1 Introduction

The subjective well-being literature has grown exponentially since the mid-1970s (Diener 2000; Frey and Stutzer 2002; Dolan et al. 2008; Stutzer and Frey 2010; Frey and Gallus 2016). Emanating from this literature has been a growing interest in the study of domain satisfactions (Møller and Saris 2001; van Praag and Ferrer-i-Carbonell 2004; Easterlin 2006; Rojas 2006; Diener and Ryan 2009) such as job satisfaction (Clark 1997; van Praag and Ferrer-i-Carbonell 2004), financial satisfaction (Joo and Grable 2004; Plagnol 2011), and leisure satisfaction (Tsou and Liu 2001). While these domains have been considered in some detail, research by Easterlin (2006) in the US has demonstrated that satisfaction with family life could be an important domain to study, for example since people are happier (Easterlin 2006) and report higher life satisfaction (Moss and Willoughby 2016) if they are more satisfied with their family life. Moss and Willoughby (2016) have also found that greater family-life satisfaction is positively related to domains such as financial-, community-, and job satisfaction.

The family life domain has not received much attention in the literature compared to some other satisfaction domains. This is despite the fact that family forms a key unit of society (Agate et al. 2009; Zabriskie and Ward 2013), is a large part of people’s lives, and influences individual psychological and social development (DeFrain and Asay 2007; Dolan et al. 2008; Alesina and Giuliano 2010, 2013; Conger et al. 2010). Although some studies (Easterlin 2006; Agate et al. 2009; Yamamura 2014; Moss and Willoughby 2016) have investigated satisfaction with family life, it was not the main focus in those studies and hence special attention is required on what the predictors of family-life satisfaction are.

This paper is the first to address the determinants of satisfaction with family life within the South African population. The paper thus contributes to the literature on domain satisfactions (van Praag and Ferrer-i-Carbonell 2004) and the satisfaction with family life domain in particular which, as mentioned previously, has received little attention in the broader literature. While more generally exploring the predictors of satisfaction with family life, the primary focus of this paper is on the importance of socioeconomic status (SES) in explaining individual satisfaction with family life. SES, generally measured by education and income, predicts many facets of people’s lives and their development (Conger and Donnellan 2007; Conger et al. 2010). South Africa provides an interesting setting for the study of family-life satisfaction and SES. South African families are diverse across a variety of spectrums, one of which is that of SES. Large SES differences exist, particularly across racial groups, due mainly to the apartheid legacy of racial categorization in terms of social spending, labour market discrimination, and where people were allowed to live (Møller and Saris 2001; Gradín 2012; Leibbrandt et al. 2012). In addition, South Africa has substantial cultural differences across racial lines, which impact on diverse views on family life, differences in household structure, and the broader socioeconomic context of families (Amoateng et al. 2007; Nkosi and Daniels 2007; Botha et al. 2017).

Since SES represents the level of economic strain that families face (Tiffin et al. 2007), SES is a potentially important factor in an individual’s perceived satisfaction with family life. Families can experience economic strain due to a number of reasons other than low income, such as a lack of sufficient assets, which in turn exacerbates family stress (Rothwell and Han 2010; Han and Rothwell 2014). This is especially the case in developing country contexts (Kabudula et al. 2016). This paper therefore takes a broader view of SES by using three constructed composite SES indices (Phongsavan et al. 2006; Georgiades et al. 2008; Sheppard et al. 2009; Kabudula et al. 2016), namely an individual-, household-, and subjective SES index (Botha et al. 2017). Apart from being able to explore objective as well as subjective SES indicators, the use of the various indices makes it possible to determine whether satisfaction with family life is correlated differently with different SES measures.

Another unique part of this paper is the inclusion of perceived family functioning (McCubbin et al. 1996; Botha and Booysen 2014) as a covariate of satisfaction with family life, which also allows for considering whether the role of family functioning in explaining family-life satisfaction differs across SES quartiles. Multidimensional in nature, family functioning reflects the intra-family relational processes by which family members interact and work towards attaining family goals and functions (Morris and Blanton 1998; Patterson 2002; Botha and Booysen 2014). The importance of optimal family functioning for individual well-being is widely established (Tiffin et al. 2007; Walsh 2016) and in South Africa a positive association of family functioning with individual happiness and life satisfaction has been reported (Botha and Booysen 2014). There is also an established interplay between the quality of family functioning and socioeconomic conditions (Conger et al. 2010; Botha et al. 2017). The concept of family functioning therefore adds a distinct dimension to the analysis and understanding of the family satisfaction domain literature that has not been considered in previous work. Theoretically, persons should be more satisfied with their family lives if they reside in families that function well.

