Examining the Determinants of Poland’s Middle Class Life Satisfaction

The middle class plays a fundamental role in creating consumer demand and economic growth (Dabla-Norris et al., 2015; Easterly, 2001) and influences socio-economic stability (OECD, 2019; Pressman, 2007). It affects the economy through consumption and capital accumulation (Chun et al., 2016), innovation and diversification, and expansion of production (Banerjee & Duflo, 2008). In addition, the middle class plays an essential role in creating inclusive institutions that secure private property, ensure markets' openness, and stimulate entrepreneurship development, which was not without significance for the systemic transformation processes in Poland (Piątkowski, 2019). However, it must be noted that Poland's middle class is still being created (Wyżnikiewicz, 2021). Additionally, social status and criteria of class positioning became connected to more extent to one’s material standard of living in Poland after the demise of the communist rule at the end of the 80s twentieth century (Zawadzka, 2018). Focusing more on individual effort to increase one’s income than on education or professional status related earlier to the affiliation to intelligentsia makes the analysis of class stratification from the economic point of view very important (Kulas, 2017).

Being part of the middle class was once a widely held aspiration, but the growing income inequalities have threatened this. Temin (2018) underlines that the middle class vanishes, resulting in American society's division between the rich and the poor. The progressive erosion of the middle class is also observed in Europe (Suter et al., 2020). However, after systemic transformation, Poland and a few other CEE countries experienced an increase in the size of the middle class (Vaughan-Whitehead, 2016). In Poland, the middle class's share in income distribution increased during the last decades from 51% in 2005 to 55% in 2021, when the middle 60% of income distribution is considered (Eurostat, 2022). In comparison, the average EU middle class share in income distribution was around 53% (Eurostat, 2022).

However, the meaning of the middle class is not clearly defined. The middle class is associated with the fundamental element of social structure in capitalist economies (Domański, 2012). The middle class is a highly complex community and does not constitute a homogenous class. The two ways to conceptualize the middle class are based most often on professional status (usually used by sociologists; Domański, 2012; Erikson & Goldthorpe, 1992; Wright, 1997) or income (traditionally used by economists; Atkinson & Brandolini, 2013; Easterly, 2001; OECD, 2019; Suter et al., 2020). Therefore, the middle class comprises people with different levels of education, age, material standards, consumption preferences, and esteem (Domański, 2012). The term middle class is used more often in sociology than in economics. From a sociological perspective, the middle class is often defined by mechanisms of exclusion over acquiring education and skills (Wright, 1997). Besides, much more attention is paid to experience based on meritocratic rules. According to Wyżnikiewicz (2021), belonging to the middle class requires fulfilling simultaneously two criteria, i.e., income (achieving high enough income that would be higher than the average income in the country) and participation (playing an important role in the economic, cultural, social, scientific, political, or artistic life in a particular country). The criteria adopted for the class classification are often connected with professional activity, education (Domański, 2012), and participation (Wyżnikiewicz, 2021) rather than only income from a strictly economic perspective. In sociology, ISCO-08 (International Classification Standard of Occupations; ILO, 2022) and EGP schema proposed by Erikson & Goldthorpe, (1992) and by Erikson and Goldthorpe (1992) are often applied. The classification proposed by ISCO-08 emphasizes occupational roles and responsibilities, while the EGP proposal takes into account other aspects that define belonging to social class i.e. the level of education required to have a job, etc. In the sociological approach, e.g., the upper middle class consists of professionals and technicians, farm proprietors, and managers. To lower middle class, sales and services workers belong. Other classes that are often distinguished are the upper, working, and lower classes.

Income-based approaches in economics identify the middle class as middle-income groups, i.e., the groups located in the middle of the income distribution, belonging neither to the poor nor the rich. There is, however, no consensus regarding the criteria for middle-class stratification (Suter et al., 2020), and therefore these are arbitrarily imposed by the researcher (Kot, 2008). In the present study, the middle class will be defined from an economic perspective as the middle-income class by taking the disposable income per person in the household, falling between 75 and 200% of its median (Atkinson & Brandolini, 2013; Easterly, 2001; OECD, 2019). Another conceptualization of the middle-income class stratum considers the income groups between the third and the eighth decile or those with an equivalised disposable income between 60 and 200% of the median income (Suter et al., 2020). What is important is that by considering equivalised disposable income per person in a household as belonging to the middle-class criterion, all the members of households are included in the middle-income group, not only those who reach a particular level of education, occupation, and income.

