Social differences are ubiquitous. Differences in resources, standing and power across individuals and groups of individuals pervade every society, even create the phenotype and dynamics of societies in the first place. Yet social differences are not simply accepted. While the concrete content, use and notion of justice norms differs around the world, justice norms itself are universal among humans and monitor the acceptability of social conditions. The Arab Spring in 2010 and the Occupy Wall Street movement in 2011 are prominent examples of what can follow if social conditions are no longer accepted and what forces perceived injustice can unleash. Perception of unjust social differences divides groups within societies and can ultimately threaten social peace and stability of governments. It is therefore of utmost political interest to understand why social differences are accepted and when they are not.
Accordingly empirical research about social differences has gained momentum over the past decades. Using the concept of Social Inequality’ research focused on where and how much of inequality exists and which groups have which attitudes towards inequality. The causal question though, why social differences are perceived that way, was given a minor role. Hadler (2005) was first to ask explicitly why different income ratios are accepted, eliciting the importance of personal income, of feeling under-rewarded and of meritocratic and functionalistic ideologies as legitimization. Newer research gives more extensive accounts about inequality and its acceptance (Lang, 2017) and examines further possible causal factors for accepting inequality like perceived upwards mobility, future time perspectives, mind-sets etc.
Contents
1 Introduction
2 Theory and Hypotheses
3 Data
4 Measurement
4.1 Acceptance of Social Differences
4.2 Personal Net Income
4.3 Relative Deprivation
4.4 Expectation: Future Economic Situation
4.5 Control Variables
5 Method
6 Analysis
6.1 Descriptive Statistics
6.2 Results Hypothesis 1
6.3 Results Hypothesis 2
6.4 Results Hypothesis 3
6.5 ModelDiagnosis
7 Summary and Discussion
References
Appendix: Tables and Figures
List of Tables
1 Descriptive Statistics of all used Variables (unweighted)
2 Average Marginal Effects of Income on Acceptance of Social Differences
3 Average Marginal Effects of Income on Acceptance of Social Differences
with Control Variables
4 Cross Table of Income and Relative Deprivation (unweighted)
5 Average Marginal Effects of Income and Relative Deprivation on Acceptance of Social Differences
6 Average Marginal Effects of Income and Relative Deprivation on Acceptance of Social Differences with Control Variables (not shown)
7 Cross Table of Income and Expectation of Future Economic Situation by Acceptance of Social Differences (unweighted)
8 Average Marginal Effects of Income and Expectation of Future Economic Situation (Interaction) on Acceptance of Social Differences with Control Variables(notshown)
9 Correlation Matrix of Independent Variables
10 Average Marginal Effects with Step-Wise Build-Up of Multinomial Logistic Regression Model
11 Comparisons of Average Marginal Effects between Multinomial and Ordinal Regression
List of Figures
1 Simple Hypothesis H1
2 Mechanism Hypothesis H2
3 Interaction Hypothesis H3
4 Control Variables
5 Control Variables with assumed Causal Paths
6 Acceptance of Social Differences by Income
7 Acceptance of Social Differences by Relative Deprivation
8 Marginsplot of Accepting Social Differences given Income
1 Introduction
Social differences are ubiquitous. Differences in resources, standing and power across individuals and groups of individuals pervade every society, even create the phenotype and dynamics of societies in the first place. Yet social differences are not simply accepted. While the concrete content, use and notion of justice norms differs around the world, justice norms itself are universal among humans (Fischer et al., 2011) and monitor the acceptability of social conditions. The Arab Spring in 2010 and the Occupy Wall Street movement in 2011 are prominent examples of what can follow if social conditions are no longer accepted and what forces perceived injustice can unleash. Perception of unjust social differences divides groups within societies and can ultimately threaten social peace and stability of governments. It is therefore of utmost political interest to understand why social differences are accepted and when they are not.
Accordingly empirical research about social differences has gained momentum over the past decades. Using the concept of 'Social Inequality’1 research focused on where and how much of inequality exists and which groups have which attitudes towards inequal- ity.2 The causal question though, why social differences are perceived that way, was given a minor role. Hadler (2005) was first to ask explicitly why different income ratios are accepted, eliciting the importance of personal income, of feeling under-rewarded and of meritocratic and functionalistic ideologies as legitimization. Newer research gives more extensive accounts about inequality and its acceptance (Lang, 2017) and examines further possible causal factors for accepting inequality like perceived upwards mobility, future time perspectives, mind-sets etc. (Cheng et al., 2019; Fehr et al., 2020; Savani and Rattan, 2012).
An interesting question seldom asked is if objectively given factors (income, GDP, societal structures etc.) or rather other individual-level perceptions shape the perception of social differences. While recognizing that such factors cannot be separated, since objective factors from the environment influence perceptions, which cause actions that again influence the environment, it appears intuitive that subjective factors on the individual level are better suited when it comes to explaining the perception of social differences as they are ‘closer’ to the explanandum. There has been some evidence that in line with this intuition the effect of objective factors on attitudes about inequality is mediated by subjective factors (Kuhn, 2019).
Following this trail we look at the theoretically solidified effect of income on the perception of social differences, trying to answer the question whether income actually has an direct effect on accepting social differences? Drawing from various theories we first prepend the hypothesis that higher income leads to higher acceptance of social differences, but argue that this effect of income isn’t a standalone effect, but has to be viewed in context with subjective factors like relative deprivation and expectations of the future. This will be elaborated in the following section. Section 3 and 4 will then outline the used data and the conceptual framework of used variables, respectively. After introducing multinomial logistic regression as the method for analysis in section 5, section 6 will analyze the data, test the posed hypotheses and present the results. Finally section 7 will summarize the results and give an outlook.
