnover is an employee’s intention to leave (He et al., 2020). While the consequences of work stress have been well-described, the specific features of work (and employee) that most contribute to an increased risk of turnover have received far less attention.

Large meta-analyses and systematic reviews have consistently shown an association between high workplace stress with increased ITL and actual job turnover (He et al., 2020; Kim & Kao, 2014; Park & Min, 2020). Although these reviews provide an important overview of the area, it is difficult to reliably determine which associations between workplace stressors and ITL are strongest as the data are pooled from different studies where most respondents are completing diverse measures of work demands (e.g. He et al., 2020; Kim & Kao, 2014; Park & Min, 2020). Additionally, most studies are not concurrently completing measures of job resources (Kim & Kao, 2014) or coping (He et al., 2020; Park & Min, 2020), so comparing the associations among constructs with ITL is problematic given employees within each meta-analysis of the work constructs with ITL come from different regions and occupations. A better way to identify the strongest predictors of ITL - to inform evidence-based intervention- is with a homogenous measure of workplace stress where employees provide responses on each of the key components of intrinsic and extrinsic work stress. The effort-reward imbalance (ERI; Siegrist, 1996) model of workplace stress is well-situated to fill this void.

The ERI measure was developed in concert with the ERI model (Siegrist, 1996) which purports that when an employee receives less reward than the effort expended, an effort-reward imbalance occurs, and workplace stress ensues which increases the risk of poor health (Siegrist, 2010). This ERI reflects the job-related characteristics of the workplace, however, the model also includes “overcommitment” (OC). As an intrinsic component of the ERI model, OC is characterised by an employee’s excessive ambition in combination with a strong desire for approval and esteem (Kinman & Jones, 2008). An overcommitted employee tends to over-invest in their work efforts and has difficulty withdrawing from their work (Feldt et al., 2016). The ERI model hypothesises that; (i), the individual components of effort (e.g. physical, emotional, psychological demands), reward (e.g. job security, opportunity for promotion, remuneration, esteem) and overcommitment (OC; a measure of maladaptive coping) will be associated with negative outcomes; (ii), the imbalance between efforts and rewards (e.g. ERI ratio) will have a stronger association with the outcome than any of the individual factors (efforts, rewards, OC) in isolation; and (iii) that OC moderates the effects of ERI on poor health (Siegrist & Li, 2016). Meta-analyses of the ERI model show it to be a strong predictor of health conditions such as coronary heart disease (Dragano et al., 2017), depression ( Rugulies et al., 2017), and dysregulated physiological responses (Eddy et al., 2016; Eddy et al., 2018, 2023), and mounting evidence appears to link the ERI model with ITL. Although the growing body of research (e.g. Fei et al., 2022; Hasselhorn et al., 2004; Jiang et al., 2022) is demonstrating the usefulness of the ERI in predicting ITL, a synthesis of these data is required to ascertain which components are most associated with an ITL.

Workplace stress models such as the ERI model can assess the separate contribution of intrinsic and extrinsic factors on employees’ ITL. Understanding if it is aspects of the workplace (efforts, reward, ERI) or maladaptive coping (OC) that is most associated with intentions to leave the occupation will provide an evidence-base to guide policy and inform person-focused or structural-level intervention(s). In addition, examining the role of potential moderators such as gender, occupation, or magnitude of stress within the workplace will further articulate the matching of intervention(s) to employee. The effects of workplace stress may differ between genders. For instance, work stress has been associated with poorer physical health in men but poorer mental health in women (Li et al., 2006; Mensah, 2021). Additionally, in the large Whitehall II study, women experienced higher workplace stress, depressive symptoms and inflammation than men, and in women only, higher depressive symptoms mediated the association between work stress and higher inflammation (Piantella et al., 2021). It is unclear, however, if the impact of workplace stress on ITL the workplace is gender-specific.

The aim of the present meta-analysis is to examine the relationships between ERI, OC, effort, and reward and the outcome of ITL. Specifically, we will assess (i), if ERI is associated with an ITL or if the association is driven by higher efforts or lower rewards (ii), if the intrinsic component OC or the extrinsic component ERI, is most associated with ITL, and (iii), if gender, occupation, or the magnitude of the stress reported by the sample, moderates the association(s) of the work stress constructs with ITL.

Method

Procedure

The first systematic search was conducted in September 2022 and the final search was repeated in June 2023 for relevant papers published since the introduction of the ERI model (Siegrist) in 1996. The search strategy was: (effort-reward imbalance OR overcommitment) AND (turnover OR quit* OR leav* OR exit*). The search was limited to English language and peer-reviewed journals.

