Life satisfaction refers to an overall positive evaluation of the quality of one’s life (Saris, Veenhoven, Scherpenzeel, & Bunting, 1996) or the subjective enjoyment of life (Veenhoven, 2014). Overall evaluations of one’s life, as well as day-to-day feelings of happiness, both independently contribute to overall life satisfaction (Rojas & Veenhoven, 2013). Satisfaction with life is a broad concept that goes beyond satisfaction with individual life domains, although it may be affected by these domains. Satisfaction with one’s job and wages, for example, may contribute to life satisfaction, but it does not automatically indicate its existence (Saris et al., 1996). Employees’ satisfaction with life seems to be a result of their expectations and the fulfilment of these expectations, which are affected by a socialisation process (Rojas & Veenhoven, 2013).

Job satisfaction is more closely related to satisfaction with one’s job or aspects of that job, and this may vary at different points in time (Diener, Suh, Lucas, & Smith, 1999). Research has established a strong link between job and life satisfaction (Mafini & Dlodlo, 2014). Spector (1997, p. 2) defines job satisfaction as ‘the degree to which an individual likes or is content with his or her job’. Moreover, job satisfaction can be viewed and measured based on an individual’s feelings towards his or her job (Thompson & Phua, 2012), as well as the individual’s thoughts regarding his or her job (Moorman, 1993). Research has shown that job satisfaction contributes to life satisfaction (Dogan & Çelik, 2014). While wage is often seen as a predictor of job satisfaction, its effects may depend on contextual factors (Carr & Mellizo, 2013).

In terms of direct income, the concept of utility is often used to explain the link between income and satisfaction. Utility, in an economic sense, refers to the perceived ability of an item to satisfy an individual’s needs or wants (Oxford University Press, 2015). The neoclassical utility model states that individuals are driven to maximise their utility (Devereux & Engel, 2003), which can be done, in part, by increasing their wage. The assumption is that the higher an individual’s wage, the more he or she can spend on satisfying his or her needs, which would result in increased satisfaction. Research suggests that life satisfaction coincides well with utility (Silva, De Keulenaer, & Johnstone, 2012).

The theory of compensating wage differentials is based on utility theory and represents another attempt to explain the link between wages and satisfaction. According to this theory, individual utility is derived from both wages and non-financial job aspects (Böckerman, Ilmakunnas, & Johansson, 2011). This approach explains the existence of wage differentials where a firm pays a higher wage and/or allowance to compensate for less favourable non-financial job aspects.

Not all researchers agree with the wages and life satisfaction link. Easterlin found that increasing individuals’ income did not necessarily increase their life satisfaction (Clark, Frijters, & Shields, 2008; Easterlin & Angelescu, 2009). Several studies have since contradicted Easterlin’s paradox (Sacks, Stevenson, & Wolfers, 2010, 2012; Stevenson & Wolfers, 2013). Stevenson and Wolfers (2008) found that the effect of wage on life satisfaction was stronger for absolute income and that income comparisons played a more limited role. Comparison income still appeared to play a role, however. Paul and Guilbert (2013) found that peer group income had a significant negative effect on life satisfaction.

Research indicates that wage satisfaction involves both satisfaction with wage amount and the wage process (Brown & Huber, 1992). The social sciences often view wages in terms of a comparison among individuals. This implies that an individual’s satisfaction with his or her wages is based on his or her perception of how well or poorly these compare to the income of others (Clark et al., 2009). Research has shown that reference-group pay is as important as own pay (Ferrer-i-Carbonell, 2005) and that life satisfaction is gained partly from how well one’s income compares to that of a reference group (Welsch & Kühling, 2014). Miceli and Lane (1990) found that wage satisfaction was associated with the number of positive or negative feelings that individuals had towards their wages; this has been proven to significantly contribute to job enthusiasm (Lee & Lin, 2014).

The link between wages and job satisfaction is well researched, with wages generally being a predictor of job satisfaction (Carr & Mellizo, 2013; Schweitzer et al., 2013). Given the interrelatedness of job and life satisfaction, it is not surprising that research has also found life satisfaction affecting gross wage, and vice versa (Dogan & Çelik, 2014). Boodoo, Gomez and Gunderson (2014) found that comparison wage, in particular, played a role in improving the life satisfaction of working individuals, while Haushofer and Fehr (2014) found that long-lasting inadequate wages could lead to a state of anxiety. Utility theory explains that higher wages may lead to higher overall satisfaction (Devereux & Engel, 2003), a sentiment that is mirrored by the income inequality theory (Hagerty & Veenhoven, 2003). Another study showed that the reverse was also true, in that higher life satisfaction was correlated with higher income later in life (De Neve & Oswald, 2012).

