Estimating the effect of urbanization on income inequality. An expansion of
the Kuznets curve
By Daniel Marcel Haerle
University of Amsterdam
Bachelor Thesis – Amsterdam School of Economics
Supervisor: Kees Haasnoot
January 2018
Abstract This paper examines the effect of urbanization on income inequality in addition to the relationship of inequality with development described by the Kuznets curve, by using panel data over the time period 1995-2015. The results show strong indication for the truthfulness of the Kuznets curve, but also demonstrate that urbanization increases inequality for higher development levels, while it decreases it for lower levels. The relationship is robust for various model specifications. This effect is, however, not evident in this form when considering developments over time. It can be concluded that a country’s level of urbanization has a significant effect on its income distribution and should therefore be taken into consideration for urban policy and spatial planning.Statement of Originality This document is written by Student Daniel Marcel Haerle who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents 1. Introduction ... 4 2. Literature Review ... 6 3. Research Method and Data ... 10 3.1 Data collection ... 10 3.2 Model ... 13 3.3 Data description ... 15 3.4 Hypotheses ... 18 4. Results ... 19 5. Discussion ... 24 6. Conclusion ... 27 References ... 28 Appendix ... 31 A. Table 1: Means and standard deviations of the independent variables ... 31 B. Table 2: Regression coefficients and average margins; dependent variable gini ... 32 C. Table 3: Regression coefficients and average margins; dependent variable gini; variables centered to their respective means ... 33
1. Introduction An unequal distribution of wealth and income within a society is associated with an array of features that are deemed undesirable from an economic perspective. In utilitarian thinking, it is distributively inefficient and can be a cause of unhappiness, as has been corroborated in empirical work (Oishi, Kesebir, & Diener, 2011). In addition, Kawachi, Kennedy, Lochner, and Prothrow-Stith (1997) show that income inequality is strongly correlated with mortality rates and reductions in social capital, which is in line with other, similar research indicating that inequality is related with health issues, but also social cohesion and trust in society (Brown & Uslaner, 2002; Oishi, Kesebir, & Diener, 2011). Inequality has also been shown to be a cause of crime, where in research homicide rates have been proven to strongly correlate with inequality across place and time (Martin, Wilson, & Vasdev, 2001). Considering these conclusions, it is therefore in the interest of society that research analyzes factors in connection with economic inequality. Despite a wide existing literature on this topic, economic research regarding the causes of unequal income distributions thus far has largely neglected the process of urbanization. Urbanization has been an ongoing development in the modern age and a phenomenon accompanying globalization and industrialization. For the first time in history, more than half of the world’s human population now lives in urban areas, while its share keeps rising continuously (World Bank, n.d.). Nowadays, an ever-increasing number of cities of previously unknown magnitudes is emerging, mainly in developing countries. Many of these urban areas are characterized by crass contrasts in income and wealth, made visible in many lesser developed countries by slums. According to World Bank (n.d.) data, over 65% of the population in low income countries lives in slums. In light of this, it is worth examining the relationship between this rapid urbanization and the inequality observed in a country. On the other hand, as Simler & Dudwick (2010) note, “urban development is an integral part of economic development”. The population shift from rural to urban areas is thus inevitably linked with economic progress. Further, it can be seen that in many nations, there is a strong urban-rural divide, in economic as well as in political and societal terms. People are flocking to the cities in search of a better life, and looking at the data of the United Nations Population Division’s Urbanization Prospects, it becomes apparent that in almost all parts of the world, industrialized or developing, the urban share of the population
is increasing (United Nations, Department of Economic and Social Affairs, Population Division, 2014). As countries develop and urbanize, some countries display a staggering degree of inequality in their income distribution, raising the concern as to how to the greater part of the populace can profit from economic development. This question of as to how economic growth relates to inequality has been a longstanding concern for economists. Kuznets’ (1955) famous contribution to this topic predicts that as countries develop from low to high income levels, inequality will rise and subsequently fall, thereby linking inequality with economic progress. In this, he did take urbanization into account, but as literature and data, especially from the World Development Report (WDR) 2009 show, there is likely to exist additional links of urbanization to how inequality in a country develops. Rather than the mere shift from rural to urban population, urbanization relates to inequality in more ways, including changes in within- and between-region inequality. It can thus prove insightful to single out the effect of urbanization on overall economic inequality in addition to the observed correlation the latter has with the stage of development in a country. If urbanization can be shown to influence inequality along a country’s development path, this could bear far-reaching policy implications for urban development policies and spatial governance. In the case that urbanization turns out to increase inequality, governments may want to invest in rural development programs to strengthen the rural sector in order to avoid excessively unbalanced growth, whereas in the opposite case it might even be desirable to stimulate urbanization instead. This paper therefore aims at answering the question: To what extent does the process of urbanization affect income inequality in a country? In order to do so I will analyze comprehensive panel data across countries and time to estimate the urbanization effect by means of an econometric regression. The presentation of the relevant literature regarding this topic will precede the regression model, an analysis of the underlying data and a discussion of the results, followed by a conclusion.