2 Theory and Hypotheses
The ‘Absolute Deprivation Thesis’ predicts, based on rational self-interest, that individuals with low income hope to gain from equality and therefore object social differences, while individuals with high income fear to lose from equality and therefore accept social differences (Szirmai, 1986).
Social Identity Theory is less concerned about rational calculation and resolves around the value membership has for someone’s identity instead. According to Social Identity Theory membership of a social group or category helps individuals to build identity within a social context. This social identity based on membership of a social group can be negatively or positively connoted depending on the esteem that social group holds compared to other social groups. It is claimed that individuals naturally strive for a positive social identity originating from a positively discrepant comparison with another social group. If the comparison is negatively discrepant, meaning the other group is more valuable in the light of relevant comparing attributes, individuals can either change the social group, compare their group with a different group, or reevaluate the comparison, e.g. by devaluating the other group in some way (Tajfel and Turner, 1979). Perceiving oneself as the member of a group and comparing this group to other groups does not need much of a catalyst, the pure awareness of groups suffices (Tajfel and Turner, 1979: 38f.) like localizing oneself within income groups in a survey. Being part of an income group cannot by easily changed so cognitive strategies have to be applied to reevaluate a comparison. With Social Identity Theory can then be argued that individuals with low income should experience a negatively discrepant social comparison and try to reevaluate this by seeing social differences as unjust and refuting them. Individuals with high income should experience a positively discrepant social comparison and try to maintain the resulting positive social identity by seeing social differences as just and accepting them.
Self-Serving Bias (Heider, 1958) complements these arguments by concentrating on identity which is not solely determined by membership of a social group. Self-Serving Bias has many facets (Shepperd et al., 2008) but relevant in this case are attributions to serve self-enhancement and one’s self-image. Individuals tend to attribute features that enhance self-worth and are in alignment with their wished identity internally, while features that devalue self-worth and contradict their wished identity are usually attributed externally (Snyder et al., 1976; Tetlock and Levi, 1982). Individuals who earn a high income should attribute this high income internally, seeing it as the result of intelligence, hard work etc. and therefore as just, being inclined towards accepting social differences. Individuals who earn a low income on the other hand should attribute this low income externally, seeing it as the result of unjust circumstances, social structures etc., being inclined towards rejecting social differences. Together those theories point in one direction (see figure 1):
H1: The higher the net income of an individual, the more likely this individual will accept social differences.
Income can be considered an objective factor that is given and stable, so to say an environment that has to be dealt with. Income is therefore an factor that does not have a direct causal effect, but only an indirect one as environment which irritates a (psychic) system according to its Eigenlogic (Luhmann, 1991; Peschl, 1990). Two individuals with the same income can have vastly different evaluations about their income being just. The perception of justice is always a relative process that compares one’s own situation with one’s own expectations and the situation of others within a shared context (Hegtvedt and Parris, 2014). Perceptions are therefore highly complex and dependent on networks of other perceptions. Rather than income itself, the internal processing that decides if one feels justly rewarded by that income is crucial (Gijsberts, 2002), which is influenced by the amount of that income, but not determined. This is captured by the concept of Relative Deprivation, which signifies “the extent of the difference between the desired situation and that of the person desiring it” (Runciman, 1966: 10). This leads to the second hypothesis (see figure 2):
H2: The effect of personal net income on accepting social differences is at least partly mediated by relative deprivation.
Another way to fine-tune the association between income and acceptance of social differences is to interact it with future time perspectives (Cheng et al., 2019). Humans include expected futures into their judgment. The same present income is differently judged if a better personal economic situation is expected in the future, a worse one or no change at all (Flavin, 1981). Arguing along the lines of Self-Serving Bias, expecting a better economic situation in the future is attributed internally and inclines towards seeing social differences as just, especially among high income individuals as high income individuals have to legitimize an expected betterment of an already advantaged income situation. Social Identity Theory would predict the same, albeit rather applying to individuals with low income, since the perceived possibility of upwards mobility, the prospect of acquiring a positive social identity by getting into a better (income) group, makes social differences appear less problematic.
Social Identity Theory further argues that perceived threat to one’s social group’s status and hence one’s social identity leads to increased discriminatory behavior towards the threatening group (Tajfel and Turner, 1979: 45). A major threat to group status is the leveling of distinctions, which positively discriminate one group from another. Individuals who perceive their social identity threatened, e.g. by expecting a worse economic situation in the future, should therefore be more likely to resort to discriminatory behavior, regarding social differences as justified, compared to individuals who don’t expect a change in their economic situation.
Putting this together you’ll notice that for high income individuals any expected change in their economic situation leads them to be more likely to accept social differences, which neatly allows the following hypothesis (see figure 3):
H3: Individuals who earn a high income and expect a change of economic situation in the future are more likely to accept social differences than individuals who earn a high income, but don’t expect any change in their economic situation.