Studies have shown that employed individuals (given that they receive a wage) experience higher life satisfaction than unemployed individuals, even those employed in ‘bad’ jobs, that is, jobs with low job quality (Arampatzi, Burger, & Veenhoven, 2015; Grün, Hauser, & Rhein, 2010; Luhmann, Weiss, Hosoya, & Eid, 2014). Given the relationship between wages and job satisfaction (Duffy, Autin, & Bott, 2015) and the prevalence of wage practices that provide allowances for uncomfortable or dangerous work, high wage satisfaction might compensate for lower job satisfaction in affecting life satisfaction.

This study aimed to investigate the relationships among wage category (derived from a gross wage), wage satisfaction, job satisfaction and life satisfaction and to determine whether gross wage category and wage satisfaction compensate for the effect of low job satisfaction on life satisfaction.

This study utilised the South African WageIndicator (WI) survey data set and focused on individuals who were actively employed. As the data set contains data on a large spectrum of variables related to respondents’ work experiences, it was necessary to determine which variables were best suited to answering the research questions. The following variables were identified: job satisfaction, life satisfaction, wage satisfaction and gross wage. The data set was then cleaned to ensure that missing and indolent responses within the four variables under study were removed. Data cleaning resulted in a final sample of 763. The sample was not limited to a single organisation or geographic region and represented different genders, marital statuses, education levels, income levels, ages and racial groups.

Regarding wage distribution in the population (N = 2178), a total of 557 individuals (25.57% of the population) were in the low wage category (R18 to R7519 per month), 1069 (49.08% of the population) were in the medium category (R7521 to R21 949 per month) and 552 individuals (25.34% of the population) were in the high category (R22 000 to R749 000 per month). In the cleaned sample (N = 763), a total of 239 individuals (31.3% of the sample) were in the low wage category (R18 to R7519 per month), 338 (44.3% of the sample) were in the medium category (R7521 to R21 949 per month) and 186 individuals (24.4% of the sample) were in the high category (R22 000 to R749 000 per month).

Table 1 shows that females comprised the majority of the sample (61.7%). The highest percentage of subjects had a Grade 12 qualification (32%), followed by those with a diploma (26.1%). The percentages regarding education and gender of the sample were close to the percentages in the population.

The WI data set contains both cross-sectional and longitudinal data regarding working conditions and wages across 65 countries. The WI survey has been active in South Africa since 2005, with the 2013 data set being the most recent available at the time of access. Access to the data set can only be obtained through a formal application process made to the International Data Service Centre (IDSC), and access is restricted to universities and research institutions.

Data collection for the WI data set takes place via an online multilingual survey hosted on dedicated sites in each of the countries covered by the survey. These dedicated sites provide users with a salary comparison tool and job-related information tailored to the specific country that the site covers. Visitors to these sites are invited to complete the WI survey. An incentive is offered for completion. The WI survey measures life satisfaction using a Likert scale, ranging from 1 (dissatisfaction) to 10 (satisfaction). Researchers have found that single-item life satisfaction measures exhibit similar criterion validity to multiple-item measures and do not produce systematically different correlations (Cheung & Lucas, 2014). Furthermore, Lucas and Donnellan (2012) indicate that single-item measures may be more reliable than previously thought. This provides support for the use of single-item measures to measure life satisfaction.

Job satisfaction is measured with a direct evaluation (‘How satisfied are you with your job?’). The survey utilises one item to evaluate an individual’s gross wage (‘What is your gross income?’) and how this is structured. The survey contains one item directly related to wage satisfaction (‘How satisfied are you with your pay?’). Both wage satisfaction and job satisfaction are measured using a five-point Likert rating scale, ranging from 1 (highly dissatisfied) to 5 (highly satisfied). Dolbier, Webster, McCalister, Mallon and Steinhardt (2005) found strong evidence for the reliability and validity of a single-item measure of job satisfaction (and by implication wage satisfaction, which is an element of job satisfaction).

To protect the confidentiality and privacy of respondents, no names, addresses or other direct identifiers were requested from respondents (Tijdens, Van Zijl, Hughie-Williams, Van Klaveren, & Steinmetz, 2010). The WI data set is specifically designed to collect detailed information concerning people’s earnings, benefits, working conditions, employment contracts and training. It is not tailored to one individual study. The purpose is to provide a large high-quality data set to universities and research institutions.

