First supervisor:
Faculty of Economics and Business Chair of Global Economics and Management
Assoc. Prof. Dr. Robbert Maseland
Second supervisor: Faculty of Economic Sciences
Chair of Economic Policy and SME Research Dr. Ann-Kathrin Blankenberg
Gender Income Inequality and Happiness –
An Empirical Analysis
Thesis prepared for the degree of
MASTER OF ARTS & MASTER OF SCIENCE
Marieke Cornelia Baaken Submitted: June 2019
J.C. Kapteynlaan 47B Place of Birth: Münster 9714CN Groningen
II TABLE OF CONTENTS
Table of Figures ... III Abstract... ... IV
1 Introduction ... 1
2 Literature Review ... 3
2.1 Happiness ... 3
2.2 Gender Income Inequality ... 4
2.3 Income Inequality and Happiness... 5
2.4 Gender Income Inequality and Happiness ... 8
3 Method and Data ... 10
3.1 Dependent Variables... 11
3.2 Independent Variables ... 12
4 Empirical Results ... 14
4.1 Intra-Household Income Inequality Regression ... 14
4.1.1 Inequality Measure ... 15
4.1.2 Multicollinearity ... 17
4.1.3 Robustness Tests ... 19
4.1.4 Regression Results and Discussion ... 20
4.2 Intra-Occupational Income Inequality Regression ... 25
4.2.1 Multicollinearity ... 26
4.2.2 Robustness Tests ... 27
4.2.3 Regression Results and Discussion ... 27
4.3 Biases and other Limitations ... 30
5 Conclusion ... 31 5.1 Summary ... 31 5.2 Limitations ... 32 5.3 Policy Implications ... 33 5.4 Future Research ... 34 6 Bibliography ... 36 7 Appendix ... 41
III TABLE OF FIGURES
Figure 1: Happiness Score and GDP per Capita, 2015………6
Figure 2: Self-reported Happiness of Females in Germany………..11
Figure 3: Self-reported Happiness of Males in Germany………..………..12
Figure 4: Fixed Effects Regression with Inequality Ratio and Inequality Gap………..16
Figure 5: Scatterplot of Household Income and Income Gap/Income Ratio……….17
Figure 6: VIF Scores in the Female Happiness Regression………...18
Figure 7: Regression Results Intra-Household Regression………...20
Figure 8: Change of Average Male Happiness Depending on Women’s Wage………24
Figure 9: Conducted Tests and Diagnostics (Intra-Occupational Regression)………25
Figure 10: Regression Results Intra-Occupational Regression……….28
Figure 11: Policy Recommendations and Implications Matrix……….33
Figure A1: Important Variables within Regressions………...………..41
Figure A2: Descriptive Statistics of Variables Used for the Analysis………..43
Figure A3: Pearson Product-Moment Correlation Matrix including Female Happiness (Intra-Household Regression)……….45
Figure A4: Pearson Product-Moment Correlation Matrix including Male Happiness (Intra-Household Regression)……….46
Figure A5: Robustness Test (Within-Household Regression)………..47
Figure A6: Robustness Test with Income Satisfaction (Intra-Household Regression)……….48
Figure A7: Regression Including a Binary mfratio Variable (Intra-Household Regression)…49 Figure A8: Regression Including Inequality Measures (Intra-Occupational Regression)…….50
Figure A9: VIF Scores Intra-Occupational Female Happiness Regression……….………….51
Figure A10: Pearson Product-Moment Correlation Matrix including Female Happiness (Intra-Occupational Regression)………..52
Figure A11: Pearson Product-Moment Correlation Matrix including Male Happiness (Intra-Occupational Regression)………..53
Figure A12: Robustness Test Intra-Occupational Regressions……….54
Figure A13: Reverse Causality Test with Lagged Variables………55
Figure A14: Sensitivity Results for Variables of Interest (Dependent Variable: female happi-ness)………..56
IV Abstract
Gender income inequality in Germany is still widespread in 2019 but how does this affect the happiness of men and women? The thesis provides insights on the relationship between gender income inequality and happiness in different spheres of life; within households and within oc-cupations. While previous studies have focused primarily on macro-level relationships, this study focuses on micro-level relationships and distinguishes the inequality effect on men and women. This thesis challenges the proposition that gender income inequality leads to gender happiness gaps by conducting a multiple linear regression analysis using large-scale longitudi-nal data of private household provided by the German-Socio Economic Panel (SOEP). After employing fixed effects regressions, results of the intra-household regressions support that fe-male happiness decreases if gender income inequality increases, whereas the relationship be-tween gender income inequality and males’ happiness is inverted u-shaped. No support for a relationship between gender income inequality and happiness can be found for the intra-occu-pational regressions. The findings suggest that the relationship between gender income inequal-ity and happiness depends on the context and comparabilinequal-ity of wages between individuals.
Keywords: Gender Income Inequality; Happiness; Gender Equality
Note: For simplicity's sake, the female form is used in the entire text; naturally, the male form
1 1 Introduction
„While happiness itself is sought for its own sake, every other goal – health, beauty, money, or power – is valued only because we expect that it will make us happy.“ - Mihaly Csikszentmihalyi, 20131
Happiness has gained a lot of attention in research during the last decades, as subjective factors are increasingly deemed as important determinants of well-being and life satisfaction (Piper, 2019). Happiness can be defined as the “degree to which people positively assess their life situation” (Katsaiti, 2010).
Different causes of people’s happiness shifts and factors contributing to happiness are being investigated within economics to assess how policy actions can be improved and to add a change of perspective to economic theories (e.g. Stevenson & Wolfers, 2009; Maseland & van Hoorn, 2009). Furthermore, economists have started to analyze whether happiness can be an appropriate proxy for individual utility. For these purposes, scholars frequently use data bases of nationally representative surveys such as the German Socio-Economic Panel (SOEP) located at the German Institute of Economic Research. Research results indicate that happy individuals tend to have a better health and are rarely absent at work, meaning that happiness is a good approximation of individual utility. Therefore, happiness is proposed to explain relationships between well-being and other factors such as income or health (Ferrer-i-Carbonell & Gowdy, 2007). Accordingly, happiness has emerged as a dimension to meas-ure societal progress. The Organization for Economic Cooperation and Development (OECD) has introduced life satisfaction as a dimension for measuring development of coun-tries in 2011. As a response, governments are increasingly showing an interest in results of happiness research in order to shift the policy focus and to analyze which policies should be implemented to improve people’s happiness. Important economic reasons for people’s hap-piness to shift are income inequality and therefore gender income inequality (Oishi et al., 2011).
Gender income inequality is a crucial topic to tackle worldwide, as it is a “chronic socioec-onomic malice” leading to far-reaching disastrous societal and ecsocioec-onomic consequences (Poddar & Mukhopadhyay, 2019). Patel et al. (2018) for example found that populations which have a higher income inequality have a greater risk of suffering from depression, the effect especially affects women. In line with these results Oshio and Kobayashi (2010) found
2
that people tend to report to have a lower level of health in areas where high income inequal-ity exists.
Germany faces a persistent gender inequality problem, as compared to other OECD coun-tries, the country ranks below average in gender equality. The main reasons for the inequality are the gender wage gap which has been constant for years and the unequal distribution of childcare (Eurostat, 2019).