3 Data
The data being used are from the Allgemeine Bevolkerungsumfrage der Sozialwissenschaften ALLBUS 2018 containing 3477 cases and 708 variables.3 The ALLBUS sample is collected every 2 years using stratified sampling on German municipalities and subsequent random sampling on the municipality registers. The 2018 version depicts all persons within Germany that reside in private households and were born before January 1 2000, which is considered representative of the German population. The merit of the 2018 version is having ‘Social Inequality and Social Capital’ as a main topic, which is ideal for the given research interest. Furthermore it offers a wide area of data in various fields (demography, economy, attitudes etc.), which allows an comprehensive amount of control variables. A usual downside of the ALLBUS is that individuals who don’t have a residence or aren’t included in municipality registers, as well as individuals who are hard to reach, e.g. commuters, are not represented or underrepresented, respectively. If those groups have systematic perceptions of social differences, that won’t find its way into the results. The same pertains to non-compliance.4 Another problem is oversampling of cases from new federal states. Since it is intended to generalize results to the whole German population and residence in a new or old federal state is considered to have an influence on the acceptance of social differences, this over-representation of new federal states cannot be ignored. All computations will therefore be weighted with the actual distribution.5
4 Measurement
4.1 Acceptance of Social Differences
Acceptance of Social Differences (ASD) will be measured by item im21 (social differences are just): ”On the whole, i consider the social differences in our country just”, with the given categories 1 ”Completely agree”, 2 ”Tend to agree”, 3 ”Tend to disagree” and 4 ”Completely disagree”. Regarding something as just doesn’t necessarily mean that it is accepted, e.g. injustice can be recognized, yet followed, if the injustice serves interests. As justice can be considered an important value in Germany that serves as a selection rule for acceptance and there’s not much space for self-interest in the survey, it is supposed that ‘accepting’ and ‘regarding as just’ are congruent. Another problem is the broad formulation of the item. It’s up to the test person to decide what she understands by ‘social differences’. Given the preceding items im19 (income differences increase motivation) and im20 (differences in social positions are acceptable) it can be assumed social differences were generally interpreted as income and status inequality due to priming.
Creating an index out of im19, im20 and im21 was considered, surprisingly though the items correlate not that strongly,6 consequently item im21 is used alone as the most suitable one in regard to the research question.
4.2 Personal Net Income
For Personal Net Income the continuous variable inc (monthly net income per person, open+list) is recoded into a categorical variable to obtain more meaningful results.7 Income is divided into three income groups. Similar to the classifications by the German Institute for Economic Research (DIW Berlin) and other research (Bosch and Kalina, 2016: 73) the first category 1 ”Low Income” comprises all individuals with income lower than 80% of the median income within the sample (0 - 1199). The second category 2 ”Medium Income” includes all individuals from 80% of the median up to 200% of the median (1200 - 2999). The third category 3 ”High Income” includes all individuals with 3000 Euros monthly net income and more.
4.3 Relative Deprivation
Relative Deprivation will be measured by item id01 (just share on living standard): ”Compared with how others live in Germany: Do you think you get your fair share [on the living standard]”, with the categories 1 ”fair share”, 2 ”more than fair share”, 3 ”somewhat less of fair share” and 4 ”very much less than fair share”. Categories 1 and 2 are combined to create the three categories 1 ”Under-Rewarded”, 2 ”Justly Rewarded” and 3 ”Over-Rewarded”. This measurement assumes the society as a whole as reference group with which the own situation is compared. This is a usual approach, but it undermines the importance of individual reference groups (Yitzhaki, 1982).
4.4 Expectation: Future Economic Situation
Expectation of the Future Economic Situation will be measured by item ep06 (economic situation respondent in one year): ”What will your own financial situation be like in one year?” with the categories 1 ”Considerably better than today”, 2 ”Somewhat better than today”, 3 ”The same”, 4 ”Somewhat worse than today” and 5 ”Considerably worse than today”. Categories 1 and 2 are combined as well as categories 4 and 5. Final categories are 1 ”Positive Change”, 2 ”No Change” and 3 ”Negative Change”.
4.5 Control Variables
Included control variables are education, education of parents, age, general health, gender, residence in new or old federal state, confession, marital status and political attitude. Control variables are chosen according to modern causal analysis, the backdoor criterion in particular, essentially adjusting for all third variables that have an independent effect on both the explanatory (Personal Net Income) and the explained variable (Acceptance of Social Differences) (Pearl and Mackenzie, 2018: 157ff.).
Education is measured using the variable educ (general education) combining categories B (finishing school without certificate) and C (Mittelschulabschluss) into category 1 ”Low Education”, D (Mittlere Reife) into 2 ”Medium Education” and E (Fachabitur) and F (Abitur) into 3 ”High Education”, dropping the remaining cases. Higher education leads on average to higher income (Anger and Geis, 2017: 45f.). Education should further shape cognitive capacities by which social differences are judged.
Education of parents is measured using the variables feduc (general education father) and meduc (general education mother). Both variables are coded as educ.Anew variable is then generated from feduc and meduc, which has category 2 (”High SEB”) if at least either the father or mother has high education and category 1 (”Medium SEB”) if at least the father or mother has medium education, yet neither has high education. The remaining cases are coded 0 (”Low SEB”). The education of parents serves as proxy for socioeconomic background, which shapes life opportunities and attitudes from early on.
Age is controlled for using the variable age, subdivided into age groups. Individuals aged 18-29 are combined into category 1, individuals aged 30-39, 40-49 and 50-66, into category 2, 3 and 4, respectively. Everyone aged 67 and higher is part of category 5. Depending on age cohorts life situations and world views differ, which also affects ASD. Income is generally increasing with age, yet decreases again when entering pension. The boundary at age 67 between the last two age groups accounts for this.
General health is measured via variable hs01 (health condition respondent) combining bad and poor health into category 0 ”Poor Health”, satisfactory health into category 1 ”Average Health” and very good and good health into 2 ”Good Health”. The influence of health on income seems to be a circular one, where low income affects health in various ways like having less money to spend on qualitative food, increasing anxiety etc. (Lenhart, 2018), while bad health lowers the chance of getting or keeping a well paid job. It’s assumed that bad health inclines towards more negative perceptions, viewing social differences more negatively.