A total of 1100 articles were identified through our initial database search (EBSCO: 75, JSTOR: 428, ProQuest Central: 77, ProQuest Psych: 20, PsycArticles: 69, PsychInfo: 83, ScienceDirect: 188, and Web of Science: 160). The study selection is depicted in Figure 1.

Figure 1. Flowchart of the Study Selection Process.

Note. ERI = effort-reward imbalance. ITL = intention to leave.

Study eligibility

After duplicates were removed, 878 articles were screened by title, abstract and keywords. Thirty-five articles were included in the full-text review. The updated search yielded an additional four articles. Two authors (MJ, RC) independently reviewed the full texts for inclusion in the meta-analysis. There were no disagreements on which studies should be included or excluded. Although the Asgarian et al. (2022) article met the inclusion criteria, the reported ERI of 70.6 seems improbable and was possibly calculated by summing the component scores (efforts, rewards, and OC) rather than calculating a ratio from efforts and rewards. Due to the lack of confidence in the data scoring and interpretation, it was excluded with full consensus from our team. The exclusion criteria are presented in Table 1.

Table 1. Exclusion Criteria for Study Inclusion in the Meta-analysis.

The paper was qualitative. The paper was not published in a peer-reviewed journal. The paper contained no original data. The paper discussed the descriptive, theoretical, or psychometric properties of the ERI model. The participants were not a working population. The paper did not utilise the ERI model. The paper did not measure turnover intentions The paper measured exiting the paid workforce due to retirement or disability. The paper did not report the effect size and the corresponding author did not provide the requested effect sizes

Note. ERI = effort-reward imbalance

Data extraction

Each article was read, analysed, and coded by one author and reviewed independently by a second author; intercoder reliability was satisfactory with both coders in total agreement with the data extracted. As such, a third reviewer was not required to resolve discrepancies.

Quality assessment

To assess the quality of the publication, each included paper was read, analysed, and coded independently by two authors, with disagreements resolved quickly by discussion until a consensus was met. The study quality was assessed using a modified version of the Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Analytical Cross-Sectional Studies (Moola et al., 2017) guidelines for appraising cross-sectional research. The JBI contains eight items and was scored 0, no problem, 1, minor problem, or 2, major problem. The included studies scored a mean of 1.26 out of a possible 16.

Data analysis

The Comprehensive Meta-analysis (CMA) software (version 3; Borenstein et al., 2005) contains in-built algorithms to convert raw data or effect sizes to various other effect sizes. Effect sizes for each study were either retrieved from the article or calculated or converted to an r value using the CMA software from information within the article, or the r value was requested from corresponding authors. Effect sizes from individual studies for each sub-group (ERI, OC, effort, reward) were combined using random-effects models to produce an aggregated effect size. Since the intentions to leave the workplace (ITL) outcomes varied by measure, they were classified into two categories: intention to leave the current job (ITLJ) and intention to leave the profession (ITLP). Fourteen studies measured ITLJ, while ten studies measured ITLP. The differences in effect sizes of ITLJ and ITLP were assessed. The heterogeneity of each sub-group was estimated by the Q statistic and evaluated using the I² statistic, which suggests scores of 0.25, 0.50, and 0.75 correspond respectively with low, moderate, and high levels of heterogeneity (Higgins & Green, 2011). Meta-regression was used to assess if the inclusion of gender, occupation, or magnitude of the stress (mean ERI) of the sample reduced this heterogeneity. As there were too few studies concurrently reporting ERI, OC and ITL scores, we were unable to assess if OC moderated the ERI-ITL association.

Publication bias was considered through visual assessment of funnel plots and Egger’s r calculations. The Fail-safe N was used to assess how robust the meta r values were as this metric informs how many additional studies with an r of zero would be required to render a significant association non-significant.

Results

Study characteristics

The main characteristics of the 23 included studies were tabulated (Table 2). The selected studies included 73,671 participants, with individual sample sizes ranging between 98 (Greenham et al., 2019) and 21,729 (Hasselhorn et al., 2004). This meta-analysis focused on cross-sectional studies only as there are limited prospective studies examining the relationships between ERI and ITL. To our knowledge, the ERI-ITL relationship has only been examined in two prospective studies: the Swedish Longitudinal Occupational Survey of Health (SLOSH) study (Leineweber et al., 2021) and the European Nurses’ Early Exit (NEXT) study (see Derycke et al., 2010; Li et al., 2011; Li et al., 2013). These NEXT study articles were not included in this meta-analysis as their baseline data were a subset of a larger included cross-sectional study (Hasselhorn et al., 2004). To harmonise our approach with the cross sectional studies, for the prospective study included in the meta-analysis (Leineweber et al., 2021), only the effect sizes from baseline data were used. The majority of the sample was female, one study was limited to females only (Tei-Tominaga et al., 2018), and 16 others were predominantly female. The employee occupations were categorised as academics (three), health workers (twelve), educators (three), or other (five).

