Research carried out over three decades studied the relationship between SWB and job performance. For example, in a series of pioneer studies, Wright and his colleagues examined the relationship between the affective component of SWB and job performance (Wright & Bonnet, 2007; Wright et al., 1993; Wright & Cropanzano, 2000; Wright et al., 2002; Wright et al., 2007; Wright & Staw, 1999). These researchers found correlations ranging from .23 to .48. Some studies showed even higher correlations (e.g., Ahmed & Malik, 2019; Haider et al., 2018; Merriman, 2016; Odle-Dusseau, 2008; Rego & Cunha, 2008; Riaz et al., 2014). Nevertheless, some studies showed very low correlations (e.g., Boyar et al., 2016; Clarke & Mahadi, 2017; Kossek et al., 2001; Law et al., 2008); some showed no relationship (Athota et al., 2020; Donaldson & Blanchard, 1995; Sargent & Terry, 1998; Staw & Barsade, 1993; Walker, 2013; Youssef & Luthans, 2007), and other studies found a negative correlation between SWB and job performance (e.g., Moradi et al., 2014; Singh et al., 2012). Overall, the findings revealed considerable variability.

Three meta-analyses summarized the relationship of the cognitive and affective components of SWB with job performance (Erdogan et al., 2012; Ford et al., 2011; Shockley et al., 2012). These meta-analyses consisted of a small number of primary studies conducted in the USA, many of them using small-size samples (Walsh et al., 2018). This last fact means that second-order sampling error can happen as in Ford et al. (2011), where the variance explained by artifacts is greater than 100% in the case of LS and self-rated performance relationship. Shockley et al.’s (2012) meta-analysis reported credibility intervals that included zero for three out of four cases. This implies that the correlation found is not generalizable beyond the studies included in the meta-analysis. Finally, the meta-analysis of Erdogan et al. (2012) did not correct the observed correlations for the effect of artifactual errors (e.g., reliability in LS and job performance and sampling error), which produced an underestimation of the true relationship and the overestimation of true variance.

The literature reviews of Lyubomirsky and her colleagues (Boehm & Lyubomirsky, 2008; Lyubomirsky et al., 2005; Walsh et al., 2018) mentioned two limitations of past meta-analyses: (a) many of the primary studies included in those meta-analyses date back many years and many used small-sample sizes and (b) they can suffer to some extent from publication bias (i.e., it is possible that a number of nonsignificant, unpublished studies were not included and only studies conducted in the USA were included).

Together with those limitations, there are two additional ones that can be mentioned: (a) previous meta-analyses did not examine the relationship between job performance and the cognitive and affective components of SWB simultaneously, and (b) apart from Ford et al. (2011), previous meta-analyses did not distinguish between self-rated job performance and supervisory-rated job performance. This last distinction is relevant, as supervisory ratings of job performance are the main criteria in personnel decisions (Campbell & Wiernik, 2015; Salgado & Moscoso, 1996, 2019a, 2019b; Salgado et al., 2016).

Research on the relationship between SWB and job performance has also been characterized by using a variety of estimates of SWB. Some studies have evaluated a global estimate of SWB, other studies have evaluated affective SWB or cognitive SWB only, and other studies have evaluated both SWB components. For instance, the series of studies by Wright and his colleagues (e.g., Wright & Cropanzano, 2000; Wright et al., 2007) used a measure of affective SWB, Law et al. (2008) evaluated cognitive SWB only, Zelenski et al. (2008) evaluated cognitive SWB and affective SWB but not overall SWB, and Salgado et al. (2019) evaluated cognitive SWB, affective SWB, and overall SWB.

The variability in SWB-job performance correlations together with the fact that the studies assessed different SWB facets and components suggest that the relationship between overall SWB, its components, and job performance might be different from the relationships of job performance and cognitive and affective SWB.

Almost all the studies in our database (with few exceptions) used one of these two types of ratings: (a) supervisory ratings of job performance or (b) self-ratings of job performance. Self-ratings were used in 69% of studies on cognitive SWB-job performance, in 33% of studies on affective SWB-job performance, and in 53.8% of the studies on overall SWB-job performance. In other research areas, for example, cognitive abilities and personality, studies using self-ratings of job performance typically found greater validity in comparison with the studies using supervisory ratings (Salgado & Moscoso, 2000). However, this moderator effect had not been meta-analytically examined in the SWB-job performance domain.