2. Literature review Urbanization refers to the increase in the urban population as a share of the total population. A country’s level of urbanization describes the share of the urban population, thus the people living in what is designated as a city or town, as part of the total population in the country. There are different ways of defining the term ‘urban’. Generally, cities and towns are “a geographic area characterized by a concentration of economic actors” (World Bank, 2009). Although researchers have been making attempts at finding universally valid measures for ‘urban concentration’ and ‘agglomeration’, it is conventional to go by each country’s administrative delineations. This is also done in the United Nations’ (2014) World Urbanization Prospects (WUP), whose classification I will follow in this paper. The data are corrected for changes in definitions and therefore achieve a high degree of accuracy over time. Urbanization can be caused by a combination of the following factors: the migration of people from rural to urban areas, natural increase, and the transformation of growing small-scale rural population centers into urban areas (Simler & Dudwick, 2010). Migration is frequently understood to be the chief driver of urbanization. Kuznets (1955) and classical models on the dual economy such as the dual-sector model by Lewis (1954) build up on this notion that urban areas, which are comprised of the industrialized sector of the economy, attract labor from the countryside, which is characterized by agriculture and lower wages. Urbanization by natural increase on the other hand depends on the urban versus the rural fertility rates. Birth rates are often reported to be considerably higher in rural areas in developing countries, which might add to the migration effect and to urbanization rates in those countries (Li & Wang, 1994; Mpando, 2000). The growth of rural areas into urban ones is likely to play a role especially in countries with high population growth in rural areas, which is again probable to be the case in many developing countries and adds to urbanization. There are different ways of measuring economic inequality. The most common metrics study disparities in income, consumption, and wealth. Despite the latter having a high relevance by itself in terms of, for instance, the inheritance of certain living standards, and it being the core generating source for the other two metrics, it does not provide a good proxy measure for wellbeing. Wellbeing can be better assessed in terms of consumption, for which income has to be spent. Using Engel curves, Aguiar and Bils (2015) show that
consumption and income inequality closely track each other. Data on income inequality are generally the most widely available, an important feature for broad cross-country and time series research. For this reason, inequality among incomes is the measure used in this paper. Finding and quantifying the causes affecting a country’s income distribution has been subject to much research with differences in findings. Kuznets (1955) posits a parabolic correlation between a nation’s level of development and income inequality, referred to as the Kuznets curve, which has been i. a. corroborated by Nielsen & Alderson (1997) for subnational data in the United States. Generally, least developed and highly developed countries experience lower inequality than middle income states, visualizing the Kuznets curve as an “inverted U”. This can be explained by a combination of economic and social factors, but is mostly explained by the population shift from agricultural, rural activities to modern ones (Kuznets, 1955). This shift has been termed the Kuznets process. The Kuznets curve and the variables involved in that process embody a wide array of factors whose combination influences the inequality in a country. Williamson (1965) furthers this conjecture and discovers a link between a country’s development level and its regional income disparities, thus spatial inequality. This relationship has been extended to urbanization. Davis and Henderson (2003) also posit that urbanization is driven by structural change in the economy along the process of development, whereby a strong link between a country’s level of urbanization and its economic development can be established. This is mainly explained by the shift in an economy’s sector composition from agricultural to industrial activity, but government policy and institutions also have a modest effect on urbanization (Davis & Henderson, 2003). Hence, more developed countries are generally more highly urbanized than less developed countries, which also shows in the World Bank’s World Development Report 2009 (World Bank, 2009). Hence, the development level is a good predictor for the level of urbanization (Simler & Dudwick, 2010). This insight is in line with Kuznets as Kuznets processes lead to more urban economic concentration for higher development levels. As Davis and Henderson (2003) note, government policy should be concerned with over- and under-urbanization in terms of economic growth. It remains to be examined whether this should be of concern for inequality, too.