Discrepancies between studies are found in the relationship between happiness and income inequality. Models which include inequity aversion2 find a negative relationship between
happiness and inequality, while other empirical studies on happiness find a positive relation-ship (Hopkins, 2008). One of the reasons for these ambiguous results is that research has largely focused on general, macro-level relationships. However, common sense dictates that effects are likely to be different for males and females. Moreover, the happiness conse-quences of inequality may differ between different notions of inequality and between differ-ent spheres of life.
This thesis addresses these challenges. In order to get a better understanding of the relation-ship between inequality and happiness, an empirical analysis of a national survey in Ger-many will be conducted. The survey data is downloaded from the SOEP, a “wide-ranging representative longitudinal study of private households” (SOEP, 2019). In more detail, the effect of gender income inequality within households and within occupations on female and male happiness will be investigated to answer the research question “Does gender income inequality lead to gender happiness gaps?”. This study aims to contribute to the literature by (1) taking different notions of gender income inequality, (2) investigating the effect of ine-quality both, on male and on female happiness separately and (3) analyzing the ineine-quality effect on the household and the occupational level.
In order to answer the research question, firstly, a brief overview of the status quo in the field of happiness and gender income inequality in Germany as well as their current rele-vance will be given. Subsequent, these topics will be linked and existing literature about models of happiness and income inequality are portrayed. Following the chapter of happi-ness and income inequality, an extrapolation of the previous mentioned insights will be given about gender income inequality and happiness. This chapter leads to the development of four testable hypothesis.
The outline of the remainder of the study is as follows. For investigating the correlation and (causal) relationship between gender income inequality and happiness of men and woman in Germany, several multiple linear panel data regressions will be conducted. First, the method and data used within the analysis will be explained. Subsequent, statistical tests are
3
conducted to find out which regression model will fit best for the data and purpose and dif-ferent measures of inequality will be discussed. Moreover, robustness tests will be imple-mented. Finally, empirical results, limitations as well as biases of the regressions will be discussed. The study concludes by giving a brief summary of the results, examining remain-ing limitations of the study, statremain-ing policy implications and potential further research ideas.
2 Literature Review
In this literature review, the concepts of happiness and gender income inequality are dis-cussed. Several perspectives on these concepts as well as on their relationship will be given.
2.1 Happiness
During the last decades, many economists have turned their research focus towards happi-ness (Sacks et al., 2012), especially since 2003 papers containing the term “happihappi-ness” and synonyms for happiness have risen strongly (McKerron, 2012). This trend can be explained by the fact that happiness is considered as a life goal and helps researchers understand eco-nomic issues and dive deeper into coherences (Piper, 2019; Maseland & van Hoorn, 2009). In fact, it can lead to the accomplishment of further societal goals such as productivity and health (Happiness Research Institute, 2019). Additionally, the OECD and many other critics, state that GDP per capita as well as unemployment rates, which have been used for decades as the preferred measure to evaluate national wealth are not sufficient to give a comparable and comprehensive overview of well-being and thus people’s happiness. However, more than ever it is crucial to measure happiness on its own in order to enhance policies that may increase the quality of people’s lives (Happiness Research Institute, 2019).
Happiness is affected by biology, policies and behavior, whereby only the last two can change according to the Happiness Research Institute. The focus of this thesis lies on poli-cies, which can be modified over time, as these have the potential to increase people’s hap-piness in the long-run (Haphap-piness Research Institute, 2019). Moreover, people’s haphap-piness may also adapt over time. The Set Point Theory is the most widely known theory regarding happiness and is linked to phenomena of hedonic treadmill and adaptation. The theory de-scribes that an individual’s happiness depends on its previous experience, thus on a specific reference point, whereby the concepts of hedonic treadmill and adaptation refer to the ten-dency of individuals to return to a stable level of happiness after a positive or negative event. People constantly draw comparisons from the past with the future. Thus, aspiration levels matter and many events may only have a temporary effect on happiness. However, more recent studies found that certain events do lead to long-lasting increases in happiness (Men-carini & Sironi, 2010).
4
policies exists. Germany ranks 16 out of 157 in happiness levels in the latest update of the World Happiness Report published in 2015 (Helliwell et al., 2016).
Happiness can be observed by conducting large surveys (Frey & Stutzer, 2002), mainly us-ing Cantril’s ladders, where participants are asked to state their happiness measured on an eleven-point scale from zero to ten, where ten is the highest grade. The SOEP, which data is used in the thesis, as well applies this method and scale of measuring happiness.
2.2 Gender Income Inequality
Gender Income Inequality substantially exists in the developed world in 2019. In a report recently published by the World Bank Group (2019), Germany is ranked 31st among 187 countries in gender equality. The rather low score of Germany (91.88/100) compared to the average OECD country (93.54/100) is due to inequality in income and inequality in childcare distribution (World Bank Group, 2019). Reimer and Schröder (2006) support this finding by stating that women having a university degree face wage discrimination in Germany. It fur-thermore is supported by the fact that the gender wage gap in Germany accounts to 21 per-cent, meaning that women on average earn 21 percent less measured in gross hourly earnings compared to males. The German gender wage gap is one of the highest and most persistent wage gaps within the EU3 (European Parliament, 2015) and even the adjusted gender wage gap4 accounts to 6% in Germany in 2014. Moreover, the yearly published Global Gender Gap Report finds that the world rank of Germany decreased significantly in the last 10 years (World Economic Forum, 2018). Blau and Kahn (2017) as well as Arulampalam et al. (2007) as well studied developments of the gender wage gap and found that the gap remains espe-cially for high-income jobs. Accordingly, the current Allbright report of 2019 found that the gender distribution of power positions within companies in Germany is highly unequal to the disadvantage of women. Within the DAX30 in Germany, not a single chairman is female and only 2.5% of the chairmen within the 160 German listed companies are women (All-bright, 2019). Throughout German firms, women occupy about 25% of the top management positions (Deutsche Welle, 2017). Finally, scholars state that the gender wage gap does not seem to change in the upcoming years (Deutsche Welle, 2019a), which makes gender ine-quality in Germany a crucial issue to address, as it significantly decreases a country’s eco-nomic performance (Cingano, 2014).
Several direct and indirect explanations of the gender wage gap exist. One of the indirect explanations in developed countries are the gender choice differences in career, which are significantly influenced by gender stereotypes, already shaping the interests of young chil-dren. In a study on societal believes it was shown that some jobs are made for men and others for women (Akerlof & Kranton, 2010). Additionally, girls at the age of six begin to avoid
3 The average EU gender wage gap is 16.4%.
5
doing activities said to be for children who are very intelligent and believe that more boys fit into the category. This increases the gender pay gap and hence income inequality, as women more often avoid choosing academic disciplines which they think are for the very smart, such as mathematics (Bian et al., 2017). Furthermore, using the same inaccurate logic of blaming women for their career choices, the choice of having children is also used to explain the gender wage gap. As employers often naturally assume that women instead of men will stay at home to care for the children if they are sick (Hughes, 2015), which is partly true as childcare still remains mainly on the shoulders of women (Gershuny, 2000). DeCicco (2018) as well found that motherhood is a major key factor for explaining the gender wage gap.