In addition, interrater reliability of supervisory ratings is typically lower than .70, while self-ratings’ reliability is typically higher than .80 (Salgado & Moscoso, 1996, 2019a; Viswesvaran et al., 1996). As the reliability of the measures is a factor that attenuates the correlation between SWB and job performance, one can expect that the magnitude of the correlation be larger for the studies in which job performance was assessed with self-reports.

Also, it must be pointed out that the correlation between self-ratings of job performance and SWB measures might also be larger because of common-method variance, particularly when cross-sectional studies are carried out. For these reasons, an examination of the potential moderator effects of the rating type seems to be in order. Consequently, we state the following research question.

Research Question 1: Does the type of job performance ratings (supervisory ratings vs. self-ratings) moderate the validity of SWB and its components as predictors of job performance?

Most of the primary studies conducted so far, and meta-analyses mentioned above examined the relationship of one of SWB components with job performance, but the joint relationship of both SWB components with job performance has been scarcely researched in primary studies, and it was not meta-analytically tested. Some recent primary research used a global measure of SWB that included both cognitive and affective measures (Jalali & Heidari, 2016; Jones, 2006; Merriman, 2016; Rabenu et al., 2017; Zheng et al., 2015). However, they did not test the specific contribution of each SWB component in the prediction of job performance.

Recently, Salgado et al. (2019) examined this issue, both concurrently and predictively. More specifically, Salgado et al. tested the predictive efficiency of cognitive SWB, affective SWB, a global measure of SWB, and a composite of cognitive plus affective SWB measures. They found that the global measure of SWB showed similar validity to the two components considered separately. Interestingly, a noteworthy finding was that the composite of cognitive and affective SWB showed a much larger validity than the other three alternative measures (i.e., cognitive SWB, affective SWB, and global SWB). However, the most significant finding was that the cognitive and the affective components of SWB showed a suppressor relationship with job performance. In other words, the magnitude of the effect of the affective component increased when the cognitive component entered in the regression equation. This suppressor effect of the two SWB components in their relationship with job performance had not previously been found in the research literature.

Salgado et al. (2019) explained this effect theoretically suggesting that the cognitive component of SWB (i.e., satisfaction with life) would function as an emotion regulation mechanism. It would operate “suppressing” (avoiding or reducing) negative emotions, which, subsequently, would produce greater frequency of positive emotions, and, therefore, affective SWB would have a greater effect on job performance. Salgado et al. also suggested that because cognitive SWB is typically more stable over time than affective SWB, the cognitive component of SWB would also have the effect of reinforcing the stability of affective SWB, which would subsequently make the effect of affective SWB on job performance stronger (because of the smaller variability and the smaller measurement error) over time.

Suppressor effects are not unknown in the SWB literature and several suppressor effects related to well-being have been described previously. For instance, Watson et al. (2013) found a suppressor effect between euphoria and well-being for predicting psychological-health problems (e.g., depression, generalized anxiety disorder, posttraumatic stress disorder, and panic). In another study, Paulhus et al. (2004) found that two emotions, guilt and shame, maintained a suppressing relationship for predicting depression.

To the best of our knowledge, this suppressor effect of cognitive SWB and affective SWB with job performance has not been re-examined or replicated yet. Moreover, it must be taken into account that the study of Salgado et al. (2019) used supervisory ratings to evaluate job performance. Therefore, it remains unexamined whether the suppressor effect also exists when job performance is self-rated. A comprehensive meta-analysis that calculates the specific correlation (effect size) of each SWB component with job performance would allow for the testing of the hypothesis of the suppressor relationship of cognitive and affective SWB with job performance. Based on the previous rationale, we posit the following two research questions:

Research Question 2: What is the magnitude of the joint relationship of cognitive SWB and affective SWB with supervisory job performance and self-reported job performance?

Research Question 3: Do cognitive SWB and affective SWB show a suppressor relationship for predicting supervisory job performance and self-rating job performance?

In summary, despite the fact that research showed that SWB consists of a cognitive component and an affective component, some empirical studies evaluated one of the components, but not both, and other studies used an overall measure of SWB that did not permit the estimation of the specific contribution of each component to the prediction of job performance (Diener et al., 2003; Fisher, 2010). Another issue unexamined in previous research is whether affective SWB shows incremental validity over cognitive SWB for predicting job performance.