Previous research on the direct effect of urbanization on income inequality has been incomprehensive. This is partly because urbanization relates to inequality in two ways. Firstly, it shifts the weights among spatial inequalities of the rural-urban income and consumption gaps and secondly, such spatial inequalities also exist within urban and rural areas, for which an increase in the share of the urban population bears implications for overall inequality. Following Gustafsson & Shi (2002) in their analysis of regional inequality, this notion can be expressed with the following equation: 𝑇𝑜𝑡𝑎𝑙 𝑖𝑛𝑒𝑞𝑎𝑙𝑖𝑡𝑦 = 𝑊𝑖𝑡ℎ𝑖𝑛-𝑟𝑒𝑔𝑖𝑜𝑛 𝑖𝑛𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦 + 𝑈𝑟𝑏𝑎𝑛-𝑟𝑢𝑟𝑎𝑙 𝑖𝑛𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦 in which the concept of region refers either to the urban or to the rural part of a country, respectively. 𝑊𝑖𝑡ℎ𝑖𝑛-𝑟𝑒𝑔𝑖𝑜𝑛 𝑖𝑛𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦 is hence the weighted sum of the inequalities in both regions and 𝑈𝑟𝑏𝑎𝑛-𝑟𝑢𝑟𝑎𝑙 𝑖𝑛𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦, or analogously 𝐵𝑒𝑡𝑤𝑒𝑒𝑛-𝑟𝑒𝑔𝑖𝑜𝑛 𝑖𝑛𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦, the inequality between them. In respect of the latter, the urban-rural gap is positive for virtually all observations across both time and countries in terms of most income based and non-income based measures (Simler & Dudwick, 2010). This becomes apparent in the data of the World Development Report (WDR) 2009 (World Bank, 2009). Moomaw and Shatter (1996) explain that internal economies of scale, reduction of transportation costs and increased specialization of labor make urban areas the drivers of economic growth, thus rationalizing the data. This is in line with the relationship between development and urbanization as elucidated above. As regards the literature on the nature of this in-between inequality, the WDR 2009 suggests an inverted U-shaped relationship with the level of development and the level of urbanization, a notion consistent with much of the literature (World Bank, 2009). Accordingly, the urban-rural gap increases with increases in development for low-income countries, but narrows once countries reach a certain middle-income level threshold. Analogously, the report conjectures an inverted U-shaped correlation of the urban-rural gap with the level of urbanization, but also claims that the vast majority of developing countries has passed the peak and shows a narrowing trend for all levels of urbanization (World Bank, 2009). The underlying reasoning is that in the initial stages of development, there are leading areas consisting in urban cores, self-perpetuated by positive feedback, a term coined by Arthur (1990), while the rural periphery lags behind. The gap is generally large and
increasing in initial stages of development due to the intensification of economic activity and accumulation of capital, consumers, and workers leading to production advantages and economies of agglomeration. The gap narrows again at higher development and urbanization levels through higher investments in the rural economy and redistribution (World Bank, 2009). Simler & Dudwick (2010) corroborate this general notion but find great variation across countries, thus implying “that policies and institutions can play a crucial role in rural-urban transformations.” There is no conclusive research regarding the relationship of within-region inequality with urbanization. Kuznets (1955), however, postulates urban areas to be inherently more unequal than rural areas. This is due to the diversity of social conditions in urban areas compared to the small economic units that constitute rural areas. This makes a case for the within-region effect of urbanization on inequality to be positive (Nielsen & Alderson, 1997). In his research, however, Young (2013) finds that most of the spatial inequality within a country can be attributed to the urban-rural gap, making it the main contributor of the effect of urbanization on overall inequality. The urban-rural gap and, in particular, the relationship of urbanization with within-region inequality certainly merit attention for further examination but exceed the scope of this paper’s research. Instead, the combined observable effect of urbanization on inequality is focused on. The implication of this is that an inverted U-shaped relationship between the inequality in a country and the level of urbanization can be expected. In their research, Castells and Royuela (2012) connect the economies and diseconomies of agglomeration brought about by urbanization with inequality, showing that the benefits in the development process derived from agglomeration depend on the underlying income distribution. The economic development level can therefore be said to be linked to both welfare inequality and urbanization, thus implying that the relationship between a country’s inequality and its urbanization level goes beyond the correlations described by the classical Kuznets curve. A recent case study by Kanbur and Zhuang (2013) regarding urbanization and inequality in four selected countries in Asia has brought about findings supporting this view. In Indonesia and the Philippines, where urbanization rates are high, urbanization appears to be a main driver of increasing inequality more so than in India, which is urbanizing at a
