Social stressors and irritability over time: Modeling the relationship via growth mixture methodology
Abstract
Most studies on stress at work and its associations with well-being and health are cross-sectional (Sonnentag & Frese, 2003). The number of longitudinal studies is growing and many of them show effects of job stress (e.g., Belkic et al., 2004; De Lange et al., 2003). Most longitudinal studies, however, involve only two waves (Zapf et al, 1996), leaving open how conditions developed in between these measurements (Garst et al., 2000).
Values obtained from multiple measurements may be approached from different perspectives. (1) One may ask if mean level of stressors across waves predicts outcomes; this approach implies that having high stressors at one time may not have a strong impact but having consistently high levels should, reflecting cumulative effects. (2) One may ask if changes over time predict outcomes. This approach implies that positive changes predict better outcomes than negative changes, even at similar mean levels across time.
Studies measuring working conditions repeatedly sometimes report cumulative effects (e.g., Amick et al., 2002; Chandola et al, 2005; DeLange, et al., 2002; Kivimäki et al., 2006; Landsbergis et al., 2003); however, changes over time have also found to be important (e.g., Godin et al., 2005; Schnall et al., 1998;).
Latent growth mixture modeling (GMM; Muthén & Muthén, 2000) allows for investigating groups with different trajectories (Garst et al. 2000). We investigated: 1) if there are different trajectories of social stressors, 2) how these trajectories influence irritability in T5, controlling for its baseline value.
Method:
We used data of the N = 308 participants in all five waves (spanning 10 years; Nt1 = 1394) of the “Work Experience and Quality of Life in Switzerland” study (Semmer et al. 2005).
Social stressors were measured with six items from the social stressors scale by Frees and Zapf (1987); irritability (e.g. reacting in an irritated manner towards others) was measured with a short version of the irritation scale by Mohr et al. (2006); alphas were above α = .70. Using GMM we estimated latent growth curves with intercept (I) and slope (S), formed by social stressors from t1 to t5, and identified different trajectories. We then tested if irritability at t5 differed between those trajectories, controlling for the initial value of irritability (ANCOVA).
Results:
Based on the Bayesian information criterion, the bootstrapped parametric likelihood ratio test (Jung & Wikrama, 2008), and entropy values (Celeux & Soromenho, 1996), we decided for a three-class model (cf. Muthén, 2003). The largest class represents individuals with stable low social stressors over time (n = 274, 88.9%). Their estimated mean for social stressors at t1 was 1.84; average growth rate was 0.006 (ns.). The smallest class (n = 12; 3.9%) was the increasing social stressors class; it started with a low level of social stressors (I=1.59), which rapidly increased over time (S = .18). The third class, the decreasing social stressors class (n = 22; 7.1%), started with quite a high level of social stressors (I = 3.04), which constantly decreased over time (S = -. 08) to a mean of 2.23 in t5.
The effect of class membership on irritation_t5 was significant; F (2, 302) = 6.219, p<.05. Planned contrasts revealed that increasing social stressors were significantly associated with a higher level of irritation_T5, compared to stable low social stressors; t (2,302) = -3.446, p>.05, and to decreasing level of social stressors; t (2, 302) = 2.178, p>.05.
Even though the mean level of social stressors across all waves was higher for the decreasing class as compared to the stable low class, the two groups did not differ in their irritability at t5.
Theoretical implications and conclusions
As expected, there were different developments of social stressors over time. Increasing social stressors predicted irritability levels at t5 that were high, and significantly different from the other two classes, showing a negative impact on well-being of increasing social stressors over time. This result is remarkable given the small size of the increasing class (n = 12); having a significant difference with such a low power indicates quite strong effects. Conversely, the lack of a significant difference between the decreasing class and the stable low class suggests that change over time is more important than mean levels across time; however, given the small size of the decreasing class, power is low, and this result awaits replication.
The low stable class was the largest by far, and the increasing social stressors class was rather small, suggesting a low level of social tension in sample. Note that the attrition analysis shows lower social stressors for „stayers“ (Kälin et al., 2000).
Multiple measurement are a strength of our study; weaknesses are the reliance on self-report only and the high drop-out (mainly between Time 1 and Time 2; cf. Kaelin et al. 2000).