This study has two main goals: (1) to examine the moderator effects of the performance measure (self-report vs. supervisory ratings) and (2) to test the hypothesis of the suppressor effect of cognitive SWB on affective SWB to predict job performance. In order to achieve these goals, we conducted a series of psychometric meta-analyses.

Using three strategies, we searched for studies that reported correlations or data which permit to calculate the correlation between a measure of SWB (i.e., overall SWB, PA, NA, EB, LS, and the like) and a measure of job performance. The first strategy was to conduct electronic searches using the following databases and meta-databases: PsycLit, Google, Scholar-Google, ERIC, Elsevier, Sage, Wiley, Academy of Management, Springer, and EBSCO. We used the following keywords: “subjective well-being,” “psychological well-being,” “satisfaction with life,” “happiness,” “positive affect,” “negative affect,” “emotional balance,” “SPANE,” in combination with “job performance,” “performance ratings,” and “work performance.” The second strategy was to examine the section of references of the meta-analyses and narrative reviews mentioned above, and the references of collected documents to identify potential papers not included in the previous set. The third strategy was to contact international researchers to obtain previously unidentified papers. The final database consisted of 34 independent samples using supervisory performance ratings and 38 independent samples using self-reported job performance. The list of studies and their data appear in the Appendix.

The Bioethics Committee of our university declared the meta-analyses carried out with published and unpublished studies exempt from approval because they do not include personal identification data.

As the primary purpose of this meta-analysis was to determine the correlation between SWB and its components with real job performance, in the final database we included only documents reporting correlational studies with real incumbents. In other words, we did not consider laboratory experiments, studies with no real people, and studies with student samples. We also excluded studies reporting validity estimates for physical symptoms and clinical scales. When studies reported a range of numbers of incumbents, we coded the smallest number to provide a more conservative estimate. When an article or document reported data from two or more independent samples of participants, they were entered into the meta-analysis as separate correlations. When a study reported correlations for the same sample obtained in different occasions, the average estimate served as the data source for that sample. Finally, when a study used two performance measures for the same sample at the same time, the average correlation was entered as data source. We also excluded 11 studies because the same coefficients had been reported in another paper included in the dataset.

The inclusion criteria fulfill the criteria and recommendations of the Meta-Analysis Reporting Standards (MARS) specified in the Publication Manual of the American Psychological Association (2019; available at https://apastyle.apa.org/manual/related/JARS-MARS.pdf) and the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA).

The authors coded all correlation coefficients independently and the following categories of data points were compared: (1) sample size, (2) correlation coefficient, (3) performance measure (self-ratings vs. supervisor ratings), (3) performance appraisal purpose (i.e., administrative vs. research), (4) SWB measure, (5) SWB reliability, and (6) performance reliability. The initial agreement was 95.3%. The authors examined the disagreements by referring back to the studies and discussing until they reached consensus.

We used three methods to detect potential publication bias: (1) comparison of the average observed correlation of published and unpublished studies, (2) correlation between sample size and effect size, and (3) cumulative meta-analyses (CMA). The total of published and unpublished studies, on average, showed practically the same effect size. Therefore, we can reject the idea that the publication source distorts average validity. The correlation between sample size and effect size was small and statistically non-significant; therefore, we can also reject this source of publication bias. Concerning to CMA, there is agreement that CMA is the most robust technique for detecting publication bias (Borenstein, 2005; Kepes et al., 2012; Schmidt & Hunter, 2015; Schmidt & Le, 2014). CMA results showed no evidence of publication bias. Therefore, the three techniques agreed that publication bias was not a relevant issue in the current case.

In this meta-analysis, we used the Sample Adjusted Meta-Analytic Deviance Statistic (SAMD) by Huffcutt and Arthur (1995; see also Beal et al., 2002) and the number of SD units below or above the mean of the distribution of correlations (Wilcox, 2014) to identify potential outliers. The two outlier criteria agreed as they identified two studies, with four validity coefficients, as potential outliers. They were the study of Moradi et al. (2014) and the study of Singh et al. (2012). The corresponding SAMD statistics were remarkably larger than 1.96 (p < .001). All SD units below the mean were larger than 3 SD units (i.e., p < .001). Therefore, the studies of Moradi et al. (2014) and Singh et al. (2012) were considered outliers in this study, and meta-analyses were done with and without these two studies. The effects of these two outliers were substantial on the observed variance, adding artifactual variance.