slower rate. All three countries have relatively low levels of urbanization. Further, urbanization has helped reduce inequality in China (Kanbur & Zhuang, 2013). The findings show that China has already reached the turning point after which urbanization decreases inequality, which also has been corroborated by Zhang & Bao (2015). Research in more developed and more highly urbanized countries shows the decreasing side of the curve. Crenshaw (1993) finds that increases in population density lead to increased “social carrying capacity” and better social organization, ultimately reducing income disparities in more advanced, service based economies. Concluding, it can be seen that by describing a relationship between inequality and economic development, the Kuznets curve bundles a diverse set of factors that affect the development of inequality in a country. From various literature regarding urbanization and inequality it shows that the level of urbanization, mainly through spatial inequalities, can alter the Kuznets curve. 3. Research method and data 3.1 Data collection The model’s purpose is to identify the effect of the curved relationship inequality-urbanization and its added explanatory value to the classical Kuznets curve. This is to be estimated by means of a beta regression analysis of annual panel data on 161 countries over the time period of 1995-2015. Income inequality, the dependent variable, is to be estimated by the Gini coefficient. The Gini index is scale invariant and a useful tool to generalize the inequality in different income distributions by representing them in a number. More specifically, an index of 0 would imply perfect equality in a country, whereas 1 suggests perfect inequality in the distribution. Despite the one-dimensionality of this number (the specifics of a country’s income distribution and hence its Lorenz curve are not represented), the index provides a good tool to analyze developments over time, but also to compare countries, which is in line with the goal of this paper. The data for the Gini coefficients are drawn from the data provided by the Standardized World Income Inequality Database (SWIID) by Solt (2016), using the band of different estimates of the Gini indices based on household disposable income data. This way, the uncertainty of the estimates and their corresponding standard errors can be incorporated into the regression by means of multiple imputation estimation and thus yield more reliable results. The graphic in image 1 demonstrates the uncertainty of the database’s Gini estimates of disposable income by
representing the 95% confidence intervals around their mean. As Solt (2016) notes, the uncertainty is generally higher for developing countries than for developed countries, which can mainly be ascribed to less data being available for those countries. Image 1 In this database, Solt (2016) uses the Luxembourg Income Study as a base and combines it with “data from the OECD Income Distribution Database, the Socio-Economic Database for Latin America and the Caribbean generated by CEDLAS and the World Bank, Eurostat, the World Bank’s PovcalNet, the UN Economic Commission for Latin America and the Caribbean, national statistical offices around the world, and academic studies”. This insures a wide coverage and great cross-country comparability of the Gini indices. As Solt (2016) notes, country-year data on Gini coefficients faces the trade-off of comparability versus coverage. The LIS ensures exceptional comparability as it harmonizes household microdata into uniform classifications, but lacks a large amount of data points. Other researchers have tried to combine different data sources, where some have attempted to adjust the additional data, marked with dummies, by running an adjustment regression. Adjustments are necessary since the measurement of inequality can be based on differing definitions of welfare, such as gross income, net income, and expenditure and can make use of differing equivalence scales for household income (Solt, 2016). He however argues that fixed adjustments will consistently yield false estimates. In the database, the different source data
are classified under the above categories, where multiple observations for a data point in the same category are averaged. For some observations, this allows for the computation of the individual adjustment ratio. Subsequently, instead of applying fixed adjustment ratios, for the reason mentioned earlier, sophisticated predictions of those adjustment ratios are made to create estimates for the missing data points in each category, where sharp increases or decreases are only permitted for reasonable cases, e.g. data involving the microdata-based LIS and estimates around the transition period of the formerly communist states (Solt, 2016). In practice, this is done by multiple imputation, meaning that there are 100 imputed values for each data point to be included in the regression to account for the uncertainty of the estimates. As a result, the SWIID ensures vast coverage over countries and time while retaining a large degree of accuracy, which