The Nature of NA as a Bias
Research has established individual dierences in the tendency to experience negative emotions. The most frequently studied of these dispositions is trait anxiety, although anger and depression have also been given attention. Watson and colleagues (e.g., Watson and Clark, 1984;Watson and Pennebaker, 1989), noting that various measures of aective disorders were strongly intercorrelated, developed the negative aectivity construct, positing that certain individuals were predisposed to experience distress and negative emotions. Thus NA is a broader construct than trait anxiety or other aective dispositions. In the organizational domain researchers have tended to adopt the NA concept when studying aective dispositions, although measures of individual pervasive emotions (e.g., trait anxiety) are often used interchangeably with broader measures that assess a variety of negative emotions. Watson et al. (1987) suggest that individuals who are high in NA tend to report high levels of distress and negative emotions, even in the absence of objective stressors. They suggested that this tendency produced a bias in the measurement of job stressors and job strains, as well as other organizational variables. Although Watson et al. (1987) focused mainly on a biasing eect, in other places Watson and colleagues have discussed the possibility of substantive eects, including the likelihood that NA itself can be aected by job strains (e.g., Watson and Slack, 1993). This distinction between bias and substantive is vital in determining the appropriate way in which personality variables in general and aective dispositions such as NA in particular are studied.
To be considered a bias, a variable such a NA must distort the assessment of a particular intended construct (Spector and Brannick, 1995). It cannot be causally interlinked with the true underlying construct as either cause or eect. For example, a response set (e.g., agreement) is the tendency to respond to items in a particular way, independent of the variable intended to be measured. When biases are common across measures, and have an in¯uence of the same direction on independent and dependent variables, they in¯ate the correlations. In other cases, biases may attenuate correlations (Cote and Buckley, 1988;Williams and Brown, 1994).
Note that bias must be interpreted in view of the intended construct. If a measure of a job stressor is intended to assess the objective work environment, any factor that is independent of that environment, yet in¯uences the measure, will serve as a bias. Hence from this perspective measures of job stressors and other work conditions are all biased because they are aected by many personal factors (Spector, 1992). Given the same objective environment, each individual likely perceives it somewhat dierently. However, if such measures are considered to re¯ect individual perceptions of the environment these personal factors are real causes of the underlying construct of interestÐperceptions. After all, it is by no means established that the objective environment can be successfully measured if we partial presumed biases, such as faking, NA, response sets, or social desirability, from self-report measures.
We note four issues that have relevance for the interpretation of biases in assessment of job stressors and job strains.
(1) NA might help explain why two variables are related without being a bias, i.e., it has a substantive role. For example, suppose individuals who are high in NA tend to perform poorly, resulting in punitive supervisor response. Furthermore, the constant negative behavior of the supervisor might produce job strains in the subordinate. We then might observe that NA correlates with incumbent reports of their supervisor's punitive behavior, as well as with job strains, but in this case NA's role is substantive. It helps explain why we have observed the relations between the incumbent measure of supervisor behavior and job strains. On the other hand, if supervisor behavior is unrelated to their subordinates' personality, a correlation between