Social stressors at work, irritation, and depressive symptoms: accounting for unmeasured third variables in a multi-wave study.(Statistical Data Included)

In work and organizational stress research, there are theories that propose complex relationships among variables. In such models, stressors and health outcomes are often mediated by one or more variables (see, e.g., Kahn & Byosiere, 1992; McGrath, 1976). There are many studies on moderator effects. In particular, control and social support are used as moderators (Kahn & Byosiere, 1992) but also other variables such as self-esteem (Jex & Elacqua, 1999), locus of control (Newton & Keenan, 1990; Parkes, 1991), a type A behaviour (Newton & Keenan, 1990), and negative affectivity (Heinisch & Jex, 1997; Moyle, 1995). However, there are only a few examples in which more complex mediation mechanisms were analysed in longitudinal studies. There are some earlier models on work and organizational stress (e.g. French & Kahn, 1962; Ivanicevich & Matterson, 1980; Marshall & Cooper, 1979). However, these models were non-specific in that mediating variables (short-term stress reactions) and outcome variables (long-term stress reactions) were only listed, but more specific propositions were lacking. These models were at least implicitly based on the assumption that some strain variables mediate the effect of stressors on other longer-term strain variables.

Models involving more specific mediation mechanisms among strain variables can be found in the burnout literature (e.g. Maslach & Jackson, 1981; Schaufeli & Enzmann, 1998). Although no rigorous propositions were made regarding the particular stressors that may trigger the chain of stress reactions, the ordering of stress reactions was clearly described. Leiter and Maslach (1988) proposed that stressors first lead to emotional exhaustion, which then leads to depersonalization which, in turn, causes a reduced sense of personal accomplishment. In contrast, Golembiewski, Munzenrider, and Stevenson (1986) suggested that the sequence starts with depersonalization, followed by reduced personal accomplishment, and finally emotional exhaustion. In testing these models, Lee and Ashforth (1993) presented one of the few longitudinal studies which analysed such mediation mechanisms, and the results were in favour of the Leiter and Maslach model. Dignam and West (1988) presented another two-wave study, which analysed mediation among stress reaction. They investigated whether stressors first lead to burnout. In this study, burnout was a latent factor with emotional exhaustion and depersonalization as indicators. They hypothesized that when burnout symptoms occur, they cause poor health. Support for this hypothesis was only found in cross-sectional analyses, but not when

longitudinal models were applied.

A theoretical model that has not yet been subjected to empirical investigation was developed by Mohr (1986, 1991). In parts, this model is tested in the present study. Therefore, it is now described in more detail.

Mohr's (1986, 1991) model suggests that stressors at work are the starting points of a temporarily ordered sequence of different stress reactions which are causally connected. First, irritation emerges, which then leads to a decrease in self-esteem and an increase in anxiety. Anxiety leads to depressive symptoms and further reduces self-esteem, which also leads to an increase in depressive symptoms. Psychosomatic complaints are also part of the model suggested by Mohr, which is shown in Fig. 1. The crucial point in this model is that increases in different responses to stressors at work occur not only with varying delays, but that an increase in irritation causes an increase in anxiety and a decrease in self-esteem, which then subsequently cause an increase in depressive symptoms.

[FIGURE 1 OMITTED]

Mohr (1991) cited prior evidence suggesting that social isolation might be critical in the development of depressive symptoms. She further supposed that social conflicts are relevant in the development of irritation. This is in line with the proposition of others (e.g. Kuhl & Helle, 1994), who also mentioned social isolation as a precursor of depressive symptoms. Schonfeld (1992) found that confrontations initiated by insolent students were a precursor of depression in a sample of first-year female teachers. For professional-managerial women, Snapp (1992) reported that trouble with the boss or subordinates was positively related to depression. Social conflicts at work were also related to depression in studies by Heinisch and Jex (1997) and Karasek and Theorell (1990). Social isolation and social conflicts are examples of social stressors (Frese & Zapf, 1987). Social stressors may thus play a critical role in the development of irritation and subsequent depressive symptoms.

Compared to other work stressors such as time pressure, role conflict or role ambiguity, there is little research on social stressors in organizations. Social stressors consist of social animosities, conflicts with co-workers and supervisors, unfair behaviour, and a negative group climate. There is reason to believe that social stressors may have strong effects on strains. Zapf and Frese (1991) have shown that social stressors are relatively strong predictors of depressive symptoms in comparison to various task stressors. In a diary study by Schwartz and Stone (1993), 75% of all recorded work-related events that indiviudals assessed as harmful were related to negative social interactions with colleagues, supervisors, and clients. Social or interpersonal stressors were an important predictor of other psychological strain variables in the studies of Heinisch and Jex (1997), Keashly, Hunter, and Harvey (1997), Spector (1987), Spector and Jex (1998), and Zapf, Seifert, Schmutte, Mertini, and Holz (2001). Moreover, research on social support at work repeatedly demonstrated that lack of support has negative consequences, underscoring the importance of personal relationships in organizations in comparison with task-related and organization-related issues (e.g. Abdel-Halim, 1982; Ganster, Fusilier, & Mayes, 1986; LaRocco, House, & French, 1980). Social stressors and lack of support are not identical. Although social stressors and social support tend to be negatively correlated (e.g. Dormann & Zapf, 1999), there are exceptions to the rule. For example, a supervisor who pushes all the time could nevertheless be very supportive in getting the job done.