Gender is measured via variable sex (gender respondent). The influence of gender on income has been extensively researched and titled gender pay gap (Anderson et al., 2001). Furthermore women are rather socialized towards being social, helpful and cooperative, while men are rather socialized towards being competitive (Carter, 2014: 255). The latter should consequently be more accepting of social differences, while the former should be less accepting of social differences.
Residence in new and old federal states is measured via eastwest, which was determined by the location of the interview. Income, living standard, expenses etc. differ between regions as well as attitudes and perceptions. People in market societies, which transformed from state-socialistic societies (Soviet Union), have been shown to be less accepting of social differences (Gijsberts, 2002).
Confession is measured via rd01 (confession respondent). All categories are retained. Different confessions could have different belief systems to evaluate social differences and different work ethics that influence income. It’s questionable though if there are systematic differences. Being of a certain confession doesn’t say as much about someone’s personality in modern days than it did in the past. Religiosity, using variables from the ISSP, might be a better control variable, but one that would vastly reduce the number of observations in the regressions, since only a subsample of the ALLBUS participated in the ISSP Religion.
Marital Status is measured via mstat (marital status) excluding categories with few cases, retaining the categories 0 ”Married”, 1 ”Widowed”, 2 ”Divorced” and 3 ”Single”. For men marriage is associated with a wage premium (Stratton, 2002).8 Moreover different partnership statuses are different life situations that influence attitudes and perceptions.
Political Attitude is measured via pa01 (self-rating left/right) which is a scale with 10 entries from left to right political attitude. The first five entries are combined into category 0 ”Left” and the last five entries into category 1 ”Right”. Political attitudes go hand in hand with world views and normative notions, which include perceptions about ASD. An effect on income might be distinct career choices, based on political attitudes, that differ in income.
An overview of the control variables is shown in figure 4. The causal graph used to test the back-door criterion via DAGitty9 is displayed in figure 5.
5 Method
Although the dependent variable Acceptance of Social Differences can be considered ordinal with ordered categories, no ordinal logistic regression is employed, since the Parallel Regression Assumption (PRA) would be dangerously close to being violated.10 An inconspicuous change of independent variables, using squared age instead of age groups, would even reject the PRA. A Brant test reveals this violation being, among others, due to the variable federal state, whose exclusion would be detrimental from a causal point of view. Violation of PRA invalidates interpreting the ordinal model as the logit/log-odds of being less than or equal to category m compared to being higher than category m, since the probabilities of adjacent categories cannot be accumulated anymore (Long and Freese, 2014: 327f.). Alternatively multinomial logistic regression (MLR) is employed, which does not consider the ranking of categories, but uses a reference category instead with which every other category is compared. As a drawback information of ranking is lost and fitting is less efficient.
MLR is a generalization of binary logistic regression. It splits the categories of the dependent variable into all possible pairs of categories and fits separate binary logits for each pair.11 The MLR model can be formally written as (Long and Freese, 2014:
Illustrations are not included in the reading sample
Conditional probabilities can be computed via:
Illustrations are not included in the reading sample
The conditional probabilities of each category m of Y are then divided by the conditional probability of the reference category b to obtain the relative odds and lastly logarithmized to obtain the logits, which depict the differences of the beta-coefficients between category m and the reference category b. Maximum likelihood estimation numerically approximates the most likely logits given the observed data. As no estimation is performed with less than 2500 observations, the results of the ML estimator here can be considered consistent, asymptotically normal and efficient (Long, 1997: 54).
Average marginal effects (AMEs) are used to interpret the results. The AME denotes the average discrete change in conditional probability of the dependent variable if one independent variable is manipulated, while holding all other independent variables constant. AMEs offer a range of advantageous properties as they are robust (Bergtold et al., 2018), intuitive and don’t suffer from unobserved heterogeneity like logits (Mood, 2009). All computations are performed using Stata.12
6 Analysis
6.1 Descriptive Statistics
Descriptive statistics are shown in table 1. Noteworthy is the low number of individuals accepting social differences. Only 24.11% are highly or very highly accepting of social differences, indicating that social differences are seen rather critically in Germany.13
6.2 Results Hypothesis 1
H1: The higher the net income of an individual, the more likely this individual will accept social differences.
The frequency distribution of ASD across the three income groups indicates support for H1 (see figure 6). The low income group has the highest percentage of individuals completely refuting social differences (31.6%) and the lowest percentage of individuals of very high acceptance (3.7%). The high income group on the other hand manifests the highest percentage of individuals with very high ASD (5.6%) and the lowest percentage of individuals completely refuting social differences (20%). The medium income group’s acceptance rates are positioned between those two poles. 21.8% of the individuals in the low income group are either highly or very highly accepting of social differences, compared to 24.7% of the medium income group and 30.4% of the high income group. This percentages suggest a trend of higher income increasing ASD.
A raw regression of ASD on income corroborates this suggestion (see table 2). The occurrence probability of completely refuting social differences (No Acceptance) is 11.6% lower in the high income group compared to the low income group (p < 0.0001). In the medium income group this occurrence probability is 4.2% lower compared to the low income group (p < 0.05), further corroborating a trend of increasing acceptance the higher the income. Furthermore having a high income increases the probability of highly accepting social differences by 6,7% (p < 0.005) and the probability of very highly accepting social differences by 2% compared to those having a low income. The last margin is not significant though (p = 0.119), possibly due to the overall low number of individuals that actually are very highly accepting of social differences (131).