The economic approach to middle-class identification has been mainly based on income. At the same time, it should also consider other key elements in the evolution of material living standards, such as wealth and debts, savings level, and earnings insecurity, when evaluating LS (Suter et al., 2020). Whether a material standard of living buys happiness needs to incorporate income (individual or per person in the household, representing a flow of money), wealth (possessed assets), and level of savings representing stocks of wealth) in the LS analyses. Wealth may increase the individuals' LS by enhancing household stability, increasing future orientation, decreasing risk-taking behavior, and fostering social influence (Han & Hong, 2011; Sheraden, 1991). However, people who prize material goals more than other values tend to be substantially less satisfied with their lives unless they are rich. Thus, more money may enhance LS when it means avoiding poverty and living in a developed nation. Still, income appears to increase LS little over the long term when more of it is gained by well-off individuals whose material desires rise with their incomes (Diener & Biswas-Diener, 2002). Even though an important moderator of the effects of income on LS is the society's wealth, the analyses are often conducted from a macroeconomic perspective (Veenhoven, 1991). Easterlin (2011) concludes that people with higher incomes, on average, are happier than those with lower incomes. However, there is a point of diminishing returns. Diener et al. (1985a; b) showed that very wealthy people were only slightly happier than a control group who had residences in the same geographical area. Diener and Biswas-Diener (2009) addressed the relationships between income, wealth, and LS. They concluded that once people have high incomes, additional increases in wealth have a minimal influence on LS, suggesting that added income beyond modest affluence no longer helps answer important needs (Diener & Biswas-Diener, 2009, p. 138). Also, very little is known about possible links between debts and LS, even though indebtedness may affect the perceived financial stability and satisfaction with the material standard of living. Individuals in the highest debt-to-wealth quartile were at risk of lower mental well-being, a higher number of depressive symptoms, and lower quality of life (Hiilamo, 2020). However, the risks associated with household debt burden at an individual level, such as its effects on mental well-being, are not much discussed in the literature.

When work-related issues are considered regarding LS, the subject literature most often focuses on the kind of professional activity, not professional status. The effect of unemployment on LS is negative, which was documented in many studies (Blanchflower & Oswald, 2000; Clark & Oswald, 1994; Graham, 2011; Lucas et al., 2004). Dolan et al. (2008), when considering causal relationships, revealed that poorer health and lower productivity of the unemployed may be the reason for their lower LS. However, when self-employment is considered, the Axerald et al. findings reveal the positive contribution of self-employment to individuals' LS (Axerald et al., 2020). Numerous analyses link professional status and job satisfaction as one of the most important dimensions of well-being (Clark, 2015). Income is a characteristic that is often thought to influence job satisfaction. It is generally believed that higher pay leads to higher job satisfaction (Redmond & McGuinness, 2020). Job satisfaction literature analyses different groups of people, i.e., academic professionals, women, and young adults (Bender et al., 2005) or different sectors, like social services, banking, and public health (George & Zakkariya, 2018; Goula et al., 2022).

Numerous studies on LS determinants can be found to the author's knowledge. However, none of them relate to the middle-income class.

Regarding the different philosophical views on the evaluation of LS, one of the concepts—life satisfaction (prudential happiness), alongside eudaimonia and hedonism, has been studied through the design of the questionnaire with the Satisfaction with Life Scale (SWLS) (Sirgy, 2021). The concept of LS, which is interchangeably used with subjective well-being, refers to the individual's well-being assessment according to their own opinion (Sirgy, 2021). When LS is considered, respondents answer questions about their general evaluation of life (Diener et al., 1985a, 1985b; Tatarkiewicz, 1976).