makes it highly appropriate for cross-national research as broad as in this paper. The measure used in this paper is the “(e)stimate of (the) Gini index of inequality in equivalized (square root scale) household disposable (post-tax, post-transfer) income”. This measure is most accurate in depicting reality as it represents a nation’s net income inequality. The explanatory variable is the level of urbanization in a country as given by the United Nations (2014). As Simler & Dudwick (2010) note, it is a somewhat imperfect measure since it is based upon national definitions of what constitutes urban versus rural, but does meet the demand of being nationally representative within the regression. One major objective of this database is moreover to “prepare consistent population estimates over historical time periods for human settlement that follow the same definition” (United Nations, 2014). This includes readjustments of past measurements and estimates and thus warrants the applicability to the research in this paper. To represent the level of development in accordance with the classical Kuznets relationship, the United Nations Human Development Index (HDI) is to be used (United Nations, n.d.). Kuznets based his work on an absolute income measure, whose measurement he contributed to greatly, in lack of a more comprehensive measure of development at the time. In contrast to this classical measure of development, national income, the HDI makes an attempt at including additional aspects of human development by augmenting the conventional gross national income measure with years of schooling and overall health while incorporating GDP adjusted at purchasing power parity. Despite not being a perfect measure of a country’s development
(the index receives criticism regarding both the methods of measurement of its components and its compositional structure), it captures the basic elements of human development while providing wide availability and comparability across countries and over time. The HDI assigns countries a score ranging from 0 to 1, where 1 represents the highest possible level of development a country could achieve. In this paper’s regression, the values also pertain to this range. The level of corruption in a country will be used as a control variable since corruption is most likely to be correlated with both the dependent variable and the explanatory variables. Gupta, Davoodi, and Alonso-Terme (2002) imply a strong positive relationship between the level of corruption in a country and inequality. In addition to that, Billger and Goel’s (2009) findings show that the level of urbanization shows significant negative correlation with the level of corruption in a country. These conclusions imply a need to include a measure of corruption in the regression. Corruption is to be measured by the scores given by the Corruption Perceptions Index by Transparency International (n.d.). The index is a composite measure of corruption perceptions among public officials and politicians and makes use of aggregate poll data from independent expert and business surveys, where corruption is defined as “the abuse of entrusted power for private gain” (Rohwer, 2009; Transparency International, n.d.). The CPI ranges from 0 to 10 in older issues and from 0 to 100 in newer ones, where 0 represents the highest possible level of corruption. For the regression in this paper, all values have been normalized to values from 0 to 1. Rohwer (2009) notes that indices such as the CPI should be viewed with caution as they only measure the perceptions of corruption in a country and are thus to a certain degree imperfect in the picture they depict. It does however, despite its flaws, give a rather comprehensive view on corruption given the lack of means for objective measurement. 3.2 Model Drawing from the literature, the level of development should have a curved relationship with the level of inequality, therefore square terms are included. Despite its hypothesized curved relationship, the WDR 2009 suggests that urbanization is now declining in inequality, so the curvature is expected to be weak, while showing mostly a downward trend. It is further expected to interact with the level of development. The dependent
variable faces a strong restriction: the Gini coefficient is by nature bound to an interval between zero and one, which requires additional econometric modeling to account for this restriction so as to ensure that regression outcomes be within these bounds. As the data show that none of the observations assumes a Gini coefficient of exactly 0 or 1 (which represent the unrealistic cases of perfect equality and perfect inequality), a regression can be run using a beta distribution. As noted by Ferrari and Cribari-Neto (2004), “the beta distribution is very flexible for modelling data on the standard unit interval” and has thus high applicability by implementing a maximum likelihood estimation. The distribution of the variable is likely to be heteroskedastic as this is typical for data within the standard unit data interval, and is thus likely to display asymmetries and long tails (Paolino, 2001). The beta regression accounts for these features by regressing on the mean of the dependent variable, assuming a beta distribution (Ferrari & Cribari-Neto, 2004). For this reason, beta regressions are widely used to regress on Gini coefficients. This results in the following specification of the regression model: 𝐸 𝑔𝑖𝑛𝑖 𝑥 = 𝜇; 𝑔 𝜇; = 𝑥𝛽 ⇔ 𝜇; = 𝑔>? 