The full regression model paints another picture (see table 3). Including all control variables leaves income with no significant effect at a 95% significance level. Being on average 5.8% more probable to be highly accepting of social differences when drawing a high income compared to a low income, ceteris paribus, is the only effect being at least significant at a 90% significance level. Figure 8 illustrates that there seems to be a slight trend of higher ASD with increasing income, mostly visible in the high acceptance group (green). But contradictorily rejection of social differences seems also to decline slightly with higher income (blue). Together with high variances, especially in the high income group, this explains the insignificant test statistics. In the light of these results Hypothesis 1 is rejected. Higher personal net income does not make it more likely to accept social differences.
6.3 Results Hypothesis 2
H2: The effect of personal net income on accepting social differences is at least partly mediated by relative deprivation.
Figure 7 reveals that the distribution of ASD across the relative deprivation groups, albeit similar to the distribution across the income groups, is generally more distinct with some particular differences. In the under-rewarded group only 17.9% of individuals are highly or very highly accepting of social differences, compared to the already low percentage of 21.8% in the low income group, promising to be a better predictor of refuting social differences. The high acceptance counterpart of the high income group appears to be the justly rewarded one. Naively one could assume that the groups of income and relative deprivation correspond. That those with low income feel underrewarded, those with medium income justly rewarded, and those with high income over-rewarded. Statistically that makes sense. Cognitively not so much. Even if there are some matches, given a correlation of 0.26, it doesn’t surprise that you can’t just fit the groups on each other, as demonstrated by the differing marginal frequencies of their contingency table (see table 4). There are way more individuals who feel under-rewarded than individuals with low income and equally less individuals that feel over-rewarded than individuals with high income. Hence relative deprivation definitely adds something new to the table that income couldn’t.
The raw regression would again support H2 (see table 5). Adding relative deprivation leaves income with none but one significant effect. On average, having a high income makes one still 6.7% less likely to completely refute social differences, rather than 11,6%, compared to having a low income, ceteris paribus (p < 0.01). Relative deprivation by contrast appears to be a powerful predictor, especially when it comes to those underrewarded. On average, feeling under-rewarded increases the probability of completely refuting social differences by 14% compared to individuals feeling justly rewarded, ceteris paribus (p < 0.0001).
Yet we know from the first model that income has no effect on ASD, so relative deprivation can’t function as a mechanism for income. The first full regression model supplemented by relative deprivation indeed reiterates that income has no significant effect on ASD (see table 6). This leaves us refuting H2 as well. Yet the second model further corroborates the prediction power of relative deprivation, showing highly significant and sizable effects on ASD, especially in the under-rewarded group. So the implicit sense of H2, namely that relative deprivation as a subjective factor is a more useful predictor than income, holds.
6.4 Results Hypothesis 3
H3: Individuals who earn a high income and expect a change of economic situation in the future are more likely to accept social differences than individuals who earn a high income, but don’t expect any change in their economic situation.
Table 7 displays the cross frequency table of Income and Expectation of Future Economic Situation (EFES) subdivided by the four ASD groups. What might catch the eye is that in the Very High Acceptance group there is not a single individual who has a low income and expects a negative change in her economic situation. In the No Acceptance group though the percentage of individuals who have a low income and expect a negative change amounts to 4,4% (40/917). A percentage that is gradually decreasing the higher the acceptance of social differences becomes. This hints at an interaction between income and EFES. Supportive of the assumed interaction in H3 specifically is the observation that the percentage of individuals who have a high income and expect a change in their economic situation increases from acceptance group to acceptance group. Only the Very High Acceptance group breaks this pattern.14
Complementing the first full regression model by the interaction term affirms parts of this observation (see table 8).15 On average, individuals who earn a high income and expect a positive change of their economic situation have a 11,8% higher probability to be highly accepting of social differences compared to individuals who earn a high income but don’t expect any change in their economic situation, ceteris paribus (p < 0.05). This does not hold true for individuals with high income who expect a negative change. For this interaction group there is not a single significant effect, also likely owing to small case numbers in the sample (24). Speaking against H3 is that individuals who earn a high income and expect a positive change have also a 3,5% lower probability to be very highly accepting of social differences. But this result is vulnerable due to low case numbers for individuals with both high income and very high ASD (26).
While caution is advised due to low case numbers in some interaction groups, the results partly affirm H3. On the one hand the Self-Serving Bias Theory is supported. Having an already high income and the prospect of even higher income increases ASD. On the other hand the Status Threat hypothesis is not supported. High income and the prospect of less income does not manifest any conclusive or significant effects. But expecting a worse economic situation isn’t necessarily connected to status threat. One might still expect to earn a high income in the future even if it is expected to be less than today, but not that much less to threaten status.16 In fact in the low income group there is support for status threat, since individuals who earn a low income and expect a worse economic situation have on average a 15,8% higher probability to completely refute social differences, compared to individuals who earn a low income and do not expect any change, ceteris paribus (p < 0.05). Although in the case of already low income it be might more sensible to speak of existential threat rather than status threat.
6.5 Model Diagnosis
A central assumption of MLR is Independence of Irrelevant Alternatives (IIA). This assumption purports that the odds to choose a category do not depend on other categories, implying that adding or deleting a category leaves the odds of other categories untouched (Long and Freese, 2014: 407). A Hausman-McFadden test indicates that this assumption is met in model 1.17 This is curious, since deleting the category Very High Acceptance would likely mean that individuals who located themselves in there would alternatively locate themselves in the High Acceptance category instead of dispersing equally among the remaining categories or refusing to choose, which would maintain the original odds.18 Yet it is argued that this two categories are no substitute for each other, but distinct and theoretically valuable to keep separated. Criteria advised to be giving the most weight (Long and Freese, 2014: 408). A Wald test for combining alternatives supports this as the two categories are distinct on a 95% significance level. The most relevant argument is though that with four categories effects in the interaction model are disclosed that are obliterated when High and Very High Acceptance are lumped together.