In summary, there is a lack of research on satisfaction with family life in general but particularly in South Africa, coupled with the unique diverse composition of South African families in terms of SES, race, cultural beliefs, and overall family contexts. This paper is therefore concerned with the questions of how SES is associated with family life-satisfaction, whether family functioning plays any role in explaining satisfaction with family life, and whether there is any interaction between family functioning and SES in explaining family-life satisfaction in South Africa.

2 Literature

Very little research has been conducted on the predictors of satisfaction with family life, with the existing studies all conducted in developed countries, moreover. There is some evidence of the important role of SES in explaining satisfaction with family life, though this is mainly in terms of narrower as opposed to broader measures of SES.

Easterlin’s (2006) US study showed that satisfaction with family life explained the largest proportion of reported happiness when compared to the domains of financial-, job- and health satisfaction. Specifically, greater satisfaction with family life was associated with greater happiness. Though the main purpose of Easterlin’s study was to relate how various domain satisfactions affect life cycle happiness, an ordered logit model was also estimated with satisfaction with family life as dependent variable. Easterlin (2006) reported an inverted U-shaped relationship between satisfaction with family life and age, with family-life satisfaction first rising and then falling after roughly 50 years of age. Men were also found to be significantly less satisfied with family life relative to women, and those with tertiary education were more satisfied with their family lives than those without a tertiary education. Finally, Black persons reported being less satisfied with family life compared to White respondents.

Agate et al.’s (2009) main goal was to examine the role of family leisure satisfaction in explaining satisfaction with family life among 898 US families. Satisfaction with family life was ascertained by five 7-point Likert scale-type questions, termed the Satisfaction with Family Life Scale. The analyses were conducted on three samples, namely a parent-, youth, and overall family sample. In the parent data, satisfaction with family life was significantly positively associated with income and currently married respondents were more satisfied with family life. Within the youth sample, satisfaction with family life was positively associated with family income, and youth were also more satisfied with family life if their parents were married. Finally, in the overall family sample average satisfaction with family life among family members was significantly positively related to family income.

Using data from the Japanese General Social Survey, Yamamura (2014) primarily investigated differences in sexual behaviour between smokers and non-smokers, and how sexual behaviour of smokers and non-smokers is related to satisfaction with family life, among married and unmarried respondents. The question measuring satisfaction with family life is assessed by asking respondents: “How much satisfaction do you get from your family life?”, with responses recorded on a 5-point scale ranging from dissatisfied to satisfied. Yamamura (2014) found a positive relationship between the frequency of sex and satisfaction with family life, although this relationship as expected was significant for married persons but not unmarried persons. Furthermore, satisfaction with family life was positively related to income whereas the unemployed were less satisfied with family life compared to the employed. There was no significant association between years of education and family-life satisfaction, and married persons were more satisfied with family life relative to unmarried persons.

Moss and Willoughby (2016) employed a large representative sample in the US to examine whether beliefs in the advantageousness of marriage were related to individual life satisfaction and several domain-specific satisfactions. In regressing satisfaction with family life on some selected control variables, Moss and Willoughby (2016) find that men were more satisfied with family life than women were, whereas the frequency of religious attendance and a respondent’s age were both associated with lower satisfaction with family life. Moreover, satisfaction with family life was higher among the more educated and among White respondents, while married persons also reported greater family-life satisfaction compared to the non-married.

In summary, to the best of our knowledge all existing research on the determinants of satisfaction with family life has been conducted within developed economy contexts. There is some evidence of a positive relationship between SES, measured mainly in terms of income and education, and satisfaction with family life. However, no study has explicitly explored the role that SES plays in relation to satisfaction with family life as an outcome, and in particular in a developing country as well. Where SES indicators were used in previous research, these indicators have been limited and narrowly defined. Furthermore, most studies employed individual-level SES indicators to evaluate an individual’s satisfaction with family life, when it is plausible that household-level SES factors may also matter in shaping an individual’s judgement about their family-life satisfaction.

