Role ambiguity (RA) is defined as the lack of clarity in understanding the actions to be taken to achieve the proposed individual goals.1 RA makes employees doubt how their objectives can be achieved and how their performance will be assessed, causing a negative relationship between RA and job performance.2-5 Hence, we consider RA as a key job stressor that forces employees to invest effort in clarifying the ambiguity of their role and correspondingly increases their psychological ill-being. Indeed, many studies have revealed that RA is related to depression,6 emotional exhaustion,7 lower job satisfaction,2, 8 and other poor mental health outcomes.
However, it might be possible that RA may work not only as a job stressor but also as an amplifier of the association between job stressors and workers’ psychological ill-being. In line with this view, RA has been found to amplify the association between abusive supervision and job burnout9 as well as between job instability and psychological distress.10 RA is considered to have a negative effect on the motivational process in the Job Demands-Control (JD-C) model11 via a perception of increased job demands due to their uncertainty. Further, RA is considered to lower perceived control over work tasks if those tasks are ambiguous. Similarly, additional effort needed to clarify RA and ambiguity about the expected evaluation of job performance may lead to a deterioration in the balance between effort and reward within the framework of the Effort-Reward Imbalance (ERI) model.12
In this study, we attempted to provide new insights into the relevance of RA in occupational health in two ways. First, we examined how RA modified the associations between key job stressors (i.e., high job demands, low job control, high effort, and low reward), which are derived from the JD-C and ERI models, and workers’ psychological ill-being (i.e., psychological distress [PD] and job dissatisfaction [JD]). Based on the observations in previous studies,9, 10 we predicted that RA would amplify the adverse impact of job stressors. Unlike previous studies, however, we compared the modifying effects on key job stressors within the same analytic framework.
Second, we conducted an analysis using data from the same participants collected at different points to address this issue, in contrast to a majority of previous studies, which relied on cross-sectional data. Specifically, we estimated fixed-effects (FE) models, which control for a participant's attributes, both observed and unobserved.13, 14 The associations observed from the cross-sectional data cannot be free from biases due to these factors, as suggested by previous FE model studies,15, 16 especially because RA, job stressors, and psychological ill-being are all subjectively evaluated, presumably leading to overestimation of their correlations.
2 METHODS 2.1 Study sampleWe used panel data from eight survey waves of an occupational cohort study on social class and health in Japan (Japanese Study of Health, Occupation, and Psychosocial Factors Related Equity [J-HOPE]). The first wave was conducted from April 2010 to March 2011; the following waves were conducted approximately one year after the first wave. The eighth wave was conducted between April 2017 and March 2018. The study population consisted of employees working for 13 firms. The surveyed firms covered 12 industries and participated in three to eight waves. The original sample consisted of 47 960 observations from 14 388 individuals. The response rates were 77.0%, 81.6%, 78.6%, 67.5%, 63.9%, 64.6%, 64.2%, and 64.8% in the first to eighth waves, respectively. After removing 4007 observations in one industry (code 11, transportation industry) over the fourth and eighth waves (because they were asked only about their experiences in sick leave) and respondents missing key variables of RA, PD, JD, and/or job stressors, we ended up utilizing 41 962 observations from 13 811 individuals (10 269 men and 3542 women). The structures of the firms, waves, and participants in the study sample are summarized in Table S1.
The Research Ethics Committee of the Graduate School of Medicine and Faculty of Medicine, The University of Tokyo (No. 2772), Kitasato University Medical Ethics Organization (No. B12-103), and the Ethics Committee of Medical Research, University of Occupational and Environmental Health, Japan (No. 10-004 and H26-115) reviewed and approved the aims and procedures of the present study. This study was conducted with the J-HOPE dataset as of June 1, 2021.
2.2 MeasuresTable 1 summarizes the key measures obtained from the survey and the definitions of the binary variables that were used in the statistical analysis. For the binary variables of high RA, high job demands, low job control, high effort, and low reward, we used the sample means of their corresponding measures as the cut-off points. More detailed explanations are provided below.