The correlation matrix between income and the other independent variables provides few indications for multicollinearity. Income correlates mildly with education (0.27) and gender (-0,32), but not worryingly (see table 9).
As Pseudo-R measures are unclear when it comes to their interpretation and significance (Long and Freese, 2014: 221f.), the Bayes Information Criteria (BIC) is used to compare models.19 The BIC of model 2 is 7,341 points lower compared to model 1, supporting the influence of relative deprivation on ASD. The opposite is true when comparing model 1 and model 3. The BIC of model 3 is almost 110 points higher, but this is expected as an interaction introduces two new independent variables and BIC strongly penalizes additional variables, especially if they do not add an exclusively strong effect on the dependent variable.
A step-wise build-up of model 1 reveals that the strongest confounders of income are gender and federal state. Otherwise the model is robust to changes (see table 10). In terms of BIC a MLR model regressing ASD only on income, gender and federal state is preferred to model 1. That is expected given the difference in number of regressors, but the Aikake Information Criterion (AIC), which does not penalize additional regressors as strongly, is only 68 points lower in the full regression model. When regressing ASD alone on income that difference in AIC is double as high compared to model 1, testifying to the strong influence of gender and federal state. In fact BIC-wise a MLR model regressing ASD only on gender and federal state is even preferred to a MLR model that regresses ASD on income, gender and federal state.
Comparing the MLR model 1 to the OLR model using the same variables manifests homologous results (see table 11). As in the MLR model in OLR income is insignificant, while significant effects of other independent variables are almost identical between the two models. This reassures that abandoning OLR in favor of MLR was no critical decision with vastly different outcomes, further corroborating the found results.
7 Summary and Discussion
In the introduction it was proposed that subjective factors are more suited to explain ASD than objective factors. In terms of income the results are in line with this proposition. Surprisingly, when adjusted for, income does not have an effect on ASD at all. Especially gender and residence in a new or old federal state seem to account for income’s effect. The subjective factor relative deprivation on the other hand, feeling justly or unjustly rewarded in comparison with others, appears to be a strong predictor of ASD. However, since relative deprivation showed to be no mediator of income, no causal effect can be asserted as the used model was causally specified only for income. When differentiated according to EFES, income again has an effect. In line with the third hypothesis individuals with high income and expectations of an even better economic situation are more likely to accept social differences than individuals with high income but no expected change. There is no effect regarding individuals with high income and expectations of a worse economic situation as hypothesized by status threat of Social Identity Theory, probably as not every worse economic situation for high income individuals is necessarily a status threatening one. Harsh economic changes that do threat status are likely an exception and even more rarely expected.
A weakness of the study is that the concept of social differences relies on one item, which different test persons might have interpreted differently. Combining various items that inquire attitudes, perceptions and actions would be a more robust approach. Elsewhere it has been shown that the GDP of a country has a particularly strong influence on ASD (Lang, 2017). Dividing federal states into groups, depending on their share of Germany’s GDP per capita, and controlling for those groups might have been an alternative approach to account for that.20 But as the effect of GDP is expected to be linearly positive on income and ASD, adding this confounder would not have changed the overall effect of income anyway.
Altogether the results suggest the complexity of perceptions and ASD, which makes it difficult to directly predict them from static, absolute factors. Perceptions are formed in relation to other perceptions which are influenced by various objective, environmental factors. This rules out a simple political ‘remedy’. A logical implication is to dig deeper into the causality of ASD. How do gender and culture influence perceptions that lead to either acceptance or rejection of social differences? Do individualistic cultures incline individuals towards demanding more for themselves, making it more likely to feel relatively deprived? How does a change in working habits, increasing free time and entertainment offers affect this acceptance? If work time is increasingly seen as time that could have been used for something else more valued and therefore feels more costly, would this increase perceptions of relative deprivation and the probability of rejecting social differences? How does media consumption affect ASD, e.g. a media climate that turns more and more attention to social inequality? There waits a plethora of questions to be approached.
References
Anderson, T., Forth, J., Metcalf, H., & Kirby, S. (2001). The Gender Pay Gap. Final Report to the DfEE.
Anger, C., & Geis, W. (2017). Bildungsstand, Bildungsmobilitat und Einkommen. Neue Herausforderungen durch die Zuwanderung. IW Trends, 44 (1), 43-58. https: //doi.org/10.2373/1864-810X.17-01-03
Bergtold, J. S., Yeager, E. A., & Featherstone, A. M. (2018). Inferences from Logistic Regression Models in the Presence of Small Samples, Rare Events, Nonlinearity, and Multicollinearity with Observational Data. Journal of Applied Statistics, 45 (3), 528-546. https://doi.org/10.1080/02664763.2017.1282441
Binelli, C., Loveless, M., & Whitefield, S. (2015). What Is Social Inequality and Why Does it Matter? Evidence from Central and Eastern Europe. World Development, 70, 239-248. https://doi.org/10.1016/j.worlddev.2015.02.007
Bosch, G., & Kalina, T. (2016). The Erosion of the German Middle Class. Intereconomics, 51 (2), 73-77. https://doi.org/10.1007/s10272-016-0580-4
Carter, M. (2014). Gender Socialization and Identity Theory. Social Sciences, 3 (2), 242-263. https://doi.org/10.3390/socsci3020242
Cheng, G. H.-L., Chan, D. K.-S., & Yeung, D. Y. (2019). Correlates of Acceptance of Wealth Inequality: A Moderated Mediation Model. Frontiers in Psychology, 10. https://doi.org/10.3389/fpsyg.2019.00900
Fehr, D., Muller, D., & Preuss, M. (2020). Social Mobility Perceptions and Inequality Acceptance. Working Papers 2020-02, Faculty of Economics and Statistics, University of Innsbruck.