3 Data and Methods

3.1 Data

The 2012 wave of the South African Social Attitudes Survey (SASAS), conducted by the Human Sciences Research Council (HSRC 2012), is used in this paper. SASAS is a nationally representative survey conducted annually since 2003 as a repeated cross-section, and monitors changes in the attitudes and values of South Africans over time. The survey is designed to provide a representative sample of individuals at least 16 years of age within households that are geographically dispersed across South Africa’s provinces. Samples are drawn from the HSRC’s master sample, which consists of 1000 Population Census enumeration areas and is stratified by province and population group. For each interview round, a sub-sample of 500 enumeration areas are then drawn from the master sample. The SASAS round used in this paper had 2547 original respondents, and the data are weighted to ensure that the sample is representative of the broader South African population.

Given the nature of the questions in the family functioning instrument employed in this paper, single-person households are excluded from the analysis since families generally consist of two or more members (Waite 2000; Patterson 2002; Williams et al. 2015). In addition, this paper excludes particular two-person households where such households comprise a single parent with a child younger than 12 years. Research has reported that children younger than 12 do not engage in meaningful bargaining, and the assumption is made that children younger than 12 generally do not make major decisions within the household (Harbaugh et al. 2001; Lundberg et al. 2009; Dauphin et al. 2011). As such, perceptions of family functioning would not apply beyond the one household member. Removing these households resulted in a total sample of 2126 observations.

3.2 Variables

The 2012 SASAS is among the few, and currently most recent, South African household surveys that includes a question on satisfaction with family life. The question measuring satisfaction with family life is on a 7-point Likert-type scale and asks: “All things considered, how satisfied are you with your family life?” Responses include “completely unsatisfied”, “very unsatisfied”, “fairly unsatisfied”, “neither satisfied nor unsatisfied”, “fairly satisfied”, “very satisfied”, and “completely satisfied”.

This paper considers SES in a broader sense (Sheppard et al. 2009; Rothwell and Han 2010; Han and Rothwell 2014), beyond mainly income and education (Conger and Donnellan 2007; Tiffin et al. 2007; Conger et al. 2010; Diemer et al. 2013). Some argue that SES indicators should be included separately to determine each factor’s individual contribution to the specific outcome (Conger and Donnellan 2007; Conger et al. 2010; Diemer et al. 2013). However, in developing countries especially (Kabudula et al. 2016) components such as household assets and living standards can be important contributors to family stress (Rothwell and Han 2010; Han and Rothwell 2014) and family relationships (Botha et al. 2017). As such, composite SES indices are used (Phongsavan et al. 2006; Georgiades et al. 2008; Kabudula et al. 2016) originally developed in Botha et al. (2017), who constructed the SES indices with the purpose of exploring how SES is associated with family functioning in South Africa.

The indices include an individual-, household-, and subjective SES index, with the various SES index components listed in Table 10. Index components were selected based on previous literature (Barbarin and Khomo 1997; Yang and Gustafsson 2004; Fotso and Kuate-Defo 2005; Howe et al. 2008; Sheppard et al. 2009; Reising et al. 2013; Kabudula et al. 2016) and data availability. The individual SES index includes the respondent’s income, education, and employment status. The household SES index includes total household income and household characteristics such as asset ownership (i.e. whether the household owns certain assets such as a washing machine and stove) and infrastructure (i.e. electricity access, toilet facilities, and so on). The subjective SES index contains items that measure a respondent’s perception of the household’s SES relative to other households (for example, the perceived income position of the household compared to that of other households in the neighbourhood). These three indices allow for an examination of whether the classification or nature of SES matters for respondents’ reported satisfaction with family life. Another advantage is that objective and subjective SES components can be explored to determine if objective and subjective SES measures relate differently to family-life satisfaction.

