Job satisfaction as an indicator of well-being at work

Conventional economists take self-reported data on job satisfaction as a proxy variable for the workers’ utility without further questioning. However, the interpretation of self-reported job satisfaction is a particularly perilous endeavor: it is never possible to know exactly what the respondents are saying. Job satisfaction is often defined as the way people feel about their job and whether they like it; it would hence have a primarily affective character. But it is also recognized that job satisfaction has a cognitive dimension and is therefore also an evaluative judgment about one or several aspects of the job. This makes job satisfaction a multidimensional construct difficult to interpret. Moreover, job satisfaction is shown to depend on personality traits (Reference Judge, Thoresen and BonoJudge et al., 2001) and its perception is culturally biased (Reference Sousa-Poza and Sousa-PozaSousa-Poza and Sousa-Poza, 2000).

Empirically, this intricacy results in the existence of substantive divergences between objective working conditions and self-reported job satisfaction. Reference Bustillo, Fernández-Macías and EsteveBustillo et al. (2011) clearly show that there is hardly any congruence between actual working conditions and job satisfaction across countries. The range of variation of the latter is extremely small even when comparing very different countries in terms of their economic development. In fact, the range of variation in job satisfaction is higher within than across countries, but it is still not clear whether it captures real differences in job quality.

Social scientists generally explain these discrepancies by the fact that workers adapt to their actual work experience. Job satisfaction is assessed by workers in part in relation to what they expect from the job. While some workers might be led to expect a lot, others might be resigned to expecting little. So everything depends on the norms against which preferences and judgments are formed. It is difficult for both the analyst and the respondent to distinguish between true satisfaction and resignation—adaptation of preferences and aspirations to the opportunities actually available. For instance, low-wage workers sometimes express higher satisfaction than high-wage workers: low earners may report being “satisfied” in their jobs simply because they have a low benchmark level of norms and aspirations (Reference Brown, Charlwood and FordeBrown et al., 2007).

Despite these important limitations, information on job satisfaction remains valuable. On one hand, it is robustly proven to relate to actual behaviors such as absenteeism and turnover (Reference SpectorSpector, 1986; Reference UtmanUtman, 1997). On another, job satisfaction transmits useful information about how workers perceive their work life, about how they feel about and evaluate their job, even though it is not an accurate indicator of the value of a job—the extent in which a job objectively provides the conditions for human flourishing. In this article, like Reference GreenGreen (2006) and Reference Green and TsitsianisGreen and Tsitsianis (2005), we consider that even if the levels of job satisfaction do not convey adequate measures of workers’ true well-being, changes in job satisfaction over time constitute reliable indications of changes in well-being at work. The analysis below therefore focuses on trends in job satisfaction and on its relationship with other variables rather than on the comparison of job satisfaction between individuals and across countries. Also, the use of successive cross-sectional samples helps in moderating the influence of individual traits and expectations, and the period of time covered—1995–2010—has not witnessed societal changes able to provoke major shifts in collective expectations toward work.

In the United Kingdom (but not in Germany), the decline in job satisfaction is found to be associated with the decrease in work autonomy ^{Footnote 1} and with the increase in work intensity (Reference GreenGreen, 2004, Reference Green2006; Reference Green and TsitsianisGreen and Tsitsianis, 2005). Indeed, job satisfaction is shown to increase in line with the degree of autonomy enjoyed in the job (Reference Nguyen, Taylor and BradleyNguyen et al., 2003), a result which holds in different countries and for a wide array of job autonomy indicators (Eurofound, 2007). Conversely, work pressure is found to be associated with negative health outcomes (Reference Danna and GriffinDanna and Griffin, 1999; Reference DejoursDejours, 2009), which predicts a strong negative association with job satisfaction. Reference Karasek and TheorellKarasek and Theorell (1990) made a decisive contribution to the understanding of both phenomena by showing that the positive effect of work autonomy on job satisfaction may be partially or totally outweighed by the negative effect of work intensity.