Fischer, R., Ferreira, M. C., Jiang, D.-Y., Cheng, B.-S., Achoui, M. M., Wong, C. C., Baris, G., Mendoza, S., van Meurs, N., Achmadi, D., Hassan, A., Zeytinoglu, G., Dalyan, F., Harb, C., Darwish, D. D., & Assmar, E. M. (2011). Are Perceptions of Organizational Justice Universal? An Exploration of Measurement Invariance Across Thirteen Cultures. Social Justice Research, 24 (4), 297-313. https://doi. org/10.1007/s11211-011-0142-7
Flavin, M. A. (1981). The Adjustment of Consumption to Changing Expectations About Future Income. Journal of Political Economy, 89 (5), 974-1009. https: //doi.org/10.1086/261016
Gijsberts, M. (2002). The Legitimation of Income Inequality in State-socialist and Market Societies. Acta Sociologica, 45 (4), 269-285. https://doi.org/10.1177/ 000169930204500402
Hadler, M. (2005). Why Do People Accept Different Income Ratios? Acta Sociologica, 48 (2), 131-154. https://doi.org/10.1177/0001699305053768
Hegtvedt, K. A., & Parris, C. L. (2014). Emotions in Justice Processes. In J. E. Stets & J. H. Turner (Eds.), Handbook of the Sociology of Emotions: Volume II (pp. 103125). Dordrecht, Heidelberg, New York, London: Springer. https://doi.org/10. 1007/978-94-017-9130-4_6
Heider, F. (1958). The Psychology of Interpersonal Relations. New York: John Wiley & Sons. https://doi.org/10.1037/10628-000
Janmaat, J. G. (2013). Subjective Inequality: A Review of International Comparative Studies on People’s Views about Inequality. European Journal of Sociology, 54 (3), 357-389. https://doi.org/10.1017/S0003975613000209
Kuhn, A. (2019). The Individual (Mis-)Perception of Wage Inequality: Measurement, Correlates and Implications. Empirical Economics. https://doi.org/10. 1007 / s00181-019-01722-4
Lang, G. (2017). Zur Akzeptanz sozialer Ungleichheit. Wiesbaden: Springer VS. https: //doi.org/10.1007/978-3-658- 14995-6
Lenhart, O. (2018). The Effects of Income on Health: New Evidence from the Earned Income Tax Credit. Review of Economics of the Household, 17 (2), 377-410. https://doi.org/10.1007/s11150-018-9429-x
Long, J. S. (1997). Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks, California: Sage Publications.
Long, J. S., & Freese, J. (2014). Regression Models for Categorical Dependent Variables Using Stata (3rd ed.). College Station, Texas: Stata Press.
Luhmann, N. (1991). Soziale Systeme. Grundriß einer allgemeinen Theorie (4th ed.). Frankfurt am Main: Suhrkamp.
Mood, C. (2009). Logistic Regression: Why We Cannot Do What We Think We Can Do, and What We Can Do About It. European Sociological Review, 26 (1), 6782. https://doi.org/10.1093/esr/jcp006
Pearl, J., & Mackenzie, D. (2018). The Book of Why. The New Science of Cause and Effect. New York: Basic Books.
Peschl, M. F. (1990). Cognitive Modelling. Ein Beitrag zur Cognitive Science aus der Perspektive des Konstruktivismus und des Konnektionismus. Wiesbaden: Springer SV. https://doi.org/10.1007/978-3-663- 14658-2
Runciman, W. (1966). Relative Deprivation and Social Justice. London: Routledge.
Savani, K., & Rattan, A. (2012). A Choice Mind-Set Increases the Acceptance and Maintenance of Wealth Inequality. Psychological Science, 23 (7), 796-804. https: //doi.org/10.1177/0956797611434540
Shepperd, J., Malone, W., & Sweeny, K. (2008). Exploring Causes of the Self-serving Bias. Social and Personality Psychology Compass, 2 (2), 895-908. https://doi. org/10.1111/j.1751-9004.2008.00078.x
Snyder, M. L., Stephan, W. G., & Rosenfield, D. (1976). Egotism and Attribution. Journal of Personality and Social Psychology, 33 (4), 435-441. https://doi.org/ 10.1037/0022-3514.33.4.435
Stratton, L. S. (2002). Examining the Wage Differential for Married and Cohabiting Men. Economic Inquiry, 40 (2), 199-212. https://doi.org/10.1093/ei/40.2.199
Szirmai, A. (1986). Inequality Observed: A Study of Attitudes Towards Income Inequality. Enschede: Febodruk.
Tajfel, H., & Turner, J. (1979). An Integrative Theory of Intergroup Conflict. In W. G. Austin & S. Worchel (Eds.), The Social Psychology of Intergroup Relations (pp. 33-47). Monterey, CA: Brooks/Cole.