Because all variables in the SES indices are categorical, the SES indices were constructed using multiple correspondence analysis (MCA) (Greenacre 2006; Sourial et al. 2010; Kabudula et al. 2016), a generalization of principal components analysis in the presence of categorical data. The MCA for the individual SES index explains 86.8% of the total inertia in the first dimension, whereas the household SES MCA explains 91.8% of the principal inertia in the first dimension. For the subjective SES index, the MCA explains 81.6% of the principal inertia in the first dimension. Table 11 contains the MCA weights assigned to each SES component. All weights have the expected sign, with positive (negative) weights for items expected to be positively (negatively) related to SES. In order to examine whether the level of SES matters as well as to conduct multiple-group analysis (discussed in Sect. 3.3) across SES groups, the continuous SES indices are also used to construct categorical SES variables with each SES index apportioned into quartiles.Footnote 1

The Family Attachment and Changeability Index 8 (FACI8) (McCubbin et al. 1996) is used as measure of family functioning. FACI8 is a self-report measure with two sub-scales, Attachment and Changeability, each with eight items. The Attachment scale measures the attachment of family members to each other, whereas the Changeability scale measures the flexibility of family members’ relationships with each other. FACI8 has been used in previous South African research (see, for example: Botha and Booysen 2014; Masquillier et al. 2014; Wouters et al. 2014) and also validated with the SASAS 2012 data (Botha et al. 2016). Table 12 contains all FACI8 items, as well as the items’ summary statistics and the proportion of responses across all categories. Cronbach alpha coefficients are acceptable at 0.78 for both FACI8 sub-scales.

The control variables included are age and age squared, gender, race, marital status, household size, religion status, geographical area, female-headed household status, and household structure. Age and age squared are included to allow for potential non-linearity in the association between satisfaction with family life and an individual’s age (Easterlin 2006). Gender consists of “male” (comparison group) and “female” categories, whereas race denotes a person’s racial group and consists of “Black” (comparison group), “Coloured” (the official South African classification for people of mixed-race), “Indian or Asian”, and “White”. Marital status includes “never married” (comparison group), “separated/divorced”, “widowed”, and “married”. Household size reflects the number of persons in the household, whereas religion status equals one if a respondent is religious and zero if not (comparison group). The geographical area denotes whether the household is located in a rural or urban (comparison group) area. A variable is also included to indicate whether the respondent lives in a female-headed household (comparison group) or male-headed household. Household structure includes “skip-generation and multi-generation households” (comparison group), “single-parent households with at least one child”, “a couple without children”, “a couple with at least one child”, and “other households” (including family forms such as mixed families with non-relatives living in the household, and siblings only).

3.3 Data Analysis

The data were analysed using Stata version 14.2 and missing values were removed from the analysis via listwise deletion (Allison 2003; Wouters et al. 2014). Alternatives to listwise deletion would have been methods such as multiple imputation and full information maximum likelihood (FIML). However, because the missing data patterns revealed that each variable had less than 0.02% of observations missing, the highly complex nature of multiple imputation may not be justify a potentially negligible improvement in observations used. In preliminary analyses the models were also estimated via FIML. The results (available on request) were very similar to the listwise deletion findings in terms of sign and significance of the path coefficients. The use of FIML requires an explicit assumption of normally distributed data (Allison 2003), however, but this assumption is not consistent with the SASAS data (see Sect. 4.1).

The structural equation modelling (SEM) model is depicted in Fig. 1. Given the many control variables included, the “controls” box is shown in Fig. 1 to reflect all control variables, as including boxes and paths for each variable would render the figure very cluttered.Footnote 2 Consistent with theory (McCubbin et al. 1996), the FACI8 sub-scales appear with correlated error terms and reflect the measurement model. For the structural part, the relevant SES index is specified as covariate for satisfaction with family life, Attachment, and Changeability. In turn, Attachment and Changeability are specified as determinants of satisfaction with family life. The same controls are included for the family-life satisfaction, Attachment, and Changeability equations.

Fig. 1
figure 1

SEM model specification. Note Controls include: age, gender, marital status, household size, religion status, whether the household is in a rural or urban area, whether the respondent lives in a female-headed household, and household structure

Full size image

All models are estimated via maximum likelihood (ML), which assumes multivariate normality. While the satisfaction with family life variable and FACI8 items can be viewed as ordinal variables and thus not normally distributed, this paper assumes cardinality of the outcomes and uses ML. This assumption is supported by previous research (Johnson and Creech 1983; Babakus et al. 1987; Dolan 1994; Hutchinson and Olmos 1998; Beauducel and Herzberg 2006; Rhemtulla et al. 2012) that argues that the treatment of ordinal variables with five or more categories as continuous and using ML is unlikely to have a serious impact on the results. To guard against violation of any normality assumptions, the Satorra–Bentler (S–B) scaled χ 2 statistic (Satorra and Bentler 1994) is also used and models are estimated with S–B standard errors that adjust for non-normality. Goodness-of-fit indices used to assess model fit are the root mean squared error of approximation (RMSEA), Comparative Fit Index (CFI), and standardized root mean square residual (SRMR). For an acceptable model fit, RMSEA ≤ 0.06, SRMR ≤ 0.08 and CFI ≥ 0.90 (Hu and Bentler 1999; Schreiber et al. 2006; Aarons et al. 2007).