SWB reliability and its cognitive and affective components were estimated using the coefficients (internal consistency) reported in individual studies. Some of the studies did not provide reliability coefficients of the instruments, and, in those cases, we used the mean value of the reliability distribution of all combined studies as the study reliability. Next, we developed an empirical distribution for overall SWB and each SWB component. Mean reliabilities were .83 (SD = .08) for cognitive SWB, .80 (SD = .08) for affective SWB, and .85 (SD = .10) for overall SWB. The meta-analysis of Busseri (2018) reported reliabilities of .86 (SD = .07), .81 (SD = .07), and .82 (SD = .04) for positive affect, negative affect, and satisfaction with life, respectively, which are very similar to reliabilities found in this meta-analysis.

In the studies included in the dataset, job performance was assessed with self-reports and/or supervisory ratings. Therefore, two different estimates of reliability are required. In the case of self-reports, an internal consistency coefficient (e.g., Cronbach’s alpha) is the estimate of choice in the absence of a coefficient of equivalence and stability (CES) (Schmidt et al., 2003). In the case of supervisor ratings, in the absence of a CES, an interrater coefficient is the appropriate one (Salgado & Moscoso, 1996, 2019a; Viswesvaran et al. 1996). As some studies did not provide job performance’s reliability, we developed an empirical distribution of internal consistency coefficients for self-reports of job performance. In the case of supervisor ratings, we developed an empirical distribution considering the nature of supervisor ratings. Salgado and Moscoso (2019a) found that the purpose of performance appraisal (research vs. administrative) is a powerful moderator of interrater reliability. As only one study in our dataset reported interrater reliability, we used the distributions developed by Salgado and Moscoso (2019a) according to the following rules: (1) if supervisor ratings were collected for research purposes, we used .61 (SD = .11) as an estimate of interrater reliability; (2) if supervisor ratings were collected for administrative purposes, we used an estimate of .48 as interrater reliability. If the job performance measure was a self-report, and the study did not provide the coefficient, we used the mean value of the distribution as reliability (.80). Average reliability was .80 (SD = .09) for self-reported job performance and .61 (SD = .11) for supervisory job performance ratings.

These meta-analyses were conducted to respond to the Research Question 1, concerning potential moderator effects of source of job performance ratings on the relationship between SWB and job performance. These meta-analyses allow us to compare true score correlation for studies which used self-reports of job performance and with the true score correlation obtained for studies which used supervisory ratings. Tables 1 and 2 show the results of these meta-analyses.

The results show that the source of job performance ratings operates differently for the various SWB components and measures. More specifically, true correlation is noticeably larger for self-reported job performance than for supervisory ratings in the case of cognitive SWB (ρ_t = .30 vs. ρ_t = .17), but the true score correlation is considerably larger for supervisory ratings in affective SWB (ρ_t = .49 vs. ρ_t = .30), slightly larger in overall SWB (ρ_t = .42 vs. ρ_t = .40), and in SWB estimate when all studies were combined (ρ_t = .35 vs. ρ_t = .33). On average, true score correlation is slightly larger (ρ_t = .35 vs. ρ_t = .33) for the supervisory ratings of job performance.

Considering all the results, Research Question 1 does not have a single answer. The results indicate that the source of job performance ratings is a relevant moderator of the relationship between SWB and job performance, but it operates differently for cognitive SWB and affective SWB. For cognitive SWB, the relationship is larger when job performance is assessed with self-reports, while for affective SWB and overall SWB, the relationship is larger when supervisors are the source of job performance ratings.

As far as observed variability was concerned, artifactual errors explained more variance for studies using supervisory ratings than for studies using job performance self-reports. Ninety percent CVs were all positive and very different from zero, which indicates that relationships between SWB and job performance generalize for both self-reports and supervisory ratings.