Prudential happiness refers to a state of well-being. It is achieved when a person reaches a high mental and physical well-being (Veenhoven, 2000) and leads a good and engaged life (Sirgy & Wu, 2009). LS is viewed as a 'cognitive' conceptualization of subjective well-being. It involves judgments of the fulfillment of one's needs and goals. In this perspective, life satisfaction is defined as 'a cognitive judgmental process dependent upon comparing one's circumstances with what is thought to be an appropriate standard (Diener et al., 1985a, 1985b)'.

When LS is measured, both emotional aspects of the construct (positive and negative affect) and cognitive-judgmental aspects should be considered (Diener et al., 1985a, 1985b). LS refers to cognitive judgments of life satisfaction that compare one's circumstances with what is thought to be a suitable benchmark (Diener et al., 1985a, 1985b). Satisfaction With Life Scale (SWLS) is a five-item scale with favorable psychometric properties (Diener et al., 1985a, 1985b), where the items are:

Each is scored on a 1 to 7 Likert scale (1-strongly disagree, 2-disagree, 3-slightly disagree,4-neither agree nor disagree, 5- slightly agree, 6-agree, 7-strongly agree). Therefore, the possible range of scores on the questionnaire is 5 (low satisfaction) to 35 (high satisfaction).

Classification is an operation that places each individual under study in one of two specified classes (binary dependent variable) according to the characteristics of the individual, which are identified as independent variables (Tuffery, 2011).

Trees are built through an algorithm that recursively partitions the data into nodes by iterated binary splits. Therefore, the root node (i.e., the whole sample) is divided into other nodes by following a set of rules that finds among all predictors the ones that allow for the most discriminative split. The procedure is accomplished by testing the impurity level of all possible splits. This procedure continues by creating branches and other nodes until certain conditions are met.

The construction was such that each terminal node (the leaves) mainly consisted of individuals of a single class. An individual is assigned to a leaf, and therefore to a particular class, with a reasonably high probability when it conforms to all the rules for reaching this leaf. The rules for all the leaves form the classification model (Tuffery, 2011). Finally, the terminal nodes, or leaves, define the predicted type for each individual whose characteristics match the traced path (Breiman et al., 1984). After a complete decision tree is built, there is usually the need to prune some branches. Indeed, by pruning the tree, the results are easier to understand and more precise in classifying alternative data sets (Han et al., 2011).

In a classification technique, the error rate, i.e., the proportion of incorrectly classified individuals, must be as low as possible. The present study used the tree-based method to identify the factors determining the LS of the middle-income class in Poland because of many predictors measured with weak scales (Likert scale). CART is one of the most effective and widely used decision trees (Breiman et al., 1984). The IBM SPSS Imago PRO software was used to build the CART, where this classification method is found under CRT.

Firstly, based on the Satisfaction With Life Scale (SWLS), the range of total scores on the questionnaire was from 5 (low satisfaction) to 35 (high satisfaction). The descriptive statistics of the results in the whole sample and the groups of low-income, middle-income (with 3 subgroups), and high-income classes are presented in Table 1.

The mean LS ranged from 18.25 in the low-income class to 23.50 in the high-income class. Furthermore, there were slight differences in the mean LS observed in the subgroups of the middle-income class. The level of LS in those subclasses was close to the whole sample average of 20.73. The median and mode also increased with the income class, ranging from 19 (median) and 20 (mode) in the low-income class to 24 (median) and 26 (mode) in the high-income class, respectively. However, when diversity in the middle class only is considered, a lower median and mode was observed in the mid-income middle class than in the remaining middle-income classes. With negative skewness, the distribution is left-skewed in all income groups. Moreover, the analysis of distribution in percentiles reveals relatively low internal diversity in the middle-income class subgroups and high diversity when intergroup comparison is studied. The bottom 20% of representants in all subgroups in the middle-income class have life satisfaction lower than 16, while the top 20% of the middle-income class have life satisfaction above 25 (lower-middle income and mid-income class respectively) and 26 (for higher-middle income class). The cut of points for 1st quintile is increasing from 13 in the low-income group to 20 in the high-income group. When the cut-off point for the 5th quintile increases from 23 in the low-income class to 27 in the high-income class. The interquartile range shows the lowest internal diversity in life satisfaction in the high-income class. The subgroups of the middle class have similar diversity in life satisfaction to the whole sample with the exception of the mid-income class with little lower diversity calculated as the interquartile range. Positive kurtosis indicates a leptokurtic distribution, except in the higher middle-income class. The histogram of the LS distribution in the whole middle-income class aw well as in the subgroups of the middle-income class is presented in Fig. 1.