TABLE 1. Summary of key measures in the survey and the definition of the binary variables Measures in the survey Cronbach's alpha Score Definition of the binary variable Range M SD Role clarity 0.88 6–42 29.7 6.0 High role ambiguity Scorea < M K6 score 0.90 0–24 5.5 5.0 Psychological distress Score ≧ 13 Job satisfaction N.A. 1–4 2.6 0.8 Job dissatisfaction Score = 1 (dissatisfied) Job demands 0.69 12–48 32.8 5.4 High job demands Score > M Job control 0.78 24–96 65.7 10.1 Low job control Score < M Effort 0.78 3–12 7.9 1.9 High effort Score > M Reward 0.76 7–28 18.1 3.0 Low reward Score < M 2.2.1 Role ambiguity (RA)We measured RA based on the Japanese version of the National Institute for Occupational Safety and Health Generic Job Stress Questionnaire (NIOSH-GJSQ).17, 18 The internal consistency reliability and validity of the Japanese version of the NIOSH-GJSQ has been reported to be acceptable.18 Respondents were asked to assess the accuracy of each of the six statements about their role clarity, such as “I feel certain about how much authority I have” on a seven-point scale (1 = very inaccurate to 7 = very accurate; see Table S2 for the full questionnaire). Cronbach's alpha for this sample was 0.88. We summed up the scores (range: 6–42; lower scores indicating higher levels of RA) and constructed a binary variable for high RA by allocating “1” to the score below the sample mean (29.7) and “0” to others.
2.2.2 Psychological distress (PD) and job dissatisfaction (JD)We considered PD and JD as workers’ psychological ill-being measures. To measure PD, we used Kessler 6 (K6) scores19, 20 as the reliability and validity have been demonstrated previously in a Japanese population.21, 22 From the survey, we first obtained the respondents’ assessments of psychological distress using a six-item psychological distress questionnaire: “During the past 30 days, how often did you feel (a) nervous, (b) hopeless, (c) restless or fidgety, (d) so depressed that nothing could cheer you up, (e) that everything was an effort, and (f) worthless.” This questionnaire was rated on a five-point scale (0 = none of the time to 4 = all of the time). The sum of the reported scores was then calculated (range: 0–24; higher K6 scores indicating higher levels of psychological distress). Cronbach's alpha for this sample was 0.90. A binary variable of psychological distress was constructed and defined as K6 ≥ 13, as this cutoff indicator has been found to indicate serious psychological distress in the Japanese population.21, 22 Regarding job satisfaction, the survey asked questions using a four-point scale (1 = dissatisfied, 2 = somewhat dissatisfied, 3 = somewhat satisfied, and 4 = satisfied). A binary variable of JD was constructed by allocating “1” to answers equaling 1, and “0” to others.
2.2.3 Job demands and controlWe utilized the items investigating job demands and control from the Japanese version of the Job Content Questionnaire (JCQ).23 It is based on the JD-C model,11 and includes scales related to job demands (five items) and job control (nine items) rated on a four-point scale (1 = strongly disagree to 4 = strongly agree). The internal consistency, reliability, and validity of the Japanese version of the JCQ have been shown to be acceptable.24 In the present sample, Cronbach's alpha coefficients were 0.69 and 0.78 for job demands and control scales, respectively. Following the JCQ User's Guide,23 we summarized the responses to these items into single indices of job demands (range: 12–48) and control (range: 24–96). Finally, we used their sample means (32.8 and 65.7, respectively) as the cut-off points for the binary variables that classified each worker as having either high or low job demands and control.
2.2.4 Effort and rewardTo assess effort and reward, we utilized data collected from a simplified Japanese version of the Effort-Reward Imbalance Questionnaire (ERIQ). The ERIQ was developed based on the ERI model,13 and its Japanese version and that of the simplified ERIQ25 used in the present study have been shown to have acceptable internal consistency, reliability, and validity scores.26, 27 The simplified version includes sub-scales for effort (three items) and reward (seven items) rated on a four-point scale (1 = strongly disagree to 4 = strongly agree). Cronbach's alpha coefficients were 0.78 and 0.76 for the effort and reward scales, respectively. We summed the responses into single indices for effort (range: 3–12) and reward (range: 7–28). Subsequently, we used their sample means (7.9 and 18.1, respectively) as the cut-off points for the binary variables classifying each worker as exhibiting either high or low effort and rewards.
2.2.5 Potential confoundersAs potential confounders, we considered gender, age (i.e., 20s, 30s, 40s, 50s, and 60s), educational attainment (i.e., high school or below, junior college, college, and graduate school), household income, job category (i.e., managerial, manual, non-manual, and others), health behavior (i.e., smoking, daily alcohol consumption, and physical inactivity), and firm codes (i.e., 1–13). Regarding household income, we divided reported household income by the square root of the number of household members to adjust for household size,28 and constructed binary variables for each quartile. We also constructed binary variables of “unanswered” for age, educational attainment, and household income. Among these variables, gender, educational attainment, and firm codes were time-invariant and were automatically removed from the FE regression.