An additional reason for the gender wage gap is direct wage discrimination. These are situ-ations where women workers receive lower wages than their male counterparts for work of equal value due to their gender. This is discrimination, as gender is not directly related to productivity levels (Grün, 2004). Pocock and Alexander (1999) interpret this phenomenon as an undervaluation of women’s work.
The consequences of a gender wage gap, regardless if provoked by direct or indirect dis-crimination, are far-reaching. It affects the concerned women’s families negatively by de-creasing social insurances and pensions. The gap as well affects society by inde-creasing the governments payments in social benefits and decreasing tax revenues. The gap additionally “[…] distorts competition between companies and can jeopardize social peace” (Schweizer-ische Eidgenossenschaft, 2019). Furthermore, Holter (2014) found that gender equality pos-itively correlates with health dimensions, which supports the fact that gender equality is cru-cial. Accordingly, contrary to the belief of many people that gender inequality being a women’s issue, gender equality does not only benefit women but also men within a society (de Looze et al., 2018).
2.3 Income Inequality and Happiness
This sub-chapter aims to connect the concepts from the previous paragraphs, namely income inequality and happiness.
6 Figure 1: Happiness Score and GDP per Capita, 2015
Source: Own research, updated Figure of Graham (2005) using World Bank (2019) & Helliwell (2016). In contrast to the study of Easterlin, several researchers found that no such satiation point exists (Stevenson and Wolfers, 2013). Deaton (2008) found that the Easterlin Paradox does not hold when using a larger sample of countries. Additionally, Sacks et al. (2010) investi-gate if raising the income of a country increases the citizen’s happiness over time and found a strong positive relationship between the two. The authors extent their prior research in 2012 by stating that absolute as well as relative income matters for the level of happiness. In order to sum this up, one can say that scholars have not yet concluded whether the rela-tionship between income and happiness is positive or negative. Various aspects such as the investigated country or the sample year alter the relationship.
The relationship between income inequality and happiness is also controversially discussed by many scholars. Hopkins (2008) for example distinguishes between three types of models analyzing the relationship, which all underlie the Model of Relative Concerns. In the follow-ing models, utility is used as a proxy for happiness. Firstly, Hopkins mentions the basic Model of Relative Concerns, whereby income is denoted by 𝑧𝑖, utility by 𝑈 and the income
of others by 𝑧𝑖−1. In this model, an increase of one’s own income will lead to an increase of utility, if the income of others remains unchanged. An increase of the own income and a simultaneous increase of the income of others can lead to different outcomes. If the other person is richer, utility will decrease, if the other person is poorer, the utility will increase, or the utility will decrease independent of the other person’s assets.
7
Subsequently, Hopkins writes about the Mean-Dependence Model. The utility of an individ-ual increases when the own income increases and when the income compared to the average income of others (𝑧̅) increases. This means that utility can increase when;
𝑈 (𝑧
𝑖, 𝑧
𝑖−1) = 𝑈 (𝑧
𝑖, 𝑧
𝑖− 𝑧̅) (2)
This model may potentially explain the above-mentioned Easterlin paradox, as happiness does not increase over time for those people whose income does not increase faster than the average income. This applies if U is linear, as the income of everyone increases at the same speed, thus the “average happiness will not rise” (Hopkins, 2008).The second model Hopkins mentions is the Model based on rank; the utility increases if the own income increases. Furthermore, happiness depends on the own rank 𝐹 in income. 𝐹 denotes the distribution of income. In this model, utility can increase if inequality increases.
𝑈 (𝑧
𝑖, 𝑧
𝑖−1) = 𝑈 (𝑧
𝑖, 𝐹(𝑧
𝑖)) (3)
This model may also potentially explain the Easterlin paradox, because the happiness of a single individual increases if her own income increases, as it enhances her rank. This means that happiness increases in cross-section. However, if the income of every individual within a society increases, their ranks remain constant and happiness would not increase despite a general increase in income.Finally, the Inequity Aversion Model is brought up by Hopkins. Utility is positively depend-ent on one’s income, but negatively on the difference between the own income and that of others. People dislike if others have more income however at the same time dislike if people have less income. An individual compares her own (𝑧𝑖) income with n other people’s in-comes (𝑧𝑖−1), whereby “𝛼 is a weight on the average of incomes that are above yours and 𝛽 is a weight on the average of incomes below yours” (Hopkins, 2008). It is assumed that 𝛼 ≥ 𝛽 and that 1 > 𝛽 ≥ 0.
𝑈 (𝑧𝑖
, 𝑧
𝑖−1) = 𝑧
𝑖−
𝛼𝑛−1
∑
𝑧𝑗>𝑧𝑖(𝑧
𝑗− 𝑧
𝑖) −
𝛽
𝑛−1
∑
𝑧𝑗<𝑧𝑖(𝑧
𝑖− 𝑧
𝑗)
(4)
Models containing inequity aversion5 assess a negative relationship between happiness and
inequality, as the utility decreases if inequality increases. Fehr and Schmidt (1999) for ex-ample found experimental evidence that individuals make decisions to minimize inequity in outcomes. These models may not explain the Easterlin paradox, as an increase in real income leads to an increase in happiness (Hopkins, 2008).
In summary, it can be said that Inequity Aversion Models assume that an increase in ine-quality leads to a decrease in happiness, whereas the Mean-Dependence Model and the Model based on rank both assume a positive relationship between inequality and happiness,
8
as an increased equality leads to more competition among individuals and thus lower utility “at each income level” (Hopkins, 2008).
2.4 Gender Income Inequality and Happiness
Several studies about the relationship between happiness and gender inequality have been conducted in the field of economics. Mencarini and Sironi (2010) found that gender differ-ences are significantly impacting individual’s happiness. The authors studied the effect of differences in labor division between men and women within a household on women’s hap-piness and found that an unequal distribution of housework towards women affects women’s happiness negatively. This is particularly intriguing, as domestic work as well as childcare are still extremely gender specific. Most of these studies have been conducted on the macro-level and found a positive relationship between gender equality and happiness (see Looze et al, 2017; Holter, 2014; Inglehart et al., 2002; Schyns, 1998; Qian, 2016). However, the stud-ies only control for gender and do not particularly investigate the effect of gender inequality on females’ and males’ happiness.
This thesis specifically aims at analyzing the relationship between gender income inequality and happiness on the micro-level, which is relatively unexplored in the literature and thus many aspects of the relationship are still neglected or require a deeper investigation. This thesis will therefore contribute to the literature by starting to close the gap and give a more sophisticated view of the relationship by studying the effect of gender income inequality on males’ and females’ happiness, within households and occupations.
Gender inequality has serious negative effects on a society. Moreover, the level is fairly high in Germany compared to the OECD countries and it does not tend to decrease in the upcom-ing years. These are reasons supportupcom-ing the fact that gender inequality in Germany is a cru-cial and current topic to deal with in research. The same applies to happiness, as happiness research helps politicians shift their policy priorities to improve societies’ productivity and health, and adequate room for improvement in Germany exists.