To explore more deeply the internal diversity in the distribution of life satisfaction in particular income groups, the SWLS scores were divided into the five equal groups (quintiles) according to the cut-off points presented in Table 1 and their percentage shares in each income groups were calculated and presented in Table 2. The results reveal the general tendency of an increasing share of those who are satisfied and very satisfied with their lives in higher-income groups (from 21% in the low-income group to 64% in the high-income group). Even though middle-income class subgroups aren’t highly diversified, two characteristics of the groups were revealed. Firstly, the mid-income middle class consists of a relatively lower share of satisfied and very satisfied with life (34%) in comparison to other middle-income groups (40% and 41%, respectively). Also, more than half (52%) of mid-income middle class members share low and very low levels of satisfaction with life, which is 7 and 11 percentage points higher share than the ones observed in the lower middle and the higher middle-income classes.

In the next step of the analysis, the sample was divided into two groups of satisfied and dissatisfied members of each income class regarding the total score in reference to the mean score in the sample. Such action is acceptable on Likert scale scores (DeVellis, 2016; Likert, 1932) and allows the CART method to be used. Therefore, when LS was considered, those who obtained more than 20.73 points in the SWLS questionnaire were satisfied (binary variable: 1 = satisfied, 0 = dissatisfied), and the dissatisfied were recognized as those with lower levels of total scores. The detailed characterization of the LS based on SWLS in each income group is presented in Fig. 2.

The interclass comparison reveals interesting results about the share of satisfied and dissatisfied respondents in each income group. When LS is considered, 33% of members of the low-income group, 55% of lower-middle-income, 52% of mid-income, 59% of higher-middle-income, and 78% of the high-income class were satisfied. In the higher-income class, the proportion of satisfied individuals increased. However, in the middle-income group, a slightly higher proportion of satisfied individuals was observed in the lower-middle-income and higher-middle-income brackets compared to the mid-income bracket.

In the next step to explore the links between income class stratification and life satisfaction the Pearson’s chi-square test of independence was applied (Benhamou & Valentin, 2018; Osińska, 2008). Because the result obtained a chi-squared probability was less than the critical point of 0.05. The null hypothesis that the LS is independent of income classes is rejected. In turn, the alternative hypothesis is accepted. It means that income class and satisfaction with life are related to each other (chi-square = 88.19, df = 4, p < 0.001). Moreover, the Spearman correlation result (0.22) revealed weak but statically significant (p < 0.001) positive correlation between the income classes and LS.

The decision tree technique was used to identify the determinants that diversify middle class members between two predetermined groups: satisfied and dissatisfied with their lives. It allowed the selection of variables that best predicted being satisfied by middle class members.

The dependent variable was LS based on SWLS, which was a dummy variable equal to 1 when an individual reports a level of LS higher than the average in the whole sample (20.7) and 0 otherwise. 53.6% of the middle-income class in the sample was more satisfied with their lives than the average (satisfied). In comparison, 46.4% were less satisfied (dissatisfied).

The LS measured by SWLS was chosen because it separated the sample into two groups, called nodes (the first node of the tree is the root), each containing the most significant possible proportion of individuals in a single group. The same operation was repeated on each new node obtained until no further separation of the individuals was possible or desirable. The root contained 800 members of the middle-income class, including three subgroups: lower-middle-income class, mid-income class, and higher-middle-income class.

Several explanatory variables that identify different aspects of the individual's life were analyzed in the CART. Regarding the literature characterizing the middle class, selected survey results were chosen for the CART analysis, namely economic factors (income, level of savings, value of assets, sources of income, aspiration for income), professional status, and form of professional activity, education level, and kind, as well as demographical factors (city size, age, etc.) that may have played an important role in the determination of the middle-class LS. The list and description of independent variables are attached in Appendix Table 8. Applying the CART approach is essential for handling a relatively large set of independent variables (13) in the analysis.