Data, empirical strategy, and the evolution of job satisfaction

The levels and trends in work autonomy, work intensity, and job satisfaction are studied using the 1995, 2000, 2005, and 2010 waves of the EWCS (Eurofound, 1995, 2000, 2005, 2010b), a cross-sectional dataset that provides unique and detailed information on work in Europe. The EWCS is a questionnaire-based survey, administered using face-to-face interviews with approximately 1000 individuals in their homes in each EU country. Every wave sample is representative of those aged 15 years and over who are in employment. In 2010, a multistage, stratified random sampling design was used in each country. Previous waves used a multistage, “random walk” procedure. ^{Footnote 2}

The data analysis follows various steps. We first analyze the evolution of job satisfaction from 1995 to 2010 in the 15 EU countries for which data is available for the four waves (see Table 6 in Appendix 1 for a description of the variables). In the second step, we build indicators of work autonomy and work intensity and examine their evolution by skill level and country (section “The evolution of work autonomy and work intensity in the EU”). The association between work autonomy, work intensity, and job satisfaction is studied in the third step by using regression models (section “Examining the relation between work autonomy, work pressure, and job satisfaction”).

As the sole observation of the aggregate trends at the national level might conceal significant divergences between workers of different skill levels, we distinguish between groups of workers by using the Eurofound classification of the occupational classes of the workers’ jobs in four categories: ^{Footnote 3} high-skill clerical (HSC), low-skill clerical (LSC), high-skill manual (HSM), and low-skill manual (LSM) (see Table 6).

The analysis of average job satisfaction by skill level reveals that, in each wave and overall, high-skill workers display higher levels of satisfaction than low-skill workers, and clerical workers are on average more satisfied than manual workers (Table 1). As for trends, average satisfaction in 1995 was higher than in any other year for all skill levels. There is a marked decline in the satisfaction of clerical workers over the years, whereas the satisfaction of manual workers seems to have stabilized in the 2000s.

Table 1. Average job satisfaction by year and skill level.

Job satisfaction goes from 1 = not at all satisfied to 4 = very satisfied.

Although declining, there is not a very pronounced trend in average job satisfaction. In contrast, the percentage of very satisfied workers decreased significantly. Differences between 1995 and 2010 are significant (p < 0.001) for all skill levels (Table 2). It appears that the steadiness of the average level of job satisfaction results from two phenomena: the decline in the number of very satisfied workers, which lowers the average satisfaction, and the decline in the not at all satisfied workers, which has the opposite effect.

Table 2. Proportion of very satisfied workers by year and skill level.

* Significant difference at the 0.001 level.

The evolution of work autonomy and work intensity in the EU

If the findings mentioned in the introduction hold, that is, if work autonomy is actually related to job performance and job satisfaction, one would expect to see work autonomy steadily increasing for all workers over time and space. However, some studies show that there is no performance advantage to autonomous motivation for low-skill jobs (Reference Gagné and DeciGagné and Deci, 2005). While work autonomy is shown to promote performance in jobs requiring high levels of commitment, control devices have been found to yield superior short-term performance in unskilled tasks. Managers seeking efficiency would therefore discriminate between workers and a polarization process would be observed, marked by a long-term decline in job autonomy for the workers in less-skilled jobs.

With regard to work intensity, the competitive pressure due to the progressive catching-up of emerging economies might lead European firms to intensify work in an effort to maintain higher levels of competitiveness. However, work intensification is not an unlimited process and there is evidence that European workers are already subject to high work strain. As predicted by Reference Karasek and TheorellKarasek and Theorell (1990), evidence shows that the combination of high work intensity and low autonomy is associated to high risks of cardiovascular disease, depression, insomnia, and musculoskeletal disorders (Reference De Lange, Taris and KompierDe Lange et al., 2003; Reference SiegristSiegrist, 2006; Reference Van der Doef and MaesVan der Doef and Maes, 2010). An increase in work intensity is hence to be expected for all skill levels, but it might be less pronounced for workers with already high work intensity and/or low work autonomy.