Access to the WI data set was obtained through a formal application process to the IDSC. On approval, the data set was downloaded in the form of an SPSS input file and stored locally. The latest available data set (2013) was selected and cleaned to remove missing or incomplete values. This was done using IBM SPSS 23 (IBM Corporation, 2015). Next, the data was filtered to include only respondents who were actively employed and working; the resulting sample included 8254 respondents (original sample and/or population size).

The data set was reviewed to determine the variables that had to be included in the study to answer the research question: Do wage category and wage satisfaction compensate for a dissatisfying job? The reasoning was to examine whether participants who received a high gross wage and were satisfied with their jobs would experience low job satisfaction and high life satisfaction; this would be most suited to answering the primary research question. This reasoning was supported by research indicating that two-way interactional relationships existed among wages (and wage satisfaction), job satisfaction and life satisfaction, with life satisfaction being a broader overarching concept (Boodoo et al., 2014). The following variables were used: gross wage category, wage satisfaction, job satisfaction and life satisfaction.

For the wage satisfaction, job satisfaction and life satisfaction variables, the decision was made to focus on contrast groups (high and low). Average scores were removed from the analyses to obtain more accurate estimates of the relations between variables when scores on the four variables were high and low. Groups were selected that scored either high or low (in terms of the upper and lower percentile) on wage satisfaction, job satisfaction or life satisfaction. These contrast groups were studied in relation to gross wage category. Four new variables were created in IBM SPSS, which divided the selected variables into three categories, namely, low (1), medium (2) or high (3). The medium group was removed for the satisfaction variables, leaving the following categories: high, medium and low gross wage category; high and low wage satisfaction; high and low job satisfaction; and high and low life satisfaction.

This allowed for the comparison of contrast groups among the different variables, for example, comparing gross wage with job satisfaction. The rationale for using contrast groups is that three categories of gross wages (high, medium, low) were retained, as these would allow the exploration of interactions between satisfaction and medium wage if the need were to arise. Gross wage was the only variable that contained continuous data, as individuals could enter any amount rather than select from a scale.

IBM SPSS 23 (IBM Corporation, 2015) was utilised to clean the initial data set. Statistical analysis was then carried out using SPSS 23 (IBM Corporation, 2015). Firstly, frequencies were calculated to describe the sample. During this process, IBM SPSS removed missing and incomplete values, leaving a total of 2178 responses across three categories. Removal of the medium (2) category for the satisfaction variables resulted in a total sample of 763 (final sample).

Cross-tabulation was used to produce a contingency table of the data. Therefore, the contingency table contained the number of cases that fell into each of the categories. This table was used to test whether the assumptions required for hierarchical log-linear analysis had been met, namely, no counts less than one and not more than 20% less than five (Field, 2013). As these conditions had been met, the study moved on to log-linear analysis.

Hierarchical log-linear analysis was used to construct a model that predicted the data. Log-linear analysis allows for the expression of categorical data in the form of a linear model, provided that logarithmic values are used (Field, 2013). The analysis constructs a model containing all possible interactions among the variables under study. This model fits the data; however, it is usually very complex. Using a process of backward elimination, log-linear analysis removes interactions until the removal of an interaction significantly affects the model fit (Field, 2013). In this study, removal of interactions followed a hierarchical progression, moving from highest-order interaction (three-way) to lower-order interactions (two-way). The purpose of this process was to end with the simplest model that best fit the data. Cross-tabulations were used to interpret significant effects. The frequencies across different categories were plotted and graphically displayed.

The full model was proposed because there was no prior reason to remove any associations. Screening and model building were used to eliminate associations that did not contribute to observed cell frequencies (Field, 2013). Table 2 contains the information needed to start the model-building procedure: the simultaneous tests for the effects of each order, each order and higher, and the test of individual association. The likelihood ratio and Pearson criteria were used to evaluate the K-way and higher-order effects. The K-way and higher-order effects are reported in Table 2.

According to Tabachnick and Fidell (2014), it is necessary to test all four-way effects combined and three- and four-way effects combined because combined effects take precedence over individual effects. If both combined tests are non-significant, the three- and four-way associations are deleted, regardless of their partial test. If the combined test is significant and the three- or four-way effects are significant, the three- or four-way effects are retained in the model. Table 3 shows that the model fit could be improved by removing the four-way and three-way interactions. Thus, the model did not need to contain associations greater than two-way.