Gender income inequality6 is thus expected to be rather high and of significant importance
in Germany. Female happiness is expected to decrease with a rising gender income inequal-ity as the Mean-Dependence Model as well as the Model based on rank link the relative income to happiness, which is expected to be lower for females. Furthermore, male happi-ness should tend to increase with an increase in gender income inequality according to these models, as their relative income would increase. The Inequity Aversion models as well pre-dict that an increase in wage discrimination decreases female’s happiness, as utility de-creases when inequity inde-creases. Thus, happiness gaps of males and females should tend to increase when income inequality increases.
9
This hypothesis will be tested in two different spheres of life; within households and within occupations.
Firstly, intra-household income inequality will be investigated, as females can directly com-pare their own income to their spouse’s income. Luttmer (2005) as well as Ravina (2007) support this by finding that people particularly compare themselves to others which are ge-ographically close. Furthermore, happiness of females is expected to be more affected within households than within occupations because (1) childcare and domestic housework are still highly gender specific and (2) gender inequality is fairly high in Germany amongst others because of an unequal distribution of childcare. Additionally, Holter (2014) found that women’s happiness is positively influenced by within-family gender equality. Accordingly, the following hypothesis will be tested within the thesis:
Hypothesis 1a: Within households a higher gender income inequality leads to a de-crease in females’ happiness.
Hypothesis 1b: Within households a higher gender income inequality leads to an increase in males’ happiness.
Female happiness is still expected to decrease with an increase in gender income inequality within occupations, especially in high-income jobs, where males earn much more relative to their female counterparts. The negative relationship might occur because the geographic proximity effect can be overcome due to available communication technology advances (Dy-nan & Ravina, 2007).
Unlike gender income inequality within households, which is mostly related to general wage gaps, gender income differences within occupations are more related to the concept of the adjusted gender wage gap. However, elements of a general wage gap still exist as females are expected to have more part-time contracts than males and differences in education re-main. When gender income inequality endures, female happiness is expected to decrease, as their relative income decreases and male happiness is expected to increase because their relative income increases. Accordingly, happiness gaps will tend to increase as well and gender income inequality may lead to an increase in or existence of happiness gaps;
Hypothesis 2a: Within occupations higher gender income inequality leads to a de-crease in females’ happiness.
10 3 Method and Data
In order to study the relationship of gender income inequality and happiness, the thesis uses cross-sectional data from the SOEP which is located at the German Institute for Economic Research. The SOEP are nationally representative household and individual surveys, which are conducted every year since 1984. The data “is centered on the analysis of the life course with objective and subjective indicators of well-being” and “strives to lead the field interna-tionally in the quality, originality, significance, and rigor of its work” (Goebel et al., 2018). The data is collected via questionnaires by Kantar Public Deutschland at the household level with currently nearly 11.000 households per year and at the individual level with around 30.000 individuals per year. For the analysis, the individual questionnaires for all persons aged 17 and older will be used from the years 1984-2017 with about 650.000 observations in total. The first wave of the data has been collected in 1984 and the last wave available, which is of 2017, has been published recently in May 2019. Thus, this study will use all available 34 waves in order to have a meaningful number of observations despite taking numerous variables which are only available for some years, and thus be able to detect rea-sonable effects. Benefits of the SOEP data are among others a high degree of sample quality, comprehensibly prepared data, a high level of available sociodemographic variables and years as well as a reliable representation of the sampled population7. In order to test hypoth-esis 1a and 1b and therefore investigate the relationship between within household income inequality and happiness, the following multiple regression estimating equations are used:
𝐻
𝑖𝑡= 𝜌
0+ 𝜌
1𝑥
𝑖𝑡+ 𝜌
2𝑦
𝑖𝑡+ 𝜌
3Φ
𝑖𝑡+ 𝜀
𝑖𝑡(5)
𝐻
𝑖𝑡= 𝜌
0+ 𝜌
1𝑥
𝑖𝑡+ 𝜌
2𝑥
𝑖𝑡2+ 𝜌
3𝑦
𝑖𝑡+ 𝜌
4Φ
𝑖𝑡+ 𝜀
𝑖𝑡(6)
𝐻𝑖𝑡 coincides to the respondent i’s reported happiness in year t, 𝑥𝑖𝑡 corresponds to the ratio of the male spouse’s income divided by the female spouse’s income or the male spouse’s income subtracted by the female spouse’s income and represents the income inequality within a household. 𝑥𝑖𝑡2 is the squared term of the income inequality measure and 𝑦𝑖𝑡 corre-sponds to the log household income of both spouses, Φ𝑖𝑡is a vector of control variables and 𝜀𝑖𝑡 is the error term. As control variables the marital status, age, health and religion are being used.All households which did not consist of at least one female and one male spouse were ex-cluded. Furthermore, observations with missing values in the variables used in the analysis were deleted. This means that also answers of respondents were excluded, which answered that a monthly net income did not apply to them, whereby others wrote that they earned zero Euros. After the exclusion about 40,146 observations for the female regression and 43,890 observations for the male regression are left, covering the period from 1984 to 2017 with
11
gaps, as the panel data is unbalanced. At least 95% of the respondents have participated 24 times in the survey and at least 75% of the respondents have participated ten times in the survey.
In order to test hypothesis 2a and 2b and therefore investigate the relationship between intra-occupational income inequality and happiness, the following estimating equation will be used:
𝐻
𝑖𝑡= 𝜌
0+ 𝜌
1𝑥
𝑖𝑡+ 𝜌
2𝑦
𝑖𝑡+ 𝜌
3Φ
𝑖𝑡+ 𝜀
𝑖𝑡(6)
Within this equation 𝐻𝑖𝑡, Φ𝑖𝑡 and 𝜀𝑖𝑡 correspond to the same variables as in thewithin-household equation. However, 𝑥𝑖𝑡 corresponds to the ratio of the males’ income within an occupation divided by the females’ income within the same occupation or the log of the gap between the men’s income within an occupation subtracted by the women’s income within the same occupation. 𝑦𝑖𝑡 corresponds to the personal income. The control variables are the same as within the household regression, thus marital status, age, health and religion.
3.1 Dependent Variables
In both, the intra-household inequality regression as well as the intra-occupational inequality regression, the first dependent variable used in the analysis is general life satisfaction of females, thus happiness. Happiness is derived by responses to the question “How satisfied are you with your life, all things considered?”, on a 11-point scale. Zero means complete dissatisfaction and ten complete satisfaction, thus higher values correspond to higher levels of happiness. Figure 2 shows the approximately bell-shaped pooled frequencies of female responses which are slightly skewed to the right, whereby the mean value is 7.082.
Figure 2: Self-reported Happiness of Females in Germany
12
Furthermore, happiness of males will be used as the second dependent variable. In Figure 3, one can observe that the shape of the pooled frequencies of male responses is similar to the shape of the frequencies of female responses. However, the mean value is 7.084 and is there-fore slightly higher (see Figure A1 and A2 in the appendix for supplementary descriptive statistics, pp. 41 ff.).
Figure 3: Self-reported Happiness of Males in Germany
Source: Own research with pooled data from 34 waves (1984-2017) of the SOEP.