Although all countries face similar competitive demands and economic constraints, national institutional settings and cultural specificities may strongly impact the levels and trends of work-related variables. In effect, the different welfare regimes exhibit substantially different levels of job quality (Reference Davoine, Ehrel and Guergoat-LariviereDavoine et al., 2008; Reference Esser and OlsenEsser and Olsen, 2012; Reference GallieGallie, 2003). Because of their more egalitarian regimes, higher trade union membership, and labor-oriented policies, we expect the Scandinavian countries to display better levels of intrinsic job quality and less difference among groups of workers. Conversely, South European countries are expected to fare worse in terms of both work autonomy levels and polarization trends, while Continental and Anglo-Saxon countries would display average positions.

As for the measurement of work autonomy, we rely on the indicators of Reference Lopes, Lagoa and CalapezLopes et al. (2014), which are based on seven questions of the EWCS that reasonably cover what Reference KarasekKarasek (1979) calls job control. We conducted the same principal components analyses (PCA) and obtained the same (standardized) scores, which we use in the analysis below. Our PCA also systematically revealed that work autonomy is a bi-dimensional phenomenon. ^{Footnote 4} We nonetheless prefer not to keep the terms used by Lopes et al. (2013) and name instead the first factor procedural autonomy (PA) and the second factor content autonomy (CA).

Despite some divergences, these two factors broadly correspond to the two distinct constructs that, for Reference KarasekKarasek (1979), constituted job control, namely, decision authority and skill discretion. In effect, PA refers to the degree of control that workers perceive that they exert on when and how they carry out their tasks and CA refers to the learning opportunities available in their job and to whether workers assess the quality of their work (see Table 6). In other words, PA shows the control workers have over their methods and schedules, while CA shows the control workers have over their work content and the use of their skills. In most studies, the two constructs are combined in a single index of work control in spite of evidence that the constructs are empirically distinct (Reference Mansell and BroughMansell and Brough, 2005). Like Karasek and Theorell, we prefer not to subsume the two dimensions into a single additive scale. In fact, results show that the behavior of both measures differs across the four skill levels in a meaningful way.

Overall, work autonomy declined significantly over the period (see last rows of Table 3). Results confirm the Scandinavian exception: work autonomy levels in Denmark, Finland, the Netherlands, and Sweden are above the EU average for all groups of workers, and there is much less difference between clerical and manual workers and between low-skill and high-skill workers than in all other countries, where manual workers systematically suffer from below average work autonomy (Reference Lopes, Lagoa and CalapezLopes et al., 2014). In addition, whereas work autonomy decreased in all other countries from 1995 to 2010, it stabilized or increased in Denmark, Finland, the Netherlands, and Sweden. As for polarization processes, the situation of manual workers in the 15 studied countries has not deteriorated when compared to clerical workers. By contrast, a clear polarization process is underway between HSC and LSC workers. Hence, our expectations regarding polarization trends and welfare regimes are confirmed, but, contrary to expectations, autonomy at work for high-skill workers slightly decreased rather than increased (for complete analysis, see Reference Lopes, Lagoa and CalapezLopes et al., 2014).

Table 3. Work pressure and work autonomy scores over time, all countries pooled.

PA: procedural autonomy; CA: content autonomy.

The 0.00 score corresponds to the average level of work pressure or work autonomy of all workers for all waves; a negative score means below average work pressure/work autonomy, while a positive score means above average work pressure/autonomy.

* Significant difference at the 0.001 level.

We now concentrate on examining the levels and trends in work intensity and analyzing the combined evolution of work intensity and work autonomy by skill level and country. Measuring work intensity—or job demands, in Karasek’s terms—requires information on work load or work effort. However, no such data are available in the four EWCS waves. To construct our indicator, we used the two questions traditionally examined in studies of work intensity based on EWCS (job involving working at very high speed and working to tight deadlines) to which we added questions related to potentially stressful patterns of work such as reliance on work done by colleagues, defined performance targets, and time constraint. The set of possible responses to the time constraint question differs across waves: the 1995 and 2000 waves required a yes or no answer, while the 2005 and 2010 waves proposed an ordered set of categories (see Table 7 in the Appendix 1). The ordinal nature of some variables and the dichotomous nature of others led us to use Categorical Principal Components Analysis (CatPCA). CatPCA also presents the advantage of incorporating a sophisticated option that only takes into account the nonmissing data when the loss function is minimized (Reference Meulman, van der Kooij, Heiser and KaplanMeulman et al., 2004), and it allows the score computation for all objects with at least one valid response. Thus, the time constraint question could be included in the analysis. CatPCA reveals that the five items load on a single factor (eigenvalue = 2.43). Cronbach’s alpha for the index is 0.71, indicating sufficient internal consistency.