Table 2. Summary of Characteristics of Studies Included in the Meta-analysis.

Author, Year	Sample size, sex, age range (M, SD), study location	Occupation	Study design	Specific predictors, ERI Mean	Outcome	Key findings
(Anker & Krill, 2021)	N = 2,863, ♀ = 51%, (Age♀ ≤ 30yrs = 11.7%, 31-40yrs = 27.9%, 41-50yrs = 25.2%, 51-60yrs = 21.4%, 61-70yrs  = 11.9%, > 70yrs = 1.9%. Age♂ ≤ 30yrs = 6%, 31-40yrs = 23.5%, 41-50yrs = 19.2%, 51-60yrs = 22.8%, 61-70yrs = 20.4%, > 70yrs = 8.2%), US	Attorneys	Cross-sectional	ERI - categorical. OC - categorical.	ITLP	Women had higher ERI and OC scores than men. OC positively related to ITLP, and this was stronger for men than women.
(Bianchi et al., 2018)^a	N = 257, ♀ = 76%, (M = 44.84, SD = 10.46), Switzerland	Teachers	Cross-sectional	ERI – continuous (M = 1.25, SD = 0.56). Effort - continuous. Reward - continuous.	ITLJ	Effort and ERI positively correlated with ITL. Reward negatively correlated with ITL.
(de Oliveira et al., 2017)	N = 3,229, ♀ = 87.3%, (M = 40yrs, SD = 10), Brazil	Nurses	Cross-sectional	ERI - categorical. OC - categorical.	ITLJ	Males more likely to report ITL than females. Younger employees more likely than older employees to report ITL. ITL increased with high ERI and OC.
(Devonish, 2018)	N = 323, ♀ = 67%, (18-34yrs = 42%, ≥ 35yrs = 58%), Barbados	Full-time employees	Cross-sectional	ERI – continuous (M = 1.15, SD = 0.47). Effort - continuous. Reward - continuous.	ITLJ	ERI had a direct positive effect on ITL.
(Dorenkamp & Weiß, 2018)	N = 421, ♀ = 53%, (M = 36.49yrs, SD = 6.01yrs), Germany	Academics	Cross-sectional	ERI – continuous (M = 1.56, SD = 0.56) OC - continuous	ITLP	Three-path-mediation showed an ERI-ITL link via strain and job satisfaction. OC-strain relationship is stronger for females than males. The direct ERI-ITL link is significant for females only.
(Fei et al., 2022)	N = 17,582, ♀ = 89.8%, (<25yrs = 19.9%, 25-34yrs = 60.0%, > 35yrs = 20.2%), China	Nurses	Cross-sectional	ERI – continuous (M = 1.26, SD = 0.61). Effort – continuous. Reward – continuous.	ITLJ	ERI and effort were positively related to ITL. Rewards negatively related to ITL. Depressive symptoms partially mediated the effects of ERI on ITL.
(Greenham et al., 2019)	N = 98, ♀ = 78.57%, 27-63yrs (M = 43.48yrs, SD = 7.62yrs), UK	School guidance counsellors	Cross-sectional	OC - continuous. Effort - continuous. Reward - continuous.	ITLJ	ITL positively related with greater effort.
(Hämmig, 2018)	N = 1,840, ♀ > 85%, < 45yrs = 58%, Switzerland	Health professionals	Cross-sectional	ERI - continuous.	ITLP	ERI was positively associated with ITL.
(Hasselhorn et al., 2004)^a	N = 21,729, ♀ = 87.3%, Belgium, Germany, France, Italy, The Netherlands, Poland, Slovakia	Nurses	Cross-sectional	ERI – continuous (M = 0.71, SD = 0.36). OC.	ITLP	High ERI was associated with high odds ratios for ITL. Small positive OC-ITL association.
(Jiang et al., 2022)	N = 10,457, ♀ = 27.02%, (≤31yrs = 29.75%, 32-35yrs = 20.56%, 36-40yrs = 24.71%, ≥ 41yrs = 24.98%), China	Emergency Physicians	Cross-sectional	ERI – continuous (M = 1.55, SD = 0.79). OC.	ITLJ	ERI and OC were positively associated with ITL.
(Kinman, 2016)^a	N = 649, ♀ = 42%, > 41yrs = 79%, UK	Academics	Cross-sectional	OC - continuous. Effort - continuous. Financial reward, esteem reward, security reward - continuous.	ITLJ	Efforts and OC positively related to ITL. Rewards negatively related to ITL. The strongest predictor of ITL was esteem rewards.