3.2 Independent Variables
The independent control variables marital status, health, income, religion and age will be used within all of the regression analysis conducted in the thesis, as they have been identified as significant determinants for self-reported happiness of individuals (Katsaiti, 2010; van Hoorn & Maseland, 2013). Additionally, Frey and Stutzer (2002) found that unemployment as well highly impacts self-reported happiness, however, the determinant is not being in-cluded in the analysis as the question was only asked in one year, and the number of obser-vations would get too small. The summary statistics are displayed in Figure A2 (see appen-dix, p. 43 ff.) and a more detailed description can be found in Figure A1 (see appenappen-dix, 41 ff.).
13
from the survey year and then taking the log, the youngest respondent is 17 and the oldest 104. The variable job is a dummy variable set to one, if the respondent is employed. Further-more, religion corresponds to an ordered variable depicting the faith of the respondent by asking how often respondents go to church or attend religious events, whereby four answers are possible; at least once a week, at least once a month, less often and never. Personal in-come is a variable depicting the log of the net monthly inin-come of the respondent,
lnfemin-come is the variable for females and lnmaleinlnfemin-come the variable for males. As a proxy for
inequality, two different variables are used in the intra-household regressions. The inequality measure lnmfratio coincides to the log of the monthly net male spouse income divided by the monthly net female spouse income within a household, whereby lnmfratio^2 is the squared lnmfratio variable. The inequality variable lngap corresponds to the log of the monthly net income of the male spouse subtracted by the monthly net income of the female spouse. Lngap^2 is the squared lngap variable. The former inequality measure is associated with relative terms, whereby the latter is associated with absolute terms. The same proxies for inequality are used in the intra-occupational regressions; lnoccgap and lnoccratio. The former is a variable corresponding to the log of the average men’s income within an occu-pation in a specific year divided by the average women’s income within the same occuoccu-pation in the same year, whereby the latter depicts the log of the ratio between the two.
In the intra-household inequality regression, the independent variable household income is included and in the intra-occupational inequality regression, the independent variable per-sonal income is used. Household income is included to observe the effect of income inequal-ity on happiness without being biased by household income. Accordingly, the estimated co-efficient of inequality will depict the effect of inequality on happiness which is not induced by household income. The same applies to the intra-occupational regressions, personal in-come is used to show the effect of inin-come inequality without being biased from the respond-ent’s income.
In order to conduct the robustness tests for the intra-household regressions, personal female income satisfaction as well as personal male income satisfaction are used, whereby the mean of the former 5.761 is significantly lower than the mean of the latter with about 6.377. Which again indicates, that males not only earn more on average than their counterparts in Germany, however, are also more satisfied with their income than females.
Within the robustness tests for the intra-occupational regressions the variables depicting fe-male income satisfaction (femincomesatis) and fe-male income satisfaction (fe-maleincomesatis) are used for replacing the variables female and male income respectively.
14
(see appendix, pp. 41 ff). The variables which are used during the analysis stem from micro-data of the SOEP and are representative on persons within Germany.
4 Empirical Results
The micro panel data used for the analysis accounts for individual heterogeneity and is un-balanced, meaning that the number of years is not the same for every individual. In the fol-lowing chapters, the unbalanced panel is used as “extracting a balanced panel out of an un-balanced one leads to an enormous loss in efficiency” and needless loss of data (Baltagi, 2011).
4.1 Intra-Household Income Inequality Regression
In the past, intra-household inequality has often been neglected in research as governments frequently assess happenings within households as private. However, inequality within households is omnipresent in both developing as well as developed countries and thus crucial to investigate (Woolley & Marshall, 1994). Within this sub-chapter the relationship between intra-household inequality and male as well as female happiness is studied, by conducting multiple regression analysis. Before conducting the analysis, different tests are performed to find out which regression model is appropriate to use with respect to the data used.
Firstly, the author conducted a Breusch-Pagan Lagrange multiplier test to observe whether a simple OLS regression or a random effects regression should be preferred. The null hy-pothesis that the variances across entities is zero can be rejected. Accordingly, a panel effect exists and a random effects model is superior to an OLS model.
In order to test whether a fixed effects model or a random effect model is more appropriate for the intra-household regressions, a Hausman test is conducted by comparing the estimates of the fixed effects model to the estimates of the random effects model. The fixed effects model captures all temporally constant individual-level effects such as race or culture, thus the net effect of the independent on the outcome variable can be assessed. Whereas within the random effects model it is assumed that the variation across entities is random and un-correlated with the independent variable within the model. The Hausman test shows that the null hypothesis stating that the unique errors are not correlated with the regressors can be rejected, as the p-value is zero (Stata, 2019). Therefore, a fixed-effects model will be used in the following analysis.
15
these control for heteroskedasticity across clusters of observations which is particularly rec-ommended when using panel data. Robust standard errors additionally control for autocor-relation.
Additionally, the author conducted a test to prove that the fixed effects model is superior to an averaged cross-section, to rule out that time variation in mfratio is minimal. However, the estimated coefficient mfratio is not significant in the average cross-section analysis due to time variation effects in mfratio.
Two last tests were conducted by the author; the Jarque-Bera test as well as the “sktest”, to test for normality. The Shapiro-Wilk test cannot be conducted, as the observations exceed 4000 within the regressions. Both tests show, that the null hypothesis of normality can be rejected. However, none of the existing normality tests can provide reliable answers if the number of observations is large enough. Ghasemi and Zahediasi (2012) state that if the sam-ple is large enough, meaning larger than 30 or 40, “the violation of the normality assumption should not cause major problems”. As around 40,000 observations within the regressions of this thesis exist, the violation of normality assumption would not cause problems. However, in order to deal with the skewed data, the natural logs of the variables mfratio, mfratio^2,
gap, gap^2, hhincome and age will be used in the following analysis.
All tests and outcomes apply to both, the female as well as the male happiness regression.
4.1.1 Inequality Measure
Two main proxies for household income inequality can be used, firstly the ratio between the male and female spouses’ income and secondly the income gap between the two. Figure 4 shows that there is only little difference between the significance of independent variables whether the income gap or the income ratio is used. The significance level of two religion variables within the male gap regression vanishes and the variable depicting age decreases its significance level from 1% to 5% in the same regression. In the female regression merely the variable single changes its significance level from 1% in the ratio regression to 5% in the gap regression. Additionally, the adjusted R2 is slightly higher for both, the female as well as for the male regression which included the gap.
16
only claims that one spouse earns a certain amount of money more than the other, it does not matter how much they earn. Furthermore, the gap is less significant for male happiness (10% level) than the ratio for male happiness (1% level).
One the other hand, the calculated variance inflation factors (VIF) are rather high in the ratio regression, compared to the gap regression, which means that multicollinearity is higher for the ratio regression, this can be supported by calculating a correlation matrix. Another way to calculate the gap might be taking the log of the absolute values function, in order to ob-serve the effect of the difference in income in general on happiness. However, this measure was not significant (p > 0.1).