2.3 Statistical analysisFollowing the descriptive analysis, which examined pairwise correlations across key variables, we estimated three linear probability models28, 29 (LPM, models 1–3), all of which linearly regressed the binary variable of PD or JD on RA, four job stressors, and potential confounders. Model 1 was a pooled cross-sectional regression model. Model 2 was a FE regression model using data from the same participants collected at different points in three to eight waves depending on the firms, as summarized in Table S1. Model 3 included the interaction terms between RA and each of the four job stressors. The estimated coefficient of the interaction term with each stressor indicates the magnitude of the modifying effect of RA on the association between each stressor and PD or JD. After regression, we calculated the sum of the estimated coefficient of each job stressor and that of its interaction term with RA to measure the RA-modified association between each job stressor and PD or JD.
In the FE models, all variables were mean-centered for each participant over the estimation period, which varied from three to eight waves depending on the participant. Unlike the pooled cross-sectional regression models, which used simply pooled data for individuals over the estimation period, FE models controlled for a participant's time-invariant attributes, both observed and unobserved, which allowed us to focus exclusively on within-participant variations.29 We further chose LPMs, which are known to provide good estimates of the partial effects of the independent variables on the response probability,29, 30 rather than probit or logistic models for two practical reasons. First, the estimated coefficient of the interaction term can be directly interpreted in LPMs.31 Second, FE models concentrate on within-participant variations in outcome and hence would remove participants who reported no change in PD (or JD), which was measured by its binary variable, over the estimation period.32
We checked the robustness of the estimation results by replacing binary variables for PD and JD with continuous variables for K6 scores (range: 0–24) and job dissatisfaction scores (range: 1–4; reversing the original order to make higher scores indicate higher dissatisfaction). We used the Stata Software Package (release 17) to perform all statistical analyses.
3 RESULTS 3.1 Descriptive analysisTable 2 summarizes the key features of the study sample, dividing the respondents into those with high PA and those with low RA. As seen in this table, higher RA was associated with lower educational attainment, non-managerial jobs, higher levels of job stressors, PD, JD, and lower household income. Table 3 also confirms a high correlation between RA and job stressors, PD, and JD.
TABLE 2. Key features of the respondents in the survey by role ambiguity Role ambiguity All High Low Gender Males 31 256 (74.5) 13 876 (69.3) 17 380 (79.2) Females 10 706 (25.5) 6133 (30.7) 4573 (20.8) Educational attainment High school or below 16 349 (39.0) 8594 (43.0) 7755 (35.3) Junior college 7122 (17.0) 3774 (18.9) 3348 (15.3) College 14 098 (33.6) 5738 (28.7) 8360 (38.1) Graduate school 4341 (10.3) 1880 (9.4) 2461 (11.2) Job category Managerial 7403 (17.6) 1889 (9.4) 5514 (25.1) Manual 19 015 (45.3) 9621 (48.1) 9394 (42.8) Non-manual 9856 (23.5) 5476 (27.4) 4380 (20.0) Other 5688 (13.6) 3023 (15.1) 2665 (12.1) Health behavior Smoking 11 656 (27.8) 5327 (26.6) 6329 (28.8) Daily alcoholic consumption 11 750 (28.0) 4988 (24.9) 6762 (30.8) Physical inactivity 25 214 (60.1) 12 750 (63.7) 12 464 (56.8) Job stressor Job insecurity (high) 15 756 (37.5) 8964 (44.8) 6792 (30.9) Effort (high) 23 574 (56.2) 11 795 (58.9) 11 779 (53.7) Reward (low) 21 156 (50.4) 13 403 (67.0) 7753 (35.3) Job demand (high) 21 054 (50.2) 10 666 (53.3) 10 388 (47.3) Job control (low) 17 293 (41.2) 10 886 (54.4) 6407 (29.2) Psychological distress 3977 (9.5) 2867 (14.3) 1110 (5.1) Job dissatisfaction 3911 (9.3) 3165 (15.8) 746 (3.4) Age (years) M 41.5 (SD 10.6) M 41.5 (SD 10.5) M 42.3 (SD 10.5) Household income (annual, thousand JPY) M 4320 (SD 2144) M 4027 (SD 2003) M 4585 (SD 2231) N 41 962 20 009 21 953 TABLE 3. Pairwise correlation coefficients across key variables (1) (2) (3) (4) (5) (6) (7) (1) High role ambiguity 1 (2) High job demands 0.053 1 (3) Low job control 0.316 0.067 1 (4) High effort 0.060 0.452 0.079 1 (5) Low reward 0.256 –0.135 0.222 –0.132 1 (6) Psychological distress 0.158 0.118 0.182 0.124 0.077 1 (7) Job dissatisfaction 0.213 0.081 0.252 0.088 0.157 0.311 1 3.2 Regression resultsTable 4 presents the key estimation results obtained from models 1 to 3 to explain the probability of PD, with more detailed results provided in Table S3 in the Supplementary file. Model 1, which used pooled, cross-sectional data, confirmed that PD was positively associated with high RA and all job stressors; notably, high RA corresponded to a 4.8% (95% confidence interval [CI]: 4.3%–5.4%) higher probability of PD, compared to low RA. The magnitude of the association between RA and PD was similar to that for the four job stressors.