Our results are preliminary but promising; they will be complemented by including other stressors as well as other well-being variables.
Values obtained from multiple measurements may be approached from different perspectives. (1) One may ask if mean level of stressors across waves predicts outcomes; this approach implies that having high stressors at one time may not have a strong impact but having consistently high levels should, reflecting cumulative effects. (2) One may ask if changes over time predict outcomes. This approach implies that positive changes predict better outcomes than negative changes, even at similar mean levels across time.
Studies measuring working conditions repeatedly sometimes report cumulative effects (e.g., Amick et al., 2002; Chandola et al, 2005; DeLange, et al., 2002; Kivimäki et al., 2006; Landsbergis et al., 2003); however, changes over time have also found to be important (e.g., Godin et al., 2005; Schnall et al., 1998;).
Latent growth mixture modeling (GMM; Muthén & Muthén, 2000) allows for investigating groups with different trajectories (Garst et al. 2000). We investigated: 1) if there are different trajectories of social stressors, 2) how these trajectories influence irritability in T5, controlling for its baseline value.
Method:
We used data of the N = 308 participants in all five waves (spanning 10 years; Nt1 = 1394) of the “Work Experience and Quality of Life in Switzerland” study (Semmer et al. 2005).
Social stressors were measured with six items from the social stressors scale by Frees and Zapf (1987); irritability (e.g. reacting in an irritated manner towards others) was measured with a short version of the irritation scale by Mohr et al. (2006); alphas were above α = .70. Using GMM we estimated latent growth curves with intercept (I) and slope (S), formed by social stressors from t1 to t5, and identified different trajectories. We then tested if irritability at t5 differed between those trajectories, controlling for the initial value of irritability (ANCOVA).
Results:
Based on the Bayesian information criterion, the bootstrapped parametric likelihood ratio test (Jung & Wikrama, 2008), and entropy values (Celeux & Soromenho, 1996), we decided for a three-class model (cf. Muthén, 2003). The largest class represents individuals with stable low social stressors over time (n = 274, 88.9%). Their estimated mean for social stressors at t1 was 1.84; average growth rate was 0.006 (ns.). The smallest class (n = 12; 3.9%) was the increasing social stressors class; it started with a low level of social stressors (I=1.59), which rapidly increased over time (S = .18). The third class, the decreasing social stressors class (n = 22; 7.1%), started with quite a high level of social stressors (I = 3.04), which constantly decreased over time (S = -. 08) to a mean of 2.23 in t5.
The effect of class membership on irritation_t5 was significant; F (2, 302) = 6.219, p<.05. Planned contrasts revealed that increasing social stressors were significantly associated with a higher level of irritation_T5, compared to stable low social stressors; t (2,302) = -3.446, p>.05, and to decreasing level of social stressors; t (2, 302) = 2.178, p>.05.
Even though the mean level of social stressors across all waves was higher for the decreasing class as compared to the stable low class, the two groups did not differ in their irritability at t5.
Theoretical implications and conclusions
As expected, there were different developments of social stressors over time. Increasing social stressors predicted irritability levels at t5 that were high, and significantly different from the other two classes, showing a negative impact on well-being of increasing social stressors over time. This result is remarkable given the small size of the increasing class (n = 12); having a significant difference with such a low power indicates quite strong effects. Conversely, the lack of a significant difference between the decreasing class and the stable low class suggests that change over time is more important than mean levels across time; however, given the small size of the decreasing class, power is low, and this result awaits replication.
The low stable class was the largest by far, and the increasing social stressors class was rather small, suggesting a low level of social tension in sample. Note that the attrition analysis shows lower social stressors for „stayers“ (Kälin et al., 2000).
Multiple measurement are a strength of our study; weaknesses are the reliance on self-report only and the high drop-out (mainly between Time 1 and Time 2; cf. Kaelin et al. 2000).
Our results are preliminary but promising; they will be complemented by including other stressors as well as other well-being variables.
Date Issued
2013-05-16
Publication Type
Conference Item
Subject(s)
Language(s)
en
Author(s)
Franziska, Tschan |
Additional Credits
Access(Rights)
metadata.only