This paper also estimates multiple-group SEM (MGSEM) to examine if the role of family functioning in explaining satisfaction with family life differs across SES quartiles.Footnote 3 In other words, is the relationship between Attachment and family-life satisfaction as well as between Changeability and family-life satisfaction similar across SES groups? Because the FACI8 sub-scales form part of the full SEM models, multiple-group confirmatory factor analysis (MGCFA) is first used to demonstrate measurement invariance among the two latent sub-scales. Separate CFA models are estimated for each SES quartile to check for adequate model fit. Configural invariance (H form) is then tested for by estimating a MGCFA across SES quartiles with no constrained parameters, with support for H form suggesting similar factor structures across SES quartiles. If H form is not rejected, metric invariance (H Λ) is tested for by imposing the constraint of equal loadings across SES quartiles. Evidence of H Λ would suggest that the latent constructs are manifested similarly across SES quartiles. Existence of metric invariance leads to a test for scalar invariance (H Λ,ν), which constrains the loadings and intercepts to be equal across groups. If H Λ,ν holds, mean levels of the latent family functioning constructs are equal across SES quartiles.

For the measurement invariance analysis Bollen’s (1989) χ 2 difference test \((\chi_{D}^{2} )\), or likelihood ratio (LR) test, examines whether a constrained model performs significantly better than a model with fewer or no contraints.Footnote 4 It is well-known that the χ 2 difference test depends on sample size, and hence could indicate lack of measurement invariance even if there is little difference in model fit (Cheung and Rensvold 2002; Meade et al. 2008; Kline 2011; Brown 2015). The ΔMcDonald’s NCI (ΔMc) (McDonald 1989) and ΔCFI statistics are therefore also used as approximate indices of model fit, as these are not affected by sample size and provide a more practical way of examining measurement invariance than the χ 2 difference statistic (Cheung and Rensvold 2002; Kline 2011). If ΔMc ≤ 0.02 and ΔCFI ≤ 0.01, the null hypothesis of invariance is not rejected (Cheung and Rensvold 2002).

Having established measurement invariance of the latent constructs across SES groups, the MGSEM analysis is conducted.Footnote 5 For each SES measure, this entails first estimating a multiple-group model without constraints on any of the unstandardized structural parameters. A second model is then estimated where in the structural model the Attachment coefficients are constrained to be equal across SES groups and the Changeability coefficients are constrained to be equal across SES groups. A χ 2 difference test is then conducted to determine whether the fit of the constrained model, which is nested within the unconstrained model, is significantly worse than the fit of the unconstrained model. If the χ 2 difference statistic is not statistically significant, the constrained model does not do significantly worse than the unconstrained version and thus would support the constraints imposed on the structural coefficients.

4 Results

4.1 Descriptive Analysis

Summary statistics are included in Table 1. In addition, to put the numbers into context, they are compared to the 2012 International Social Survey Program (ISSP) data (ISSP Research Group 2012), as SASAS 2012 formed part of the 2012 ISSP module. Mean satisfaction with family life is 5.47. This seems high within the context of a 7-point scale and may suggest that South Africans are in general quite satisfied with family life. However, compared to the 2012 ISSP where individuals across countries were asked the same 7-point question regarding satisfaction with family life, South Africa fares rather poorly. For instance, the mean satisfaction with family life score across all countries, excluding South Africa, is 5.44. Of the 37 countries in the ISSP, South Africa ranks 24th in terms of reported satisfaction with family life.