Figure 4: Fixed Effects Regression with inequality ratio and inequality gap
(1) (2) (3) (4)
VARIABLES FEMALE
HAP-PINESS FEMALE HAP-PINESS MALE HAPPI-NESS MALE HAPPI-NESS lnmfratio -0.0838*** 0.0704*** (0.0174) (0.0158) lngap -0.0594*** 0.00763* (0.0143) (0.00389) lnhhincome 0.275*** 0.335*** 0.247*** 0.235*** (0.0349) (0.0517) (0.0322) (0.0574) lnage -0.312*** -0.532*** -0.299*** -0.263** (0.0808) (0.103) (0.0744) (0.131) health 0.232*** 0.224*** 0.242*** 0.247*** (0.00639) (0.00837) (0.00617) (0.0121) religionw 0.0887 0.0997 0.130** 0.0152 (0.0582) (0.0746) (0.0547) (0.117) religionm 0.0548 0.0535 0.0866** 0.0397 (0.0403) (0.0510) (0.0365) (0.0696) religions 0.0224 0.00471 0.0448** 0.0885** (0.0247) (0.0336) (0.0217) (0.0448) o.religionn - - - - married 0.405*** 0.439*** 0.472*** 0.526*** (0.0775) (0.144) (0.0769) (0.137) single 0.374*** 0.338** 0.454*** 0.436*** (0.0863) (0.156) (0.0840) (0.153) divorced 0.405*** 0.403*** 0.450*** 0.463*** (0.0815) (0.156) (0.0842) (0.159) widowed 0.0765 -0.0578 0.212 0.580 (0.152) (0.245) (0.185) (0.396) Constant 2.814*** 2.562*** 2.751*** 2.863*** (0.317) (0.446) (0.306) (0.546) Observations 40,146 40,297 43,890 44,035 Adjusted R2 0.092 0.093 0.100 0.107
Fixed Effects YES YES YES YES
17
Source: Own research with data from 34 waves (1984-2017) of the SOEP.
What has been additionally considered for the choice of the inequality measure is the exist-ence of multicollinearity between the household income and the income ratio as well as the household income and the income gap. Figure 5 shows a strong positive correlation between the household income and income gap, which means that the higher the household income, the higher the gap. This is rather not the case in the ratio regression, a higher household income does not necessarily lead to a higher ratio on average. This argument is in favor of using the ratio.
In summary, it can be said that arguments for using either of the inequality measures exist. However, as most of other papers measuring intra-household inequality use the ratio such as Woolley and Marshall (1994) or Lise and Seitz (2011), the ratio will be used during further analysis.
Figure 5: Scatterplot of Household Income and Income Gap/Income Ratio
Source: Own research with data from 34 waves (1984-2017) of the SOEP.
4.1.2 Multicollinearity
In order to test if multicollinearity within the regressions exists, the VIF scores as well as the Pearson product-moment correlation matrix are being calculated in the following sub-chapter.
18 Figure 6: VIF Scores in the Female Happiness Regression
Source: Own research with data from 34 waves (1984-2017) of the SOEP.
Another indicator for multicollinearity is the Pearson product-moment correlation coeffi-cient, denoted as r, whereby the value can range from -1 (denoting a perfect negative linear relationship) to 1 (denoting a perfect positive linear relationship). A strong correlation exists if |r| > .5, a moderate correlation if 0.3 < |r| < .5 and a small correlation if 0.1 < |r| < .3 (Cohen, 1988), whereby such a correlation only depicts a first look at the data, as no control or other independent variables are taken into account. Figure A3 (see appendix, p. 45) shows the Pearson correlation matrix, whereby in contrast to the VIF scores, no evidence for mul-ticollinearity can be found. This supports the fact that no severe problem with multicolline-arity exists, as only three bivariate correlations are strongly correlated; inequality and its squared variable, the variable depicting if the respondent never goes (religionn) to church and if the respondent seldomly attends church (religions) as well as single and married. The same applies to the male happiness regression, expect that the squared and “normal” ine-quality measure are not strongly but moderately correlated (see appendix, Figure A4, p.46). The correlation between the religion variables is the reason for why the variable depicting that the respondent never goes to church is being omitted in the regressions. Additionally, female and male happiness is significantly positively related to the inequality measure (fe-male regression: r = 0.06; p < 0.01/ (fe-male regression: r = 0.03; p < 0.01) which stands in contrast to the findings of the female regression analysis. However, as stated above the rela-tionship should not be over-emphasized, as it only provides a first glance at the data. Addi-tionally, female and male happiness are positively related to the independent variable house-hold income (female regression: r = 0.15; p < 0.01/ male regression: r = 0.16; p < 0.01) and to the squared income inequality variable lnmfratio^2 (female regression = 0.05; p < 0.01/ male regression: r = 0.01; p < 0.01).
Variable VIF 1/VIF
lnhhincome 47.80 0.020919 married 23.58 0.042409 lnage 15.26 0.065541 health 13.49 0.074144 single 8.8 0.113590 divorced 4.75 0.210326 widowed 1.87 0.533511 religions 1.60 0.625275 lnmfratio 1.50 0.665328 religion 1.20 0.831077 religionw 1.16 0.861123
Variable VIF 1/VIF
19 4.1.3 Robustness Tests
In order to test whether the variables of the core regression coefficient estimates, thus ine-quality (lnmfratio) and household income (lnhhincome), are plausible and robust, robustness checks are conducted as these increase the likelihood of structural validity of the results (White & Lu, 2014). In the first robustness check, the regression specification will be changed by replacing the dependent variables female happiness and male happiness with the variables female satisfaction with personal income and male satisfaction with personal in-come respectively. Another robustness check has been conducted, by using the Stata com-mand checkrob and defining the core as well as testing variables. Stata then “estimates a set of regressions where the dependent variable is regressed on the core variables and all possi-ble combinations of the testing variapossi-bles” (Barslund et al., 2007). If the signs and magnitudes of the coefficients do not change when replacing, adding or removing regressors, the esti-mated coefficients can be “reliably interpreted as the true causal effects of the associated regressors” (White & Lu, 2014). The robustness check of removing and including testing variables, such as age, health, marital status or religion, has shown that the estimated coef-ficients of inequality and household income are robust, as the sign has never changed and the magnitude only to a certain extent. The same applies to the robustness test, where the variable depicting household income has been replaced by household income satisfaction (see Figure A5 in the appendix, p. 47).
If the variable corresponding to employment is included, the magnitude and sign changes which can be due to the rather low adjusted R2. It is a well-known phenomenon that if the model accuracy is rather low, as it is the case in the regression model in Figure A5 (see appendix, p. 47), adding or removing variables can change the magnitude and/or the sign. Nevertheless, the signs do not change, and the significances and magnitudes change only slightly. The author conducted a last robustness test by replacing the dependent variable female happiness and male happiness by female income satisfaction and male income satis-faction respectively. The estimated coefficients did not change signs8 whereas the magnitude and the significance changed slightly (see appendix, Figure A6, p. 48). The conducted ro-bustness tests indicate validity of the results.
Moreover, the author conducted a sensitivity analysis. In specific, the Leamer’s extreme bound analysis, as the code for a Sala-i-Martin extreme bound analysis (1997) is not availa-ble for Stata yet. Leamer’s analysis tests if certain explanatory variaavaila-bles can robustly explain the dependent variable by computing the extreme upper and lower bound of the estimated coefficient of the variable of interest. If the lower bound is negative and the upper bound is positive, the variable of interest is not correlated with the dependent variable. Conducting the analysis, it can be said that the variables of inequality and the variable of household income are highly robust for the female regression as well as for the male regression (see
20
Figure A14 and Figure A15 in the appendix, p. 56). The sign of the inequality measure is not negative for the female regression, this can be due to the many limitations of Leamer’s analysis, for example that it can only be conducted with OLS models and no options such as robust standard errors can be included (Poprawe, 2012).