TABLE 4. Estimated associations with psychological distress (N = 41 962 observations from 13 811 individuals) Pooled cross-sectional Fixed effects Model 1 Model 2 Model 3 Coef. (95% CI) Coef. (95% CI) Coef. (95% CI) Main effects High role ambiguity 0.048 (0.043, 0.054) 0.032 (0.025, 0.039) −0.008 (–0.022, 0.006) High job demands A 0.038 (0.032, 0.045) 0.024 (0.017, 0.031) 0.014 (0.004, 0.023) Low job control B 0.030 (0.024, 0.036) 0.019 (0.011, 0.027) 0.014 (0.003, 0.025) High effort C 0.040 (0.034, 0.046) 0.029 (0.022, 0.037) 0.019 (0.009, 0.029) Low reward D 0.078 (0.072, 0.084) 0.049 (0.041, 0.056) 0.037 (0.027, 0.047) Interaction terms High role ambiguity ×High job demands a 0.021 (0.008, 0.034) ×Low job control b 0.009 (–0.004, 0.022) ×High effort c 0.021 (0.008, 0.035) ×Low reward d 0.024 (0.011, 0.037) Post-regression calculations High job demands with high role ambiguity A + a 0.035 (0.025, 0.045) Low job control with high role ambiguity B + b 0.023 (0.013, 0.033) High effort with high role ambiguity C + c 0.040 (0.030, 0.051) Low reward with high role ambiguity D + d 0.061 (0.051, 0.071) Abbreviation: CI, confidence interval.We observed the associations of PD with RA and job stressors in model 2, even after controlling for a participant's time-invariant attributes. However, the magnitude of the observed associations was somewhat attenuated compared to those in model 1, suggesting that the associations observed from cross-sectional data were overestimated. Although we did not report the results, the F test showed that the null hypothesis that individual-specific effects were equal to zero could be rejected (P < .001), and the Hausman test showed that the null hypothesis that individual-specific effects were not correlated with independent variables could be rejected (P < .001). The results of these tests confirmed that the FE model was preferred to pooled cross-sectional and random-effects models.
Model 3 showed that the coefficient of the interaction term with high RA was significantly positive for high job demands, high effort, and low reward. For example, the coefficient of the interaction between high job demands and high RA was 2.1% (95% CI: 0.8%–3.4%; denoted by “a” in the table). As seen in the bottom part of the table, post-regression calculations showed that a combination of high job demands and high RA added to the risk of PD by 3.5% (95% CI: 2.5%–4.5%; denoted by “A + a”), compared with 1.4% (95% CI: 0.4–2.3; denoted by “A”) for a combination of high job demands and low RA (denoted by “A”), both using low job demands as a reference. These results indicated that high RA amplified the association between high job demands and PD by approximately 2.5 times (=3.5%/1.4%). Such an amplifying effect of PD was observed for high effort and low reward, while it was non-significant for low job control. Meanwhile, the estimated coefficient of high RA became slightly negative and non-significant, suggesting that the association between high RA and PD was mainly through RA’s amplifying effects on the association between job stressors and PD.
Figure 1 graphically illustrates the amplifying effect of RA for each job stressor to help understand the estimation results in Table 4. For each stressor, except for job control, the line for high RA has a greater slope than that for low RA, reflecting the RA’s amplifying effect on the association between that stressor and PD. The line for high RA is also located above that for low RA for each job stressor, reflecting RA’s amplifying effect on the associations between the other three stressors and PD.
The probability of psychological distress corresponding to a combination of different levels of each job stressor and role ambiguity†
Table 5 presents the estimation results obtained by replacing PD with JD as a dependent variable, with more detailed results provided in Table S4. We obtained results similar to those in Table 4 and confirmed RA’s amplifying effect, except for job control. Tables S5 and S6 present the detailed estimation results for the continuous variables of the K6 and JD scores, respectively. The results in these tables were similar to those in Tables 4 and
留言 (0)