Table 3 provides the basis of a search for the best model of one- and two-way effects.

Table 3 shows that various associations were statistically significant (p < 0.01): job satisfaction by wage satisfaction (partial χ² = 85.612, p < 0.001), job satisfaction by life satisfaction (partial χ² = 54.871, p < 0.001), wage satisfaction by life satisfaction (partial χ² = 52.564, p < 0.001) and wage by wage satisfaction (partial χ² = 24.523, p < 0.001). Two of the two-way associations were non-significant: wage by job satisfaction (partial χ² = 5.853, p > 0.01) and wage by life satisfaction (partial χ² = 3.096, p > 0.001). All first-order effects needed to be included in the final hierarchical model because they were highly significant or part of a significant two-way interaction. Although one three-way effect, job satisfaction by wage satisfaction by life satisfaction, was statistically significant (partial χ² = 8.084, p < 0.01), the three-way associations were not considered for inclusion because the simultaneous tests took precedence over the component associations (Tabachnick & Fidell, 2014).

Stepwise selection by simple deletion of the model with six two-way terms was conducted with IBM SPSS HILOGLINEAR. The selection process stopped after the second step because the criterion probability (0.01) was reached. Each potential model generates a set of expected frequencies. The goal of the model selection is to find the smallest number of effects that still provides a fit between expected and observed frequencies. Criteria for the optimal model are that (1) it must have a non-significant likelihood ratio chi-square value and (2) the selected model should not be significantly worse than the next more complicated model (Tabachnick & Fidell, 2014).

The first model (Step 0) included six effects. The model was not significant, meaning that it provided an acceptable fit between expected and observed frequencies: χ² = 10.934, df = 9, p = 0.280. In Step 1, effects were deleted one at a time. Wage by job satisfaction (p = 0.054) and wage by life satisfaction (p = 0.213) were deleted in this step because these effects produced the smallest chi-square change from Step 0. This model was also statistically non-significant (χ² = 14.030, df = 11, p = 0.231). The second criterion for model selection is that the model should not be significantly different from the next more complicated model. Deletion of the two effects (wage by job satisfaction and wage by life satisfaction) did not result in a significant difference between the models: χ²(1) = (14.030 – 10.934) = 3.096, ∆df = 3, p = 0.213. Therefore, the model in Step 1 was retained because it was not significantly worse than the next more complicated model. The model of choice for explaining the observed frequencies included all first-order effects and the two-way associations between job satisfaction and wage satisfaction, job satisfaction and life satisfaction, wage satisfaction and life satisfaction, and wage and wage satisfaction.

The likelihood ratio χ² (14.030, p = 0.231) indicated a good fit between observed and expected frequencies. Confidence limits for χ² were found by using an IBM SPSS syntax developed by Smithson (2003). Confidence limits (90%) were 0–16.861. The upper level was slightly higher than the critical value of 14.030 (df = 18, α = 0.05). Assessment of fit of the model in individual cells proceeded through inspection of the standardised residuals for each cell. The cell counts and residuals are reported in Table 4.

Table 4 shows the observed and expected frequencies for each cell and the standardised residual values from which discrepancies were evaluated. Most of the standardised residual values were quite small. None of the cells had values that exceeded the critical value of 1.96. Therefore, the model fit in individual cells was acceptable.

Figure 1 shows a normalised probability plot of residuals.

Figure 1 shows that the observed standardised residuals were close to those that were expected (the diagonal line). The closer the observed frequencies are to the expected frequencies, the better the specified model fits the data.

Two types of information were useful in interpreting the model, namely, parameter estimates for the model and marginal observed frequency tables for all included effects. The log-linear parameter estimate, the z-value (coefficient divided by standard error) and confidence intervals for estimates are reported in Table 5. Because life satisfaction, job satisfaction and wage satisfaction had only two levels of each variable, each effect was summarised by a single parameter value where one level of the effect had the positive value of the parameter and the other the negative value of the parameter. Effects with the largest standardised parameter estimates (z) held the most importance in influencing frequency in a cell (Field, 2013; Tabachnick & Fidell, 2014).

The strongest predictor of cell size was wage satisfaction (z = 11.343). The least predictive of all the effects in the model was the interaction between wage satisfaction and wage category. Parameter estimates indicated the relative strength of the effects.

Table 6 summarises significance tests and their confidence intervals (see Smithson, 2003). Note that the expected value of chi-square when the null hypothesis is true is equal to the degrees of freedom.