Table 1 Summary statistics
Full size table

The distribution of the satisfaction with family life question is presented in Table 2. The distribution is clearly skewed towards the higher ends of the 7-point scale, in particular from “fairly satisfied” and higher: The Shapiro–Francia W’ test rejects the null of normality (z = 10.9, p < 0.001). This distribution is also remarkably similar across SASAS waves and the ISSP 2012 data. Around 83.77% report being at least “fairly satisfied” with family life in the SASAS, compared to roughly 84.85% in the 2012 ISSP.

Table 2 Distribution of satisfaction with family life, compared to ISSP 2012 data
Full size table

Spearman correlation coefficients between reported family-life satisfaction and the SES indices are presented in Table 3. Satisfaction with family life is positively correlated with all SES indices, with the association being strongest with subjective SES (ρ s  = 0.401, p < 0.001) and weakest with individual SES (ρ s  = 0.179, p < 0.001). Unsurprisingly, the SES indices are also positively correlated, with the largest correlation being between household SES and subjective SES (ρ s  = 0.695, p < 0.001). Thus, there is a notable association between a household’s objective level of SES and a respondent’s subjective perception of the household’s SES.

Table 3 Spearman correlations
Full size table

Figure 2 plots average satisfaction with family life according to individual SES quartile. The relationship between satisfaction with family life and the individual SES index is statistically significant (F = 23.0, p < 0.001). There are no significant mean differences between quartiles one and two (p = 0.145) and three and two (p = 0.166), but the differences between all the other individual SES quartiles are statistically significant (all p < 0.001). Mean satisfaction with family life according to household SES quartile is presented in Fig. 3. The relationship is also statistically significant (F = 97.9, p < 0.001) with a strong positive association evident. For example, average satisfaction with family life is 4.76 among persons in household SES quartile one, compared to 6.08 among those in the fourth quartile. Figure 4 shows how average reported levels of family-life satisfaction differ by subjective SES quartile. The relationship between subjective SES and satisfaction with family life is significant (F = 154.1, p < 0.001). Again, persons report higher average levels of satisfaction with family life if they fall in a higher subjective SES quartile. Pairwise mean comparisons also demonstrate significant differences between all household- (all p < 0.001) and subjective SES (all p < 0.001) quartiles. Overall, therefore, mean family-life satisfaction thus clearly increases as the SES quartile rises, especially among the household- and subjective SES indices.

Fig. 2
figure 2

Mean satisfaction with family life and individual SES quartile

Full size image
Fig. 3
figure 3

Mean satisfaction with family life and household SES quartile

Full size image
Fig. 4
figure 4

Mean satisfaction with family life and subjective SES quartile

Full size image

4.2 Measurement Invariance

Table 4 reports the goodness-of-fit statistics for each quartile, according to SES index. Although the fit results for the first subjective SES quartile are relatively poor (S–B χ 2 = 251.7, p < 0.001; RMSEA = 0.055; CFI = 0.888; SRMR = 0.065), fit statistics for all other quartiles of all SES indices are acceptable. As a whole, therefore, there do not seem to be any major issues with the individual analysis of the various SES quartiles when considered in isolation.

Table 4 Goodness-of-fit results for SES group CFA models
Full size table

Table 5 contains the measurement invariance findings with the purpose of demonstrating invariance for the two latent family functioning sub-scales across the three SES indices. Considering the results for the individual SES index, the test for configural invariance (H form) suggests a good overall fit (χ 2 = 894.9, p < 0.001; RMSEA = 0.049; CFI = 0.927; SRMR = 0.054). As such, there is configural invariance for the individual SES index. There is also evidence of metric invariance across individual SES indices: The \(\chi_{D}^{2}\) statistic is not statistically significant (\(\chi_{D}^{2} = 51.3, p = 0.155\)), whereas ΔMc < 0.02 and ΔCFI < 0.01. In testing for scalar invariance across individual SES indices, the results support scalar invariance (\(\chi_{D}^{2}\) = 62.4, p < 0.001; ΔMc < 0.02 and ΔCFI < 0.01).