(Kinnunen et al., 2008)^a	N = 323, ♀ = 30%, M = 47.88yrs, SD = 8.52yrs, Finland	Managers	Cross-sectional	ERI – continuous (M = .55). OC - continuous. Effort - continuous. Reward - continuous.	ITLJ	Efforts, ERI, and OC positively correlated with ITL. Reward negatively correlated with ITL. ERI-ITL strongest under high OC and perceived organisational support.
(Lang et al., 2010)	N = 789, ♀ = 6.3%, (M = 32yrs, SD = 9.12yrs), US	Military reservists	Cross-sectional	Effort - continuous. Financial reward, esteem reward, meaning of military reward - continuous.	ITLP	High efforts/low financial reward positively associated with increased ITL.
(Lavoie-Tremblay et al., 2008)	N = 309, ♀ = 96.4%, 20-25yrs (M = 22.6yrs, SD = 1.28yrs), Canada	Nurses	Cross-sectional	ERI - categorical.	ITLJ ITLP	ERI positively related to both ITLJ and ITLP.
(Loerbroks et al., 2014)	N = 425, ♀ = 80.24%, Age (M = 35.74yrs, SD = 6.65yrs), China	Teachers	Cross-sectional	ERI – categorical (M = 1.17, SD = 0.35). Effort. Reward.	ITLP	ERI positively related to burnout and ITL.
(Leineweber et al., 2021)^a	N = 7,375, ♀ = 58.79%, Age (M = 49.94yrs, SD = 8.69yrs), Sweden	Employees	Prospective^b	ERI – continuous (M = 1.08, SD = 0.43).	ITLJ	ERI, OC and Effort positively related to ITLJ. Rewards negatively related to ITLJ. Higher income related to lower ERI. Higher income related to higher ITLJ.
(Martinez et al., 2022)^a	N = 605, ♀ = 85%, (≤30yrs = 17.02%, 31-40yrs = 36.86%, ≥ 41yrs = 46.12%), Brazil	Nurses	Cross-sectional	ERI – continuous (M = 0.74, SD = 0.43). OC - categorical.	ITLP	ERI and Effort positively related to ITLJ. Rewards negatively related to ITLJ.
(Panatik et al., 2012)	N = 267, Malaysia	Academics	Cross-sectional	OC - continuous. Effort - continuous. Reward - continuous.	ITLJ	Reward was negatively correlated with ITL. OC positively related to ITL. Effort not associated with ITL.
(Scanlan et al., 2013)^a	N = 103, ♀ = 91.3%, (<25yrs = 15.55%, 25-35yrs = 42.7%, 35-45yrs = 18.4%, > 45yrs = 21.4%), Australia	Occupational therapists	Cross-sectional	Effort - continuous. Reward - continuous.	ITLJ	Higher efforts, lower rewards related to higher well-being at work. Well-being at work was negatively related to ITL.
(Schug et al., 2022)	N = 757, ♀ = 76.9%, (18-30yrs = 24.2%, 31-40yrs = 23.8%, 41-50yrs = 20.9%, ≥ 41yrs = 31.2%), German	Nurses	Cross-sectional	Effort - continuous. Reward - continuous.	ITLJ	Higher rewards were related to lower ITL.
(Tei-Tominaga et al., 2018)^a	N = 5,074, ♀ = 100%, (M = 36.24yrs), Japan	Nurses	Cross-sectional	ERI (M = 0.43, SD = 0.15). OC - continuous. Effort - continuous. Monetary reward, Esteem reward, career opportunity reward - continuous.	ITLJ	Participants categorised into four age groups (≥50yrs, 39-49yrs, 25-34yrs, < 24yrs) Effort and monetary reward were associated with ITL across all age groups. Career opportunity reward was only associated with ITL for the 39-49yrs group. Esteem reward was not related to ITL. OC was associated with ITL in all but ≥50yrs group.
(Tominaga & Miki, 2011)^a	N = 1364, ♀ = 91.5%, (M♂ = 22.4yrs, SD = 2.06yrs; M♀ = 22.2yrs, SD = 1.43yrs), Japan	Nurses	Cross-sectional	OC - continuous. Effort - continuous. Money reward, esteem reward, career opportunity reward - continuous.	ITLJ	Effort was positively related to ITL. OC and reward were not related to ITL.
(Vu-Eickmann et al., 2018)	N = 887, ♀ = 98.41%, (M = 39.23yrs, SD = 11.43yrs), Germany	Physician assistants	Cross-sectional	ERI –categorical (M = 1.28, SD = 0.42). Effort - continuous. Reward - continuous.	ITLP	ERI associated with poorer self-reported health, elevated symptoms of depression and anxiety, poorer quality of care, increased ITL.