𝑥𝛽 ∈ (0,1) where 𝜇; follows a beta distribution within the (0,1) interval. Here, the beta regression works with a logit link, hence: ln 𝜇; 1 − 𝜇; = 𝑥𝛽 𝜇; = 𝑒;I 1 + 𝑒;I where 𝑥𝛽 = 𝛽J+ 𝛽?𝑢𝑟𝑏𝑝𝑜𝑝LM + 𝛽N𝑢𝑟𝑏𝑝𝑜𝑝LM∗ ℎ𝑑𝑖LM+ 𝛽Qℎ𝑑𝑖LM+ 𝛽Rℎ𝑑𝑖LMN + 𝛽R𝑐𝑝𝑖LM and where 𝑔𝑖𝑛𝑖LM is country 𝑖’s Gini coefficient measuring its income inequality in year 𝑡, 𝑢𝑟𝑏𝑝𝑜𝑝LM the share of the urban population among the country’s total population, ℎ𝑑𝑖LM the Human Development Index, and 𝑐𝑝𝑖LM the Corruption Perceptions Index. The Kuznets curve is represented by the quadratic polynomial of the HDI, whereas the urbanization effect is reflected in a linear as well as an interaction term with the level of development. The above specification will be referred to as the main regression. In a second and third control regression, the interaction will be expanded to a quadratic interaction term. As we will see, this does not add much explanatory value. The conditional variance of the beta-distributed dependent variable is given by
𝑉𝑎𝑟 𝑔𝑖𝑛𝑖 𝑥 =𝜇;(1 − 𝜇;) 1 + 𝜓 where 𝜓 > 0 is a scale factor that is also estimated in the regression. The comprehensive panel data yields a total of 2256 observations. As mentioned above, the Gini coefficients provided by the Solt database act as a passive variable. In the regression, this is incorporated by 100 imputations. All explanatory variables have been scaled such that they can range from 0 to 1 in order to ensure comparability of the individual coefficients in the regression. 3.3 Data description A first glance at the data shows the strong correlation the level of development has with the level of urbanization: Generally, more developed countries are also more urbanized, as seen below for cross-country data in 2014 (image 2), where the red line represents the OLS-estimated fit: Image 2 Three countries seem to be exceptional: Djibouti is highly urbanized despite a low level of development, whereas Trinidad and Tobago and Liechtenstein experience the opposite. All three can be considered very small countries and have extraordinary geographical positions that might explain the outliers. This might indicate that such features can influence the relationship observed and will be considered when conducting robustness checks for the regression. Likewise, there is a strong visual indication of the Kuznets curve, which can be seen in both cross-country data for the year 2014 (image 3) and panel data observation
Image 3 Image 4 Note, however, that the Gini indices above are merely the averages of the passive variable estimates for the purpose of visual representation and therefore do not reflect the uncertainty of those estimates. These can easily be seen below (image 5), where the 95% confidence intervals of each observation’s Gini estimate in the cross-country data of 2014 are visualized.
Image 5 The regression will take the standard errors of each individual observation into account. The data further show that the picture becomes less clear when looking at individual countries. Due to the limited time span covered, there is no single country going through the entire development process from least to highly developed. Only certain development intervals can therefore be considered for the longitudinal aspect of the panel data. As seen in image 5 for a selected number of countries, trends for increasing inequality for lower income countries and narrowing inequality for higher middle-income countries can nevertheless be seen. However, turning points, if existent, seem to be different for each state. China’s and Rwanda’s graphs in particular yield an insight in their individual Kuznets developments. Both seem to have just reached their “turning point”. In the case of China, this seems to support Kanbur’s and Zhuang’s (2013) as well as Zhang and Bao’s (2015) findings. Moreover, some nations display contrary developments in equality, especially the high-income countries in the graph’s sample, such as Switzerland and Japan, but also Pakistan, for instance. This supports the view by Simler and Dudwick (2010) that government policy and institutions can have substantial influence on the inequality in a country, apart from the mere relationship with development. As a consequence, policies should be devised that take factors into account that play a role in altering the Kuznets curve, one of which is likely to be
be included in the graph. The example of Rwanda in image 5 also shows the problem this data transformation faces especially for low levels of development, namely that the uncertainty is high for those observations. This shows the importance of their inclusion in the regression, making the input more reliable. Image 6 3.4 Hypotheses The regression is expected to change the shape of the Kuznets curve for high levels of urbanization. Following from the literature, higher urbanization is hypothesized to decrease inequality due to the narrowing trend in spatial inequality (World Bank, 2009), while the Kuznets curve is hypothesized to