The resulting indicator, named work pressure rather than work intensity for the sake of rigor, fully exploits the information available in the EWCS and contains reliable information on an important dimension of job demands. The (standardized) score on this factor is used hereafter as the Work Pressure (WP) indicator (dataset pooled).

Table 3 shows that, with minor exceptions, perceived work pressure has been steadily increasing over the years for all skill levels. Differences for the period are significant at the 0.001 level for all workers. The increase in work pressure has been higher for HSC workers—who suffer from the highest levels of work pressure at the end of the period while enjoying relatively low levels at the beginning—and lower for LSM workers. These findings suggest that it is HSC workers who have seen their situation deteriorate the most, but definite conclusions can only be drawn from the analysis of the combined evolution of work autonomy and work pressure.

The analysis reveals that, overall, work pressure increased and work autonomy decreased over the period. More specifically, Figure 1 shows that job strain clearly increased for all workers over the period: work pressure rose for all skill levels without being compensated for by any increase in work autonomy. Work autonomy decreased more intensely and work pressure increased the most between 2000 and 2005, since then, both phenomena seem to have stabilized, but no reversion of the overall negative trends can be observed. Work pressure for HSC workers increased markedly but they still benefit from high, though declining, work autonomy levels, ^{Footnote 5} especially when compared to LSC workers. But according to Reference Karasek and TheorellKarasek and Theorell (1990)’s model, it is HSM workers who suffer from the highest job strain since they experience the larger excess of work pressure over decision latitude. The large discrepancy between work autonomy and work pressure for manual workers may partly explain their lower levels of job satisfaction.

Figure 1. Work autonomy (PA and CA) and work pressure by skill level and year for all countries.

PA: procedural autonomy; CA: content autonomy.

Turning now to the analysis of countries, we can see in Figure 2 that Finland and Sweden have the highest perceived work pressure over the period, followed by the Netherlands and Denmark—the four countries where work autonomy is also the highest. But we can also note that work pressure was already very high in Finland and Sweden in 1995. While closer to the Continental countries for work autonomy, the United Kingdom stands closer to the Scandinavian countries regarding work pressure, which indicates high levels of job strain in this country.

Figure 2. Work autonomy (PA and CA) and work pressure by country and year.

PA: procedural autonomy; CA: content autonomy.

Another interesting finding is that Continental countries (Belgium, Austria, Germany, France, and Luxemburg) and Ireland ^{Footnote 6} display quite high work pressure at the end of the period—in fact, these are the countries where work pressure increased the most—and yet are still characterized by rather low work autonomy. Finally, South European countries—with the exception of Greece for work pressure—display low levels of both work pressure and work autonomy. Italy stands between Continental and South European countries.

As for trends, the evolution for workers in Continental countries and Greece is the most negative: work pressure rose significantly without being compensated for by increases in work autonomy. Workers in Scandinavian countries have seen their situation slightly worsen or stabilize. Analysis by skill level and country (results not shown, available on request) shows that whereas most clerical workers benefited from lower work pressure than manual workers in Continental countries over the period, workers of all skill levels in Scandinavian countries have always experienced similar—high—levels of work pressure. The conclusions concerning polarization trends and welfare regimes are the same for work pressure and work autonomy: overall, a polarization trend is observed between HSC and LSC workers with a deeper worsening of the latter group’s situation, and Scandinavian countries offer markedly more egalitarian and higher-quality jobs.

Examining the relation between work autonomy, work pressure, and job satisfaction

To analyze the extent to which the evolution of work autonomy and work pressure affects job satisfaction, we estimate econometric models using worker-level data and controlling for several explanatory variables of job satisfaction.