4.1.4 Regression Results and Discussion
Female and male happiness are specified as a weak form of interval-scaled variables and are thus not clearly metric, meaning that the model is not a typical linear regression model. Therefore, it is not surprising, that when a scatterplot is created between the interval-scaled variable female or male happiness and the metric inequality variable lnmfratio, a direct linear relationship cannot be observed (Backhaus et al., 2016). Nevertheless, a linear relation is assumed, as the happiness variable can arguably be categorized as an interval variable, where the order and difference between the given values is known (Kalmijn et al., 2010). Further-more, a quadratic regression will be conducted to analyze if an (inverted) u-shape relation-ship between household income inequality and female as well as male happiness exists, thus whether the relationship wears off at a certain point. The model validity in all four regres-sions in Figure 7 is given. Conducting the f-test, the hypothesis that all coefficients of the independent variables are equal to zero can be rejected (p < 0.01). The adjusted R2, a measure of goodness of fit, may be rather low with about 10% which means that only 10% of the variance can be explained through the model. However, Stoetzer (2017) states that the level of the adjusted R2 heavily depends on the data and problem analysis; while a macroeconomic panel data analysis usually results in a very high adjusted R2, the adjusted R2 in micro panel data where individual behavior is analyzed, is often very low and under 30%. Thus, a high share of unexplained scattering within the model exists, and results can still be economically properly interpreted. This study is a unique investigation of gender income inequality and happiness within households and therefore other papers and comparable adjusted R2 on the same topic do not exist.
Figure 7: Regression Results Intra-Household Regression
(1) (2) (3) (4)
VARIABLES FEMALE
21 health 0.232*** 0.232*** 0.242*** 0.242*** (0.00639) (0.00638) (0.00617) (0.00617) religionw 0.0887 0.0888 0.130** 0.129** (0.0582) (0.0582) (0.0547) (0.0547) religionm 0.0548 0.0549 0.0866** 0.0864** (0.0403) (0.0403) (0.0365) (0.0365) religions 0.0224 0.0228 0.0448** 0.0445** (0.0247) (0.0247) (0.0217) (0.0216) o.religionn - - - - married 0.405*** 0.405*** 0.472*** 0.473*** (0.0775) (0.0775) (0.0769) (0.0769) single 0.374*** 0.374*** 0.454*** 0.456*** (0.0863) (0.0864) (0.0840) (0.0840) divorced 0.405*** 0.405*** 0.450*** 0.450*** (0.0815) (0.0815) (0.0842) (0.0842) widowed 0.0765 0.0747 0.212 0.212 (0.152) (0.152) (0.185) (0.184) Constant 4.253*** 4.261*** 4.129*** 4.142*** (0.314) (0.314) (0.310) (0.310) Observations 39,986 39,986 43,780 43,780 Adjusted R2 0.092 0.092 0.100 0.100
Fixed Effects YES YES YES YES
Notes: Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. Source: Own research with data from 34 waves (1984-2017) of the SOEP.
Looking at the first regression in Figure 8, the variables lnmfratio, lnhhincome, lnage, health,
married, single, divorced and the constant are statistically significant at the 1% level. If an
independent variable is used in its logarithmic form and the dependent variable in a linear form, it is called a linear-log model. A 1%-change of the independent variable is associated with a change of the dependent variable of 0.01 times the estimated coefficient. If however, the independent and the dependent variable are both used in their linear forms, a one unit increase in the independent variable is associated with a change of the dependent variable of the estimated coefficient. Thus, if the inequality increases by one percent, female happiness decreases by 0.0019 units c.p. on average. Age and female happiness as well have a negative
relationship, if age increases by one percent, female happiness decreases by 0.003 units. Health and household income both have a positive significant relationship with female hap-piness, if health increases by one unit, female happiness increases by 0.232 and if the house-hold income increases by one percent, female happiness increases by 0.003 units c.p. on
22
average. Furthermore, the marital variables married, single and divorced are associated with a positive relationship with female happiness. If the respondent is married, females enjoy a happiness bonus of 0.405 units compared to a respondent who is not being married c.p. on average. If the respondent is single, females’ happiness increases by 0.374 units and if the respondent is divorced, females enjoy a bonus of 0.405 units of happiness c.p. on average. The religion variables and the variable widowed are not significant.
Within the first male regression in Figure 8 all variables expect for widowed are significant. If inequality increases by one percent, male happiness increases by 0.001 units c.p. on aver-age. The other signs and magnitudes of the estimated coefficients are similar to those of the female happiness regression, except for the religion dummies which are now statistically significant. If the male goes to church or attends a religious event once a week, he enjoys a happiness bonus of 0.130 units, whereby if the respondent attends such an event once a month, the bonus would be 0.087 units and if he seldomly attends such an event, the bonus would be 0.045 units c.p. on average at a 5% significance level. It is important to mention that in many recent papers the significance level and therefore the p-value, as a value of implied truth is heavily criticized because scholars e.g. conduct p-hacking10 or use very lib-eral alpha levels within their hypothesis testing (Branch, 2019). The inequality measure within the intra-household regressions however are highly significant; even if a very small alpha level of below 0.01 is used, the variables remain significant.
Recapitulating hypotheses 1a and 1b of the thesis:
Hypothesis 1a: Within households a higher gender income inequality leads to a de-crease in females’ happiness;
Hypothesis 1b: Within households a higher gender income inequality leads to an increase in males’ happiness;
one can say that within households a higher gender income inequality is associated with a decrease in females’ happiness. Females’ happiness may decrease because of an aver-sion against inequality, their income rank within the household and also their low relative income within the household. Moreover, the economic dependence of women on their spouse may lead to the decrease in females’ happiness. It might also be the case that the decrease stems from women who must forego their personal career ambitions in order to ensure that the family’s domestic (e.g. childcare) obligations are met, while their male part-ner is not similarly restricted.
Within households a higher gender income inequality is associated with an increase in males’ happiness. However, a restriction exists. If the squared income inequality variable
lnmfra-tio^2 is included, male happiness decreases at a significance level of 10%, which means that
the relationship between household income inequality and male happiness is an inverted
23
shape, whereby the turning point is 2.334. This can be supported by conducting a u-test in Stata. The null hypothesis of the test, stating that the relationship is u-shaped, can be rejected and an inverted u-shape for males exists between the inequality measure and the squared inequality measure. This means that the male needs to earn 10.319 times11 the income of the
female spouse for the inequality to have a negative effect on male happiness. The squared inequality variable is not significant within the female regression, the relationship is there-fore not u-shaped or inverted u-shaped for females. Accordingly, household income inequal-ity has a constant negative effect on female happiness. The negative relationship between inequality and male happiness after the turning point and the positive relationship before may be explained by various factors.