Table 5 Measurement invariance results
Full size table

There is support for equal form invariance across household SES quartiles (χ 2 = 903.5, p < 0.001; RMSEA = 0.049; CFI = 0.925; SRMR = 0.057). Although the \(\chi_{D}^{2}\) statistic is statistically significant (\(\chi_{D}^{2}\) = 69.6, p < 0.01), ΔMc < 0.02 and ΔCFI < 0.01. As such, the findings suggest the existence of metric invariance across household SES quartiles. There is somewhat mixed evidence regarding the existence of scalar invariance (\(\chi_{D}^{2}\) = 124.8, p < 0.001; ΔMc < 0.02; ΔCFI > 0.01) in the household SES index. There is support for configural invariance (χ 2 = 962.0, p < 0.001; RMSEA = 0.052; CFI = 0.918; SRMR = 0.056) as well as for metric invariance (\(\chi_{D}^{2}\) = 50.4, p = 0.174; ΔMc < 0.02; ΔCFI < 0.01) across subjective SES groups. In addition, scalar invariance cannot be rejected for the subjective SES index (\(\chi_{D}^{2}\) = 145.9, p < 0.001; ΔMc < 0.02; ΔCFI < 0.01).

As a whole, the measurement invariance results indicate that the form of the two latent family functioning sub-scales is similar across quartiles for all three SES indices. Moreover, failing to reject the equal loadings models suggests that the two latent constructs are conceptualized in very similar ways across the quartiles of all the SES indices. As invariance has been demonstrated in the measurement model, we proceed by estimating full structural equation models.

4.3 SEM Results

All estimated SEM models (Tables 6, 7, 8) have an acceptable model fit, with RMSEA ranging between 0.027 and 0.029, CFI between 0.911 and 0.913, and SRMR between 0.026 and 0.027. The SEM results from Table 6 reveal that persons ranking higher on the individual SES index are generally more satisfied with their family life. Persons in quartiles three and four are significantly more satisfied with family life relative to those in the first quartile. Also, people in quartile four are significantly more satisfied with family life relative to those in quartile two (χ 2 = 16.0, p < 0.001) and quartile three (χ 2 = 22.8, p < 0.001). Post-estimation tests show a significant difference in the Attachment and Changeability coefficients when explaining satisfaction with family life (χ 2 = 22.8, p < 0.001), whereas there is no significant distinction between how the individual SES index is related to Attachment and Changeability (χ 2 = 1.7, p = 0.192).

Table 6 SEM results—individual SES
Full size table
Table 7 SEM results—household SES
Full size table
Table 8 SEM results—subjective SES
Full size table

From the results reported in Table 7, persons in quartile one are significantly less satisfied with family life when compared to people in all other household SES quartiles. Moreover, individuals in the second household SES quartile are less satisfied with family life compared to people in quartile three (χ 2 = 6.2, p < 0.05) and quartile four (χ 2 = 16.7, p < 0.001), while those in quartile four are more satisfied than those in quartile three (χ 2 = 4.1, p < 0.05). In addition, there is a statistically significant difference in the relationship of Attachment and Changeability with family-life satisfaction (χ 2 = 19.2, p < 0.001). In contrast to the results for individual SES, the difference in the household SES coefficients across the Attachment and Changeability equations is statistically significant (χ 2 = 7.5, p < 0.01).

As with the findings for the household SES index, individuals in the first subjective SES quartile are on average less satisfied with family life relative to those in all other subjective SES quartiles (Table 8). Furthermore, people in subjective SES quartile two are less satisfied with family life relative to those in quartile three (χ 2 = 10.7, p < 0.01) and quartile four (χ 2 = 46.5, p < 0.001), while individuals in quartile four report higher family-life satisfaction than those in quartile three (χ 2 = 20.2, p < 0.001). In the satisfaction with family life equation the Attachment and Changeability coefficients are not statistically equal (χ 2 = 20.2, p < 0.001). There is a significant difference in the subjective SES coefficients across the Attachment and Changeability equations (χ 2 = 18.9, p < 0.001), thus subjective SES has a different relationship with Attachment than with Changeability.

The SEM results suggest a clear relationship between higher SES and higher satisfaction with family life, although the relationship with household and subjective SES seems slightly stronger than with individual SES. This is not necessarily surprising, as persons are probably likely to place more weight on household-level SES factors (and hence their subjective evaluation of the household’s SES position) than on individual-level SES variables when assessing their family life.