Note. ^aAdditional data provided by author/s. ^bbaseline data used in meta-analysis. ERI = effort-reward imbalance, ITL = intention to leave, ITLJ = intention to leave the job, ITLP = intention to leave the profession, OC = overcommitment

ERI and intention to leave work

Higher ERI was related to higher intention to leave work (k = 17, r = .33, p < .001) for the combined outcomes (ITLJ and ITLP). Subgroup analysis revealed that there was no difference in effect size between outcomes assessing ITLJ (k = 8, r = .36, p < .001) or ITLP (k = 9, r = .32, p < .001), Q_between = 0.25, df = 1, p = .62 (Figure 2). Evaluation of a funnel plot of results for the combined outcomes indicated more studies toward the right, suggesting the possibility of publication bias. Egger’s regression, however, indicated that asymmetry was not significant (p (2-tailed) = .65). Additionally, Duval and Tweedie’s trim and fill method indicated it was unlikely there were missing studies. The Fail-Safe N analysis revealed that the meta r finding was robust; 22510 more studies with an r value of zero would be required to render the current association non-significant.

Figure 2. Forest Plot of the Association Between Effort-Reward Imbalance and Intention to Leave.

Note. ITLJ = intention to leave the job, ITLP = intention to leave the profession.

As the heterogeneity was high, I² = 99.22, Q = 2048.85, df = 16, p < .001, it was appropriate to assess if potential moderators reduced the heterogeneity (Higgins & Green, 2011). Moderation analyses were conducted using meta-regression to assess if the covariates mean ERI, the female percentage of N, or occupation (health workers, educators, others) reduced the between-study heterogeneity. The meta-regression findings (k = 13, with moderator information) revealed that when the covariates were added to the model simultaneously, I² was significantly reduced R² = .53, I² = 97.92, Q = 336.42, df = 7, p < .001. The magnitude of stress (mean ERI, p = .008) attenuated the association between ERI and intention to leave work. Specifically, the association between higher ERI and intention to leave work was higher within samples that reported a higher magnitude of stress. Gender, p = .81 and occupation (educators, health workers, others), p = .61 did not attenuate the association between ERI and intention to leave work.

Overcommitment and intention to leave work

There was a significant association between higher OC and a higher intention to leave work (k = 13, r = .26, p < .001). There was no difference in effect size between outcomes assessing ITLJ (k = 7, r = .28, p < .001) or ITLP (k = 6, r = .22, p < .001), Q_between = 0.95, df = 1, p = .33 (Figure 3). Evaluation of a funnel plot of results for the combined outcomes indicated more studies toward the right, suggesting the possibility of publication bias. Egger’s regression however indicated that asymmetry was not significant (p (2-tailed) = .58). Additionally, Duval and Tweedie’s trim and fill method indicated it was unlikely there were missing studies. The Fail-Safe N analysis revealed that the meta r finding was robust; 5,249 more studies with an r value of zero would be required to render the current association non-significant.

Figure 3. Forest Plot of the Association Between Overcommitment and Intention to Leave.

Note. ITLJ = intention to leave the job, ITLP = intention to leave the profession.

As the heterogeneity was high, I² = 97.41, Q = 463.94, df = 12, p < .001, it was appropriate to assess if potential moderators reduced the heterogeneity. Moderation analyses were conducted to assess if the covariates mean ERI, or female percentage of N, or occupation (health workers, educators, others) reduced the between-study heterogeneity. Meta regression analyses were conducted singularly for each moderator (mean ERI (k = 8), the female percentage of N (k = 12), and occupation (health workers, educators, others, k = 13)) as there were not enough studies for all to be entered simultaneously. The meta regression findings revealed that gender, R² < .01, I² = 97.41, Q = 386.61, df = 10, p < .001, mean ERI, R² < .001, I² = 98.23, Q = 339.73, df = 6, p < .001 or occupation, R² = < .01, I² = 97.97, Q = 443.28, df = 9, p < .001 did not moderate the association between OC and intentions to leave work.

Effort and intention to leave work

Higher perceptions of work efforts were associated with an intention to leave work (k = 16, r = .25, p < .001). There was no difference in effect size between outcomes assessing ITLJ (k = 10, r = .27, p < .001) or ITLP (k = 6, r = .19, p = .001), Q_between = 2.04, df = 1, p = .15 (Figure 4). Evaluation of a funnel plot of results for the combined outcomes indicated more studies toward the right, suggesting the possibility of publication bias. Egger’s regression however indicated that asymmetry was not significant (p (2-tailed) = .73). Additionally, Duval and Tweedie’s trim and fill method indicated it was unlikely there were missing studies. The Fail-Safe N analysis revealed that the meta r finding was robust; 8,655 more studies with an r value of zero would be required to render the current association non-significant.