Tetlock, P. E., & Levi, A. (1982). Attribution Bias: On the Inconclusiveness of the Cognition-Motivation Debate. Journal of Experimental Social Psychology, 18 (1), 68-88. https://doi.org/10.1016/0022-1031(82)90082-8
Yitzhaki, S. (1982). Relative Deprivation and Economic Welfare. European Economic Review, 17 (1), 99-113. https://doi.org/10.1016/0014-2921(82)90045-9
Appendix: Tables and Figures
Table 1: Descriptive Statistics of all used Variables (unweighted)
Illustrations are not included in the reading sample
Table 2: Average Marginal Effects of Income on Acceptance of Social Differences
Illustrations are not included in the reading sample
Table 3: Average Marginal Effects of Income on Acceptance of Social Differences with Control Variables
Illustrations are not included in the reading sample
Table 4: Cross Table of Income and Relative Deprivation (unweighted)
Illustrations are not included in the reading sample
Table 5: Average Marginal Effects of Income and Relative Deprivation on Acceptance of Social Differences
Illustrations are not included in the reading sample
Table 6: Average Marginal Effects of Income and Relative Deprivation on Acceptance of Social Differences with Control Variables (not shown)
Illustrations are not included in the reading sample
Table 7: Cross Table of Income and Expectation of Future Economic Situation by Acceptance of Social Differences (unweighted)
Illustrations are not included in the reading sample
Table 8: Average Marginal Effects of Income and Expectation of Future Economic Situation (Interaction) on Acceptance of Social Differences with Control Variables (not shown)
Illustrations are not included in the reading sample
Table 9: Correlation Matrix of Independent Variables
Illustrations are not included in the reading sample
Table 10: Average Marginal Effects with Step-Wise Build-Up of Multinomial Logistic Regression Model
Illustrations are not included in the reading sample
Table 11: Comparisons of Average Marginal Effects between Multinomial and Ordinal Regression
Figure 1: Simple Hypothesis H1
Illustrations are not included in the reading sample
Figure 2: Mechanism Hypothesis H2
Illustrations are not included in the reading sample
Figure 3: Interaction Hypothesis H3
Illustrations are not included in the reading sample
Figure 4: Control Variables
Illustrations are not included in the reading sample
Figure 5: Control Variables with assumed Causal Paths
Illustrations are not included in the reading sample
Illustrations are not included in the reading sample
Figure 8: Marginsplot of Accepting Social Differences given Income
Illustrations are not included in the reading sample
[...]
1 A concept that is still not well defined and often used as a “catch-all concept” (Binelli et al., 2015: 239) for various forms of inequality regarding income, wealth, status, health etc.
2 For a overview see Janmaat, 2013.
3 https://www.gesis.org/en/allbus/contents-search/study-profiles-1980-to-2018/2018
4 Imaginable is the exclusion of individuals who distrust organizations. Those individuals will likely deny cooperation in the survey and might see social differences more negatively, maybe even because they consider inequality a consequence of the actions of organizations.
5 wghtpew is a weighting variable given in the ALLBUS data set, which adjusts the proportion. In Stata every computation will be appended with [pweight=wghtpew] to apply this weighting. Unfortunately this prevents the use of certain commands.
6 Im21 & im19: 0,35. im21 & im20: 0,51. im20 & im19: 0,52. The reason might be different framing. Im19 shifts the attention towards positive consequences of social differences, namely higher motivation. The following items shift away from this positive perspective, accordingly disacceptance grows.
7 The change in occurrence probability of being in a certain acceptance group when increasing income by one Euro would be negligibly small and rather uninformative.
8 Mechanisms might be higher productivity and aspirations when married as well as a better position when negotiating wages.
9 http://www.dagitty.net/
10 PRA assumes that beta-coefficients are equal for all categories m of the dependent variable Y, implying that the effect of an independent variable X on Y is equal for every category m.
11 MLR theoretically computes N 2 binary regressions, N being the number of categories of the dependent variable. Since these regressions are linearly dependent and partly redundant, only a minimal set of N-1 binary regressions is computed, comparing N-1 categories to one reference category.
12 https://www.stata.com/
13 Unless distorted by social desirability.
14 No Acceptance: (17+3)/90 = 22%. Low Acceptance: (39+14)/211 = 25,1%. High Acceptance: (32+6)/109 = 34,9%. Very High Acceptance: (4+1)/26 = 19,2%. These are the unweighted numbers to keep the frequencies natural numbers. Weighting the cases according to the actual proportion of populations in the new and old federal states gives the Very High Acceptance group a share of 30% with high income and expected change. Furthermore showing the influence of living either in a new or old federal state.
15 Relative deprivation is not included in this model since it is now considered a variable with an independent effect on ASD and not a mediator or control variable of income.
16 This could have been accounted for by keeping all 5 categories that ep06 offers. Yet there would have been no meaningful results, since only 3 individuals with high income and expectation of a considerably worse economic situation were included in the regression.
17 Mind that IIA tests in Stata only work on unweighted data, meaning not the weighted distribution used in the regressions is tested, but the sample distribution.
18 This might generally be the case when an ordinal dependent variable is used, since deleting one category should make it more likely to chose a remaining neighbor than choosing randomly among all remaining categories. Advise to resort to MLR, if assumptions of ordinal logistic regression are not met, is therefore questionable. Other ordinal models might be better suited, like the generalized ordered logit model.
19 In the following ‘model 1’ references the full regression model of H1, ‘model 2’ the full regression model of H2 (relative deprivation added to model 1) and ‘model 3’ the full regression model of H3 (interaction added to model 1).
20 At the expense of perfect collinearity as federal states would have been included twice. Since the division of east and west federal states has mostly a cultural effect different from the GDP effect, deleting this control variable is unwise. Some information of the GDP is also coded in the east and west dummy as the share in the GDP per capita is on average higher in the west federal states. Controlling for west and east federal states as done here might therefore be the best compromise.
- Quote paper
- Marco Hauptmann (Author), 2020, The Effect of Income on Acceptance of Social Difference. The Importance of Subjective Factors, Munich, GRIN Verlag, https://www.grin.com/document/1597582