The findings for the control variables reveal some interesting observations. The Attachment and Changeability coefficients within the satisfaction with family life equations are not statistically equal in any of the estimated models (all p < 0.01), suggesting that the FACI8 sub-scales have different relationships with satisfaction with family life. Higher reports of Changeability relate significantly to a higher satisfaction with family life score. However, there is no significant association between satisfaction with family life and Attachment. Thus, better family functioning relates to higher reports of satisfaction with family life, but this is only the case for family flexibility and not family attachment. In the models that control for individual SES (Table 6), Black persons report significantly lower family-life satisfaction scores when compared to all the other race groups. When controlling for household- and subjective SES (Tables 7, 8), however, White and Indian individuals are no longer more satisfied with family life than Black persons are. This might suggest that greater household living standards as well as better perceptions of household SES position are key explanations for why White and Indian persons are more satisfied with family life than Black persons. Married persons are on average more satisfied with family life than the never married, as are people who identify themselves as being religious compared to those who are not religious.

4.4 Multiple-Group Analyses

Table 9 reports the MGSEM results with the purpose of investigating whether the relationship between family functioning and satisfaction with family life differs depending on the particular SES quartile. All models were compared to a model where no constraints were imposed on the structural coefficients. For all three SES indices, the Chi square difference test indicates that the models with constraints do not fare significantly worse relative to a model with no constraints. This provides support for the assertion that family functioning has a similar relationship with family-life satisfaction across all SES quartiles. Thus, in the general SEM analyses it was reported that there is no significant relationship between Attachment and satisfaction with family life, and this also holds true across SES quartiles. Moreover, while Changeability is positively related to satisfaction with family life, this relationship does not differ according to SES quartile. The MGSEM findings therefore imply that family Changeability is an important predictor of satisfaction with family life and that this is the case irrespective of the SES quartile.

Table 9 Multiple-group results
Full size table

5 Discussion and Conclusion

This paper examined the determinants of satisfaction with family life in South Africa, with primary emphasis on the role of SES, specifically individual-, household-, and subjective SES. The findings reveal that higher levels of SES are associated with higher reported satisfaction with family life. Thus, people report higher satisfaction with their family lives when their personal level of SES is higher, if they live in households with higher SES, and if respondents perceive their SES to be higher. However, the nature of the relationship between SES and family-life satisfaction differs slightly depending on the particular SES index considered, in that household and subjective SES have the strongest association with family-life satisfaction.

As expected, people are likely to place more weight on household-level SES factors, as well as subjective household-level SES, than on individual-level SES factors when assessing how satisfied they are with family life. The positive association between SES and satisfaction with family life is in a sense consistent with the findings of Agate et al. (2009) and Yamamura (2014), who found a positive relationship between satisfaction with family life and income. However, the results are not directly comparable given this paper’s use of broader SES indicators and not income only.

Multiple-group SEM was also conducted with the purpose of examining whether family functioning relates differently to satisfaction with family life depending on the particular SES quartile. The MGSEM results suggest that family flexibility is positively related to satisfaction with family life and that this relationship does not differ depending on the SES quartile considered. Thus, family flexibility remains an important factor in determining family-life satisfaction, irrespective of SES.

Overall, this paper’s findings imply that poverty alleviation programs and improvements in factors such as household living standards and infrastructure are likely to improve satisfaction with family life via an associated improvement in SES. Moreover, specially designed family strengthening programs can facilitate greater flexibility of family relationships, which in turn may enhance satisfaction with family life across all SES classes.

This is the first study to examine the determinants of satisfaction with family life in a developing country setting, with particular emphasis on the role that SES, at various levels, plays in explaining satisfaction with family life. Besides these strengths the paper, however, also has some limitations. Firstly, no comments can be made about causality since the data are cross-sectional. Secondly, the question measuring satisfaction with family life is only asked of the respondent and not all other household members. It is therefore not possible to consider potential intra-family differences in reported satisfaction with family life. Another important limitation is that it is not possible to know how respondents may think of “family” when asked about satisfaction with their family life, as it is self-defined. For some, “family” may mean only those close members living in the same household, whereas for others it may mean family members within the household as well as outside the household. However, the data do not allow for any determination of how respondents define “family”.

There are interesting avenues for future research. Firstly, though this paper focused only on South Africa, it would be worthwhile to consider the predictors of satisfaction with family life across various countries using cross-national data. Secondly, the availability of panel data would make it possible to control for unobserved heterogeneity and move towards making assertions about causal relationships between satisfaction with family life and SES.