Figure 4. Forest Plot of the Association Between Effort and Intention to Leave.

Note. ITLJ = intention to leave the job, ITLP = intention to leave the profession.

As the heterogeneity was high, I² = 96.68, Q = 451.24, df = 15, p < .001, it was appropriate to assess if potential moderators reduced the heterogeneity. Meta regression analyses were conducted singularly for each moderator; mean ERI (k = 9), the female percentage of N (k = 15), and occupation (health workers, educators, others, k = 16) as there were not enough studies for all to be entered simultaneously. The meta-regression findings revealed I² was significantly reduced when gender, R² = .15, I² = 96.17, Q = 339.51 df = 13, p = .01 was added to the model. Specifically, the association between higher efforts and intention to leave work was higher within samples that included more female employees (p = .01). Neither mean ERI, (p = .33), nor occupation, (p = .31), attenuated the relationship between effort and intention to leave work.

Reward and intention to leave work

Lower perceptions of work rewards were significantly associated with an intention to leave the workplace (k = 15, r = -.41 p < .001). The effect size between outcomes assessing ITLJ (k = 10, r = -.44, p < .001) and ITLP (k = 5, r = -.26, p = .001) differed, Q_between = 5.81, df = 1, p = .016 (Figue 5). As such, the outcomes were treated separately in moderation analyses. Evaluation of a funnel plot of results for both outcomes indicated more studies toward the left, suggesting the possibility of publication bias. Egger’s regression however indicated that asymmetry was not significant (p (2-tailed) > .999). Additionally, Duval and Tweedie’s trim and fill method indicated it was unlikely there were missing studies. The Fail-Safe N analysis revealed that the meta r finding was robust; 20,419 more studies with an r value of zero would be required to render the current association between rewards and ITL non-significant.

Figure 5. Forest Plot of the Association Between Reward and Intention to Leave.

Note. ITLJ = intention to leave the job, ITLP = intention to leave the profession.

As the heterogeneity was high, I² = 99.23, Q = 1801.28, df = 14, p < .001, it was appropriate to assess if potential moderators for each measure of ITLJ and ITLP reduced the heterogeneity.

Reward and ITLJ. Meta-regression analyses were conducted to assess if the female percentage of N (k = 9), ERI mean (k = 5), or occupation (health workers, educators, others, k = 10) moderated the association between rewards and ITLJ. The meta-regression findings revealed that I² was not reduced when gender, R² < .01, I² = 94.76, Q = 133.50, df = 7, p = .86, ERI mean, R² = < .04, I² = 96.21, Q = 79.07, df = 3, p = .56, or occupation R² = .02, I² = 93.37, Q = 90.47, df = 6, p = .72 were added to the model.

Reward and ITLP. Assessing moderators in the ITLP studies showed that gender moderated the association between rewards and ITLP. Specifically, studies with a greater proportion of males were more likely than females to report an intention to leave the profession when rewards were low, k = 5, R² = .68, I² = 90.86, Q = 32.84, df = 3, p = .02. Neither ERI mean k = 4, R² < .001, I² = 90.20, Q = 20.42, df = 2, p = 61, or occupation type k = 5, R² < .001, I² = 99.46, Q = 368.57, df = 2, p = .72, moderated the association with ITLP.

Occupation type and intention to leave work

Given the variety of occupations included in the meta analysis, sub-group analysis by occupation type was conducted. Occupation type moderated the association between OC and ITL (Q = 9.38, p = .03). Specifically, health employees with higher OC were more likely to report an ITL than academics (Q = 9.29, p = .01) and “other” (Q = 6.98, p = .01) employees. Academics, educators, and others did not differ in the magnitude of association between OC and ITL. Occupation did not moderate the association of ERI, effort or reward with ITL (Table 3). Finally, a subgroup assessment of differences in effect size by outcome (ERI, OC, effort, reward) revealed that reward had a stronger association with intention to leave work than ERI (Q = 102.69, p < .001), OC (Q = 115.69, p < 0.01), and effort (Q = 118.80, p < .001). All other contrasts were non-significant.

Table 3. Associations Between Effort-Reward Imbalance, Overcommitment, Efforts, and Rewards and Intentions to Leave by Occupation.