Since job satisfaction is measured through an ordered multilevel scale, we use an ordered logit regression model. Our main interest is in the work autonomy and work pressure variables, measured using the scores obtained previously. We also introduce an interaction term between work autonomy and work pressure to capture the effect of the relation between both variables on job satisfaction. As mentioned, our hypothesis is that an increase in work pressure is less harmful to the worker’s satisfaction if he or she enjoys high autonomy. This means that the coefficients of the interaction terms should be positive.

We introduce sequentially our explanatory variables, starting with the control variables. Next, we test the autonomy variables, followed by work pressure. The quadratic terms of both variables are then introduced. Finally, the interaction terms between autonomy and work pressure are considered. The quadratic terms are introduced prior to the interaction terms to rule out the possibility that the significance of the interaction is the spurious result of a curvilinear impact of work autonomy or work pressure on job satisfaction (Reference Mansell and BroughMansell and Brough, 2005). One advantage of this sequential approach is that it allows understanding what occurs to the unexplained time trend as additional variables are introduced.

Our first estimated model includes only control variables (Model 1, Table 4)—that is, the variables available in the four EWCS waves that theory and previous evidence show to influence job satisfaction: socio-demographic features, skill level, hours of work, dummies for years, countries, fixed-term contract, and economic sector (description of variables in Table 6). Results confirm those of the descriptive analysis reported earlier: low-skill workers are less satisfied than high-skill workers, and manual workers experience less satisfaction than clerical workers. The coefficients for the other control variables are in line with previous studies on the determinants of job satisfaction. The time dummies indicate a significant unexplained decline in job satisfaction between 1995 and 2010.

Table 4. Regressions for job satisfaction.

WP: work pressure; PA: procedural autonomy; CA: content autonomy; LSC: low-skill clerical; HSM: high-skill manual; LSM: low-skill manual.

The base categories are as follows: high-skill clerical for skill level, services for economic sector, and United Kingdom for country. Standard deviations in brackets. Other control variables also included: dummies for economic sector and country dummies.

* Significant at 10% level of significance

** significant at 5%

*** significant at 1%

In the second model, the introduction of the autonomy variables (Model 2, Table 4) reveals that work autonomy has a significant and positive effect on job satisfaction, a result consistent with the literature. Even though the time dummies continue to be statistically significant, the reduction in their size shows that work autonomy contributes to explain the observed decline in job satisfaction.

Next, we introduced work pressure, which, as expected, proved to be negatively and significantly related to job satisfaction (Model 3, Table 4). Any one unit increase in work pressure (the value of the standard deviation) decreases the odds of a higher level of job satisfaction compared with a lower level by a factor of 0.40. Likewise, PA and CA increase the odds ratio by a factor of 0.25 and 0.22, respectively. It can also be observed that the 2005 and 2010 time dummies become insignificant after introducing work pressure, which means that the increase in work pressure is highly significant in explaining the decline in job satisfaction.

Quadratic terms for CA, PA, and WP were then introduced because, besides the reason pointed out above, autonomy and work pressure may have nonlinear effects on job satisfaction (Reference KarasekKarasek, 1979). Indeed, the detrimental effect of work pressure on job satisfaction may increase as work pressure increases: an increase in work pressure would have a greater negative impact on job satisfaction for a worker with an already high work pressure. Results confirm the curvilinear effects of autonomy and pressure on job satisfaction (Model 4, Table 4). For example, assuming autonomy constant, if work pressure increases one unit for a worker with low work pressure (work pressure = −1), the odds of a higher level of job satisfaction compared to a lower level decreases by a factor of 0.34, while the same increase in work pressure in a worker with high work pressure (work pressure = 1) has an impact of −0.45.