To build the CART, firstly, 70% of individuals in the middle-income group were randomly chosen (i.e., 'training sample') to grow the tree. The remaining 30% of the sample (i.e., 'testing sample') was used to test the precision of the model predictions. Secondly, the tree was constructed by imposing restrictions on the minimum size of the node (the minimum number of items in the parent node = 60; the minimum number of items in the child node = 30) and the maximum tree depth equal to 5. The tree was pruned to minimize the misclassification error by considering a penalty for additional nodes. The Gini index was used to measure the level of node impurity as the standard criterion used in classification trees.

Among the 13 explanatory variables in Appendix (Table 8), the tree selected only three (two economic and one professional criterion). The variables that significantly diversified the middle-income class LS were the value of assets (wealth), professional status, and level of savings.

The first split depends on the individual's assets: those whose value of assets was lower than 200 K PLN went to the left branch, and the others (with assets higher than 200 K PLN) went to the right. Thus, the first split shows the importance of economic factors. However, it was not net individual income but wealth that appeared to be an important determinant of LS in the middle-income class in Poland. The following splitting variable, in the left branch, was professional status. Those working in services, qualified and unqualified workers, unemployed, and technicians were mainly unsatisfied. Administrative workers, pensioners, retirees, self-employed, housewives, students, and farmers mainly were satisfied (Fig. 3).

Professional status was also a substantial following splitting variable for those with assets higher than 200 K PLN. Qualified and unqualified workers, retirees, higher education specialists, self-employed, and pensioners were more satisfied than administrative workers, students, technicians, unemployed, and housewives (some of the professional statuses are presented in Fig. 3). However, having savings of more than 20 K PLN was the variable that significantly differentiated those satisfied in this group. 81.5% of those with more than 20 K PLN savings had a particular professional status (qualified and unqualified workers, retirees, higher education specialists, self-employed, pensioners), and assets higher than 200 K PLN were satisfied members of the middle-income class in Poland. Interestingly, different independent variables in the leaves (economic and professional) are identified, confirming that middle-class LS is determined by many aspects of life (Fig. 3).

In the next step, the evaluation of the classification was conducted based on the test sample (Table 3), and the following measures were calculated: accuracy = 57.5% with the overall error rate = 19.2%; sensitivity = 66.1%; specificity = 47.2%; precision = 60.0%; F1 score = 62.9%.

Among the factors with the highest relative importance in differing the middle-income class into satisfied and dissatisfied among all the explanatory variables were the professional status, value of assets, and monthly net personal income (Fig. 4). Among the identified variables (Fig. 4) three factors, i.e. wealth, professional status, and savings are repeated from the presentation of CART results as those which mostly increased the probablility of achieving relatively high satifacion with life.

In the next step of the analysis, based on the factors identified by CART analysis that, to the greatest extent, increase the probability of middle-income class LS, namely wealth, professional status, and savings level, the cross-tables were studied. That action allowed to identify the differences in the structure of satisfied members of each income class in regard to professional status (Table 4), wealth (Table 5), and savings (Table 6).

The higher the income class, the higher the share of satisfied managers and high education specialists. This result suggests that for the high life satisfaction of those professions, a higher disposable income per person in the household is needed. Also the higher concentration among satisfied students, unemployed, and housewives is observed among members of low and lower-middle-income classes. High diversity among the shares of satisfied technicians, office workers, farmers, self-employed, students, unemployed, and housewives in the subgroups of middle class was identified. The highest share of dissatisfied housewives was identified in the low-income class (Table 4).

Almost half of those, who are satisfied, and own assets valued at more than 500 K PLN are members of a high-income class. Those, who possess relatively low-value assets (below 200 K PLN) and are satisfied, are mostly concentrated in the low and lower-middle-income classes. Similarly, as for those, who are dissatisfied. It suggests that higher assets owned by members of higher income classes increases life satisfaction (Table 5).

When life satisfaction in terms of income classes and savings is analyzed the higher concentration of those satisfied is observed among low and lower-middle income classes when individuals possess relatively low levels of savings (below 10 K PLN) and among high income class when individuals have high level of savings (more than 100 K PLN) (Table 6).