Association by occupation		r	95% CI	I^2	K	N	Contrast
ERI	Academics	.29**	.20, .38	0.00	1	421
	Health workers	.37**	.24, .48	99.55	10	61,756
	Educators	.37**	.12, .58	92.93	2	682	Q = 1.37
	Other	.29**	.17, .40	93.33	4	9853	p = .71
OC	Academics	.15**	.09, .20	2.96	3	1337
	Health workers	.32**	.22, .42	98.87	6	49,413
	Educators	.19	-.01, .37	0.00	1	98	Q = 9.38
	Other	.18**	.13, .22	0.00	3	2892	p = .03
Efforts	Academics	.17*	.01, .33	82.63	2	916
	Health workers	.28**	.21, .36	98.22	8	46,587
	Educators	.22**	.13, .31	42.97	3	780	Q = 4.34
	Other	.17**	.07, .26	91.62	3	9465	p = .28
Rewards	Academics	-.37**	-.42, -.31	0.00	2	916
	Health workers	-.36**	-.48, -.22	99.59	8	47,504
	Educators	-.34**	-.53, -.11	89.90	3	780	Q = 3.36
	Other	-.45**	-.52, -.37	90.34	2	8676	p = .34

Note. * = p < .05, ** = p < .001. ERI = effort-reward imbalance, OC = overcommitment,

Academics [academics], Health workers [nurses, occupational therapists, physician assistants, health professionals], Educators [teachers, school counsellors], Others [managers, military reservists, attorneys, mixed samples].

Discussion

The current meta-analysis synthesised the associations between ERI, effort, reward, and OC, with turnover intentions in a sample of 73,671 employees. Our analysis revealed that higher ERI, effort, and OC and lower rewards were associated with higher turnover intentions. Rewards (r = -.413, CI: -.46, -.37) were more strongly associated with ITL than ERI (r = .327, CI: .27, .38), OC (r = .257, CI: .21, .31), and effort (r = .250, CI: .20, .30). Additionally, samples with more females were more likely than males to have an ITL when efforts were high, and samples with more males were more likely than females to have an ITLP when rewards were low. The association between OC and ITL was stronger in health workers than in other occupations. Finally, the association between ERI and ITL was strongest in samples that reported higher workplace stress. With the exception of the rewards to ITL association, the effect sizes did not differ between studies with ITLJ or ITLP outcomes.

Workplace stress (ERI, effort and rewards) and ITL

Our findings showed that an ERI, high efforts, and a lack of rewards were all individually related to ITL and suggest that rewards play a stronger role in the ERI-ITL relationship than workplace efforts. Although subgroup analysis of the outcomes of ITLJ and ITLP showed no difference in the ERI-ITL effect, the association of rewards-ITLJ was greater than the effect size for the reward-ITLP. While leaving a current job for another organisation may appear to have limited impact on a profession, earlier research regarding turnover in nursing has described a withdrawal process whereby nurses may first leave their ward, followed by leaving the organisation and finally abandoning the profession (Krausz et al., 1995; Morrell, 2005).

The present meta-analysis used the composite scales of rewards in the reward-ITL association, however studies that considered subscales of rewards showed varying findings. Perceptions of career advancement and promotional prospects are negatively related to an ITL (Boag-Munroe et al., 2016; Leone et al., 2015; Li et al., 2013), however, job security is not related (Kinman, 2016; Li et al., 2013). Although there is some evidence that salary is related to an ITL (Almalki et al., 2012; He et al., 2020; Kinman, 2016; Tei-Tominaga et al., 2018), the vast majority of studies suggest it is the perceptions of poor pay (Gardulf et al., 2005; Kudo et al., 2006) rather than objective pay itself that lead to higher ITL, and this is supported by meta-analytic evidence (Kim & Kao, 2014). Esteem has also been shown to be an important factor related to ITL (Kinman, 2016; Li et al., 2013), which may provide a more malleable avenue to improve rewards than improved pay and promotion opportunities. For instance, experimental ERI research has shown how positive verbal feedback targeting esteem can lead to adaptive physiological responses to a work stressor (Brooks et al., 2019).