While Table 2 gave us an indication that one- and two-way interactions significantly affected model fit, partial associations could show which interactions affected the model. Firstly, from Table 6, it was clear that removing the variables being studied would significantly affect the data. Secondly, the table showed that job satisfaction and wage satisfaction, job satisfaction and life satisfaction, and wage satisfaction and life satisfaction significantly affected the model. The relation between wage satisfaction and wages was the only direct income relationship that affected the data. From the c² values, it was also clear that the satisfaction interactions had a much greater effect on the model than the wage interaction.

Table 7 shows cross-tabulations among the significant associations in the model.

As far as the association between job satisfaction and wage satisfaction was concerned, most of the participants who reported job dissatisfaction (95%) also reported dissatisfaction with their wages. A slight majority of those who were satisfied with their jobs (54.3%) were dissatisfied with their wages. If wage satisfaction was high, only 45.7% of the participants showed high job satisfaction. Therefore, although job satisfaction and wage satisfaction were strongly related at a low level of wage satisfaction, fewer people were satisfied with their jobs at a high (compared to a low) wage satisfaction level.

Regarding the association between job satisfaction and life satisfaction, most participants who reported low job satisfaction (76%) also reported dissatisfaction with their lives. Although most of the participants (66.6%) who reported job satisfaction also reported satisfaction with their lives, 33.4% of those who were satisfied with their jobs were not satisfied with their lives. Therefore, although job and life satisfaction were strongly related at a low job satisfaction level, relatively fewer people were satisfied with their lives at a high job satisfaction level.

Concerning the association between wage satisfaction and life satisfaction, 65.5% of the participants who were dissatisfied with their wages were dissatisfied with their lives. At a high level of wage satisfaction, 81.9% of the participants were satisfied with their lives. Therefore, wage dissatisfaction was associated with dissatisfaction with life, but wage satisfaction was even more strongly associated with life satisfaction at a high level of wage satisfaction.

With regard to the relation between wage category and wage satisfaction, most of the participants in the low gross wage category (86.19%) reported low wage satisfaction. Only 13.8% of the participants in the low gross wage category reported high wage satisfaction. Furthermore, in the high wage category, 47.3% of the participants reported high wage satisfaction. However, it should be noted that 52.7% of the participants in the high wage category were dissatisfied with their wage.

The results of this study showed that gross wage was associated with dissatisfaction with wage at a low wage level. However, at a high wage level, this association was much weaker. Job dissatisfaction was associated with dissatisfaction with life at a low level of job satisfaction. However, at a high level of job satisfaction, the association between job and life satisfaction was substantially lower. Wage satisfaction was strongly associated with life satisfaction. Wage dissatisfaction was strongly associated with job dissatisfaction when the wage satisfaction level was low. However, at a high wage satisfaction level, job satisfaction was less strongly associated with wage satisfaction.

High wage and high wage satisfaction did not compensate for lower job satisfaction in affecting life satisfaction. The results showed that of the people who experienced low job satisfaction, 19 (of the expected 26 cases) experienced high life satisfaction when gross wage was high. A total of 31 people (compared to an expected number of 19 cases) who experienced low job satisfaction experienced high life satisfaction when gross wage was low. The hierarchical log-linear analysis also confirmed that interaction did not exist among gross wage, wage satisfaction, job satisfaction and life satisfaction.

The results of this study showed that wage satisfaction, job satisfaction and life satisfaction were related. Therefore, if organisations want to reap the benefits of a satisfied workforce, they must focus on all three of these areas. This study, furthermore, showed that a wage that satisfied employees might affect employees’ life satisfaction. The findings of this study have implications for managers in organisations. Firstly, managers should attend to the perceptions of wage dissatisfaction at low wage levels. Such dissatisfaction may have a negative impact on the job and life satisfaction of employees. Secondly, managers should realise that employees in the high wage category are not necessarily satisfied with their wages.

The study utilised convenience sampling methods, and as such its results might not be generalisable to the general population. It did not examine how biographical factors might affect the specified model; future studies could focus on factors such as age, gender, education or geographic region that might affect the relationships among gross wage, job satisfaction and life satisfaction. Furthermore, this study did not consider job type or situational factors; it might be useful to examine whether the position an employee holds in a company would affect the relationship between gross wage and satisfaction. As with most studies examining wages and life satisfaction, this study could only comment on the correlations among variables. Future studies should make use of longitudinal designs to assess whether wage and wage satisfaction interact with job satisfaction to affect life satisfaction.