First, the positive relationship between wage inequality and male happiness may reflect a pervasive form of gender stereotyping, as most males like to earn more than their spouse since they are socialized to care more about job satisfaction and material rewards than fe-males. This socialization has been mentioned in previous chapters, in the form of career interests and decisions of young children. Lu (2000) supports this by stating that “husbands [are] more committed to the worker role, whereas the wives [are] more committed to the parental role”. The statement of Lu (2000) cannot explain the negative relationship of female happiness and inequality. Secondly, if inequality exists, the rank of the men in income is higher than the one of females which contributes to an increase in happiness. The Mean Dependence Model as well states that if one’s own income is higher compared to the income of others, happiness increases. The income of spouses is directly comparable, as usually exact knowledge about the income of the partner exists.
However, if the gender income inequality is very high and the male earns much more than his spouse (10.319 times the income of the spouse), inequality has a negative effect on male happiness. According to the Inequity Aversion Model this effect could derive from an aver-sion against high inequality. Other explanations might be that men who earn 10.319 or more than their spouse feel pressured by their role of being the main earner. Additionally, the high number of working hours may lead to a smaller time with their children and a relatively poor work-life-balance compared to a situation with a lower inequality, where the man earns less and the woman earns more (Holter, 2014). In a situation where the man earns the same and the woman earns more, pressure and respective over time can be decreased and therefore work-life-balance presumably increased.
Moreover, the author investigates if males’ happiness is dominated by the inequality effect or the household income effect. If the inequality within a household is high, the household income is relatively low, compared to the same situation where the female earns approxi-mately the same as the male. Whether the effect on male happiness is dominated by the household income effect or by the inequality effect can be tested by taking the average male
24
in the intra-household regression and observing his happiness if his spouse earns different amounts of money. Within the SOEP data used in the thesis (SOEP, 2019), the average male earns approximately 1940€ (net value) per month. Figure 9 shows the change in men’s hap-piness if the income of his spouse changes, using the formula12:
𝐻
𝑖𝑡= 0.0942ln (𝑥
𝑖𝑡) + 0.242ln (𝑦
𝑖𝑡) (7)
If the male earns 1940€ per month and the female within the household earns 50€ per month, male happiness would be 2.1828, if only the inequality measure and the household income are considered. Figure 8 shows that until the female spouse earns 1500€, the inequality effect dominates, since male happiness decreases if female income increases. After the turning point the household effect dominates as despite the inequality decreases, males’ happiness slightly increases again, this holds also true if the woman earns more than her spouse. Within this exemplary scenario, the female must earn approximately 70% or more of the men’s income for the household income effect to dominate. If she earns less than 70% of his wage, the inequality effect dominates. This finding may indicate that the inequality is not substan-tial enough for many men to enjoy a status advantage until the women earns less than 70% of his wage. It may also be the case that the relatively small gender wage gap of 30% or less is typically present in households comprised of more gender egalitarian couples.Figure 8: Change of Average Male Happiness Depending on Women’s Wage
Source: Own research with data from 34 waves (1984-2017) of the SOEP.
Furthermore, the author investigates whether a binary relationship between male happiness and gender income inequality could exist which turns into a linear one if the male earns more than the female within a household. Such a relationship would mean that most men are happy as long as their wife earns less but are still supportive regarding their work and wage. This could be supported by the Model based on rank, only two ranks exist in a household and as long as the man comes in first, he would be happy. In order to test a binary relationship, a dummy variable of the inequality variable is created and set to one if the ratio exceeds one, thus if the male earns more than his spouse. Figure A7 (see appendix, p. 49) shows that a
25
binary relationship between male happiness and inequality does not exist, as the binary var-iable is not significant.
In summary, it can be said that male happiness is positively associated with income inequal-ity and a very high income inequalinequal-ity is negatively associated with male happiness. Further-more, taken the household income and the inequality into account, the inequality effect dom-inates until the woman earns about 70% of the males’ wage. The household income effect dominates after the turning point. Even if the female earns more than her spouse, male hap-piness increases because of the relatively high household income.
Female happiness is negatively associated with an increase in income inequality. Hypothesis 1a, stating that within households a higher gender income inequality leads to a decrease in females’ happiness, can be supported by this study. Hypothesis 1b, stating that within house-holds a higher gender income inequality leads to an increase in males’ happiness, can only partly be supported, as the relationship between male happiness and gender income inequal-ity is inverted u-shaped.
4.2 Intra-Occupational Income Inequality Regression
Intra-occupational inequality is a crucial topic to investigate, Moore (2018) for example finds that the gender wage gap is persistent mainly because of intra-occupational wage dis-persions between men and women. Therefore, this sub-chapter analyzes the relationship be-tween gender income inequality within occupations and happiness of men and women. Figure 9 shows a summary of the conducted tests for the intra-occupational female regres-sions and their definitions, outcomes, results and solutions.
Figure 9: Conducted Tests and Diagnostics (Intra-Occupational Regression)
Test Name Definition Outcome Result Solution
Breusch-Pagan Mul-tiplier Test
Null hypothesis defines that variances across enti-ties are zero. Thus, no significant panel effect exists.
chibar2(01) = 4262.09 Prob > chibar2 = 0.0000
Reject the null hy-pothesis and con-clude that a random effects model is su-perior to a simple OLS model.
Hausman Test
Decide between a fixed effects and a random ef-fects model. The null hy-pothesis defines that the unique errors are not cor-related with the regres-sors. chi2(11) = (b- B)'[(V_b-V_B)^(-1)](b-B) = 603.66 Prob>chi2 = 0.0000
Reject the null hy-pothesis and con-clude that the fixed effects model is pre-ferred.
Modified Wald Test
Tests for groupwise het-eroskedasticity. The null
chi2 (17511) = 4.6e+38
The null hypothesis can be rejected, which means that
26
hypothesis defines the ex-istence of homoskedas-ticity. Prob>chi2 = 0.0000 heteroskedasticity exists. used (clus-tered) Jarque-Bera Test for normal-ity
The null hypothesis de-fines that normality ex-ists. Jarque-Bera statis-tic= 9007.7423 Jarque-Bera p-value = 0 Null hypothesis of normality can be re-jected.
Use the log of occratio,
oc-cgap,
femin-come,
ma-leincome and age.
Source: Own research with data from 34 waves (1984-2017) of the SOEP.
All tests and results apply to both, the female as well as the male happiness regression. As an inequality measure the ratio and not the gap will be used, reasons for this choice have been given in previous chapters. Nonetheless, a comparison of both regressions can be found in Figure A8 (see appendix, p. 50).
4.2.1 Multicollinearity
In order to test whether multicollinearity within the intra-occupational regressions exists, the variance inflation factors (VIF) and the Pearson product-moment correlation matrix are be-ing calculated in the followbe-ing subchapter.
As well as in the within household inequality regression, the VIF scores are rather high. In order to cope with this situation, the variables corresponding to age, married and health, which exhibit rather high VIF scores, are excluded. Figure A9 (see appendix, p. 51) shows that the VIF scores in the female happiness regression without these variables are all below 4. As the results in both regressions are qualitatively the same, one can say that the relatively high VIF scores and thus multicollinearity do not bias the results in the regression. This as well applies to the male happiness regression and can be supported by the Pearson correla-tion matrix, which shows no evidence for multicollinearity.