In order to establish the joint capacity of cognitive and affective SWB to predict job performance and to know whether cognitive SWB shows a suppressor effect on affective SWB, we conducted a series of hierarchical multiple regression analyses using the validity estimates found in previous meta-analyses. Multiple regressions were conducted in two steps. In the first step, affective SWB was the predictor, as it showed a larger correlation with job performance. In the second step, both affective and cognitive SWB were entered into the equation. Multiple regression requires knowing the correlation between cognitive SWB and affective SWB, but the studies included in the database rarely reported this information. For this reason, we used the values found in the meta-analyses of Busseri (2018) as the best estimate of the correlation between cognitive SWB and affective SWB. Finally, we created three matrices of correlations reported in Table 3, using cognitive SWB and affective SWB and three correlation estimates of job performance.

We carried out three hierarchical multiple regression analyses. Table 4 reports the results of these analyses. The first one used the correlations of cognitive and affective SWB with job performance, without distinguishing if job performance was assessed by supervisory ratings or self-reports. In this case, R was .374, and cognitive SWB added practically no explained variance over affective SWB (13.7% vs. 14.0%), but beta weights were significant for the two SWB components. In this case, the compound of cognitive and affective SWB showed a multiple correlation similar to the correlation found for overall SWB in the prediction of both self and supervisory ratings of job performance, as it was reported in Tables 1 and 2.

Self-reported job performance was the dependent variable, and cognitive SWB and affective SWB were the independent variables in the second multiple regression analysis. In this case, cognitive SWB showed incremental validity over affective SWB. Multiple correlation R was .334, which is higher than the two bivariate correlations (.30 and .30, respectively). The increment of explained variance was 2.2%, and beta values were significant. Betas were similar for cognitive and affective SWB, although slightly higher for affective SWB (.202 vs. .177). Therefore, the two components of SWB added validity for predicting self-reported job performance. However, the magnitude of the multiple correlation of the compound is smaller than the bivariate correlation of overall SWB with self-reported job performance (.334 vs. .40, respectively).

In the third multiple regression analysis, the dependent variable was supervisory performance rating, and cognitive and affective SWB were the independent variables. In this case, R is larger than the true correlation of affective SWB and supervisory ratings of job performance (.516 vs. .49). Therefore, the addition of cognitive SWB to the equation increased predictive validity, and also increased explained variance by 2.6%. However, beta for affective SWB increased by 26% (.615 vs. .49), while beta for cognitive SWB was negative (-.205); nonetheless, its zero-order correlation was positive (ρ = .17). These two beta values indicate that cognitive SWB acted as a suppressor variable for predicting supervisory ratings of job performance. Suppressor effect suggests that the presence of this variable in a regression equation increases the predictive power of an independent variable on the dependent variable.

More specifically, we found a net or cross-over suppression. Salgado et al. (2019) also found this cross-over suppression effect of cognitive SWB in their longitudinal study. The net or cross-over suppression refers to cases in which two independent variables and a dependent variable correlate positively with each other, but the inclusion of the two independent variables in regression equations increases beta of the most influential variable and changes beta sign of the weakest variable; that is, positive zero-correlation becomes a negative beta (Cohen & Cohen, 1975; MacKinnon et al., 2002; Paulhus et al., 2004; Salgado et al., 2019; Watson et al., 2013).

In order to test if the increment of affective SWB beta is significant, we computed Sobel test, z test, and 95% confidence interval (Sobel, 1982; MacKinnon, 2008; MacKinnon, Lockwood, et al., 2007). For the Sobel test, we used a calculator developed by Preacher and Leonardelli, which is available online at http://www.quantpsy.org/sobel/sobel.htm. We also computed 95% confidence interval for the suppressor effect using the distribution of the product of two regression coefficients (z test). In this case, we used the PRODCLIN program developed by MacKinnon, Fritz, et al. (2007). Sobel test was -9.97 (p < .001), z test was -.12 (p < .01), and upper and lower limits of the 95% confidence interval were -.146 and -.105. Therefore, the three criteria showed that the suppressor effect was significant.

The finding that cognitive SWB suppresses some affective SWB variance in the case of supervisory ratings of job performance agrees with previous finding of Salgado et al. (2019). Also, it supports Diener’s suggestion that both components of SWB should be assessed and measured independently. This suppressor relationship has both theoretical and practical implications that we will discuss later.

In summary, findings also revealed that the source of job performance ratings (i.e., self-reported vs. supervisory ratings) moderates the relationship of SWB and its components with job performance. Also, we found that there was a cross-over suppression relationship between affective and cognitive SWB in the prediction of supervisory ratings of job performance.