One can assume that there is a relationship between job stressors and social stressors. If there is time pressure, there may be a supervisor being responsible for it. In particular, one may associate role conflict and role ambiguity with the social system at work. However, most researchers consider them to be related to the task structure and the organization of work focusing on unclear or contradictory task goals. Therefore, these concepts are clearly distinct from social stressors comprising social conflicts, personal animosities or unfair behaviour.

The model of Mohr (1986, 1991) is quite complex and a huge sample size would be necessary to test the whole model in a longitudinal fashion. Therefore, only the relationships among social stressors, irritation, and depressive symptoms were investigated in the present study. This corresponds to the grey-shaded area in Fig. 1. Although the model does not include a direct effect of irritation on depressive symptoms, such an effect would be established if the two mediating variables anxiety and self-esteem were excluded from the model. This represents hypothesis 1 of the present study.

Hypothesis 1: Social stressors at work affect irritation, and irritation affects depressive symptoms.

Moreover, it is assumed that there is a pure mediation mechanism. That is, as long as irritation is included in a model, a direct effect of social stressors on depressive symptoms should be absent. This represents hypothesis 2 of the present study.

Hypothesis 2: Social stressors do not directly affect depressive symptoms.

In organizational stress research, authors have repeatedly commented on the methodological shortcomings of many studies (e.g. Frese & Zapf, 1988; Kasl, 1978; Spector, 1992). Among other things, authors have suggested (a) using longitudinal designs to analyse causal effects; (b) taking into account that there may be third variables responsible for the stressor-strain relationship; (c) taking reverse causation into account, that is, the possibility that some strains may affect the stressors perceived at one's job; and (d) considering the time lag necessary for a stressor to develop its effect on strain.

These issues are all considered in the present paper. The first two hypotheses were tested using a multi-wave design, which was based on a representative sample of the working population of Dresden in Germany.

Hypotheses 1 and 2 are shown in Fig. 2. In this figure, social stressors, irritation and depressive symptoms are measured at three time points and it is assumed that the variables have some stability, reflected by the arrows from social stressors Time 1 to social stressors Time 2, from social stressors Time 2 to social stressors Time 3, etc. According to hypothesis 1 there are causal paths (arrows) from social stressors to irritation and from irritation to depressive symptoms. According to hypothesis 2, there are no direct causal paths from social stressors to depressive symptoms.

[FIGURE 2 OMITTED]

One shortcoming of many published longitudinal studies is that the potential of longitudinal data to rule out alternative explanations for the hypotheses under study has often not been used (exceptions are, e.g., Dignam & West, 1988; Marcelissen, Winnubst, Buunk, & de Wolff, 1988). The most important alternative explanations refer to reversed causation and third variables. The latter means that an observed relationship is due to the effect a third variable has on the independent and the dependent variable. An example is negative affectivity which has recently been hypothesized to explain self-reported stressor-strain relations (e.g. Burke, Brief, & George, 1993). If third variables, which are held responsible for the spuriousness of a relation, are known and have been measured, the treatment of these variables is straightforward. They can, for example, be partialled out in correlative designs (see, however, Spector, Zapf, Chen, & Frese, 2000). The problem is that researchers do not always have all the relevant information about a causal system, so it can always be argued that a relevant third variable has been left out from the analysis. Therefore, one cannot be sure whether or not a causal interpretation of empirically derived relationships is valid.

There are, however, some methods available when considering the effects of third variables that have not actually been measured (e.g. Dormann, 2001; Dwyer, 1983; Finkel, 1995; Kenny, 1975, 1979). They provide means of protecting one's findings from special kinds of omitted third variable explanations. Basically, such methods rest on assumptions regarding the properties of the unmeasured third variable that should be ruled out as a source of spuriousness. Dormann showed how an extended version of the synchronous common factor (SCF) model, which is the basic model underlying the cross-lagged panel correlation technique (Kenny, 1975), can be used to rule out unmeasured variables as sources of spuriousness when more than two variables, each of which is measured at least twice, are analysed. Compared to the SCF, the model proposed by Dormann is less restrictive and therefore called the `less restrictive synchronous common factor' (LRSCF) model. The LRSCF model is described in some more detail in the Methods section, but it should be noted here that an LRSCF represents a latent variable, (a) which affects the variables of interest (i.e. social stressors, irritation, and depressive symptoms in this instance), (b) which may change over time to some degree, and (c) which is not actually measured but is estimated on the basis of its relations with other variables. According to Dormann, the so-called occasion factors, which are shown in Fig. 3, can be considered as a special case of LRSCF.

[FIGURE 3 OMITTED]

Occasion factors represent third variables that affect other variables in a given situation and which are completely unstable (Dwyer, 1983). The impact of good weather may be an example. Good weather may reduce the level of depressive symptoms, may make individuals feel less irritated, and may even reduce the number of social conflicts because everybody is in a good mood. In Fig. 3, this is shown by causal paths from the occasion factors on social stressors, irritation and depressive symptoms. If weather is assumed to be a completely unstable variable this means that current weather does not allow a prediction about what the weather will be like in the future. In Fig. 3, this is shown by the fact that the occasion factor Time 2 is unrelated to the occasion factor Time 3.

A more general LRSCF model is shown in Fig. 4. In addition to the occasion factor model, this model assumes that the third variable possesses some stability over time shown by the causal …