Interestingly, the impact of both autonomy indicators increases with the level of autonomy, as can be seen by the positive sign of the squared terms of autonomy. In other words, both work pressure and work autonomy have an increasing marginal impact on job satisfaction, the former being negative and the latter positive. Workers with already large autonomy are those who most benefit from an increase in autonomy. Other aspects constant, an increase in PA for a worker with high PA (= −1) has a two times larger effect on job satisfaction than for a worker with low PA (= −1): the odds ratios increase 0.38 and 0.19, respectively. A possible explanation for this result is that there may be a learning process on how to cope with autonomy, a process yielding satisfaction.

Finally, we introduced the interaction terms between work pressure and work autonomy (Model 5, Table 4). The interaction terms are positive, confirming our expectation: an increase in work pressure has a smaller negative impact on job satisfaction when the workers’ autonomy is high. This indicates that the moderating influence of autonomy on the effect of work pressure on job satisfaction—the buffer effect—is significant. Furthermore, a simultaneous increase of one unit in PA, CA, and WP increases the odds ratio of job satisfaction by a factor of 0.11. ^{Footnote 7} This confirms Reference KarasekKarasek (1979)’s prediction that high work pressure and high work autonomy may be associated with higher job satisfaction than low work pressure and low autonomy.

We also estimated the model for each country separately (results not shown, available on request). For all countries, the coefficients of the work autonomy variables are significant and positive, while the work pressure coefficients are negative and of larger amplitude. It is in the countries where work pressure increased the most and where work autonomy is around average (Austria, Belgium, Denmark, France, Germany, and Luxemburg) that the negative effect of work pressure on job satisfaction is higher (above EU average). Conversely, for low levels of work autonomy and work pressure (Spain and Portugal), the negative effect of work pressure is below average, as is the positive effect of work autonomy. In other words, increases in work autonomy are more valued by workers in Continental and Scandinavian countries, which confirm the results obtained with the aggregate data at the European level.

Attending to the limitations of job satisfaction as an indicator of well-being at work (see section “Job satisfaction as an indicator of well-being at work”), we wanted to know whether our results hold when job satisfaction is replaced by work-related health problems. We constructed a variable with the number of diseases likely to be generated by job strain that the employee says have been caused by his job: anxiety, heart diseases, overall fatigue, sleep difficulties, and some muscular pains. This variable was used as dependent variable in the estimation of an equation similar to Model 5 (in Table 4). Both the correlation and the regression analyses show that more PA is associated to a smaller number of reported diseases, while more CA and work pressure contribute to an increase in the number of diseases (results available on request). The interaction terms between each of the autonomy indicators and work pressure are not statistically significant, but they have a negative sign. Therefore, when job satisfaction is replaced by the number of work-related diseases, we find weaker but still positive evidence supporting the Karasek–Theorell’s model, which is consistent with results reported in meta-analyses (Reference De Lange, Taris and KompierDe Lange et al., 2003; Reference Van der Doef and MaesVan der Doef and Maes, 2010).

Finally, to assess the extent to which the evolution of work pressure and work autonomy accounts for the average decline in satisfaction, we followed Reference GreenGreen (2006) and Reference Bartolini, Bilancini and PugnoBartolini et al. (2013). First, to simplify calculations and the interpretation of results, we estimated the job satisfaction model using ordinary least squares (OLS). Then, we computed the average change between 1995 and 2010 in each regressor. Finally, we multiplied each variable’s coefficient obtained in the first step by the respective average change in the variable, but using only the coefficients significant at the 10% level. We thereby obtained the contribution of each regressor to the explanation of the average change in job satisfaction. The sum of the contributions of each explanatory variable gives us the overall predicted change in job satisfaction, which can be compared with the observed variation in job satisfaction (−0.047).

The application of this method shows that the increase in work pressure is the main factor explaining the decline in job satisfaction, accounting for around 67% of the decline (Table 5). The decrease in CA was also an important factor, explaining around 23% of the reduction in job satisfaction. PA accounted for only 6% of the decline in satisfaction. As suggested in the literature (Reference ClarkClark, 2005; Reference RoseRose, 2003), the deterioration of the quality of intrinsic aspects of work outweighed the beneficial effects on job satisfaction of other features of work.

Table 5. Explaining the average change in job satisfaction.

PA: procedural autonomy; CA: content autonomy.