Moderators of the workplace stress with ITL relationship

The meta-analysis demonstrated that gender did not moderate the ERI, or OC to ITL relationship. This is at odds with previous research showing men are more likely to report an ITL than women (Derycke et al., 2010; He et al., 2020), with men under high strain, twice as likely to leave compared to men under low strain conditions (Söderberg et al., 2014, N = 940). Additionally, the effort to ITL effect was stronger for women than men, but the low reward to ITLP effect was stronger for men than women. Although this finding aligns with findings that men and women may differ in the value they attach to rewarding aspects of their work or experiences of job demands (Derycke et al., 2010), the finding that women had higher ITL than men when efforts were high is unusual. Specifically, in an examination of the job demands-resources (Demerouti et al., 2001) paradigm, the job demands→strain→turnover relationship was stronger for men, and the job resources→job-satisfaction→turnover relationship was stronger for women (Hoonakker et al., 2013, N = 624). Potentially, the use of the reward measure in the ERI model, the smaller sample in the Hoonakker study, or the use of meta-regression which can increase Type I error (Higgins & Thompson, 2004), may explain the discrepancy. Neither occupation type nor the average level of stress within the study (i.e. mean ERI) moderated the association of ERI, effort or reward with ITL. However, occupation type did moderate the association between OC and ITL. Specifically, health workers with higher OC were more likely to report an ITL than other workers (academics, managers, military reservists, attorneys, mixed samples).

Maladaptive coping (OC) and ITL

In the current meta-analysis, OC was found to be associated with turnover intentions. The magnitude of this association was moderate for health workers and small for other occupations. Previous research has provided mixed support for the association of OC (and maladaptive coping more generally) with an ITL. In the large study by Hasselhorn et al. (2004), the OC-ITL relationship was present and in the expected direction, albeit weak, whereas other research failed to support the OC-ITL relationship (e.g. Derycke et al., 2010; Kinnunen et al., 2008; Tominaga & Miki, 2011).

Overcommitment is related to reduced job satisfaction (Huyghebaert et al., 2018) and there is evidence to suggest that job satisfaction moderates the association between OC and burnout (Avanzi et al., 2014), with both burnout (Geurts et al., 1998) and low job satisfaction (Al-Hamdan et al., 2017, Kim & Kao, 2014) strongly related to an ITL. As maladaptive coping is related to job dissatisfaction and ITL (Cheval et al., 2019), this provides the impetus to further explore the moderating or mediating pathways between OC and job satisfaction that may lead to an ITL. It was not possible to assess the proposed moderating effect of OC on ERI with an ITL in the present meta-analysis as too few studies provided the necessary data for all measures. Understanding the underlying mechanisms of these pathways will provide a clearer picture for intervention.

Strengths & limitations

To the researchers’ knowledge, this study is the first to meta-analyse the relationship between measures of the ERI model and an ITL. One strength of this study is that it is based on a large sample, with 23 primary studies and an overall sample size of 73,761. As 22 of 23 studies included in this meta-analysis were cross-sectional in nature, these findings need to be confirmed with prospective investigations before inferring causality between constructs. It is also important to note, that the moderating effect of gender on effort with ITL and rewards on ITLP may not be robust as it was based on a small proportion of male employees, with most studies recruiting a high proportion of female employees. Additionally, as intentions to leave and not actual turnover were assessed, caution should be taken interpreting our findings as evidence of ERI resulting in turnover behaviour. While the intention-behaviour gap (Sheeran & Webb, 2016) may restrict the interpretation of our findings as evidence of the effect of ERI predicting actual turnover, ITL is widely accepted as a strong predictor of actual turnover (Siong et al., 2006). There is a growing interest in physiological indices in workplace stress research, however, the current body of ERI-ITL literature has relied on self-reported evidence. Future research may be encouraged to assess if physiological measures of stress improve the prediction of an ITL the workplace.

Implications

In comparing the relative contributions of the ERI model components, our findings showed low perceived rewards to be the stronger risk factor for ITL. Based on these data, it would appear that for those reporting an ERI that increasing rewards rather than reducing job demands may be more likely to reduce ITL. The few studies that have prospectively considered what improves an ITL suggest that an increased intention to improve professional capabilities reduces an ITL (Chang et al., 2019; Parry, 2008). Of particular interest are the findings from Chang et al., who report that improved work capabilities mediate the association between outcome expectation (salary and future benefits) with a reduced ITL the profession. When this research is considered alongside the present findings, the continued prospective examination of the impact of job rewards (e.g. improved social support, salary and benefits, esteem) on an ITL appears justified- especially for male workers.

Conclusion

Although the ERI model was developed to predict adverse health outcomes, our study has demonstrated its usefulness in predicting ITL the workplace. Across a range of occupations, high ERI, OC, and efforts, and low rewards were individual risk factors for ITL. Considering the findings together, the ERI-ITL relationship is, for this large sample, driven by a perception of low rewards in the workplace. The findings from this study align with other reviews of the relationship between high workplace stress with ITL. Unlike these earlier reviews, however, the use of the ERI model in the present investigation provides an evidence-based avenue for intervention that is gender-specific (i.e. reduce demands for women, increase rewards for men) and may assist in reducing job turnover.

Disclosure statement

No potential conflict of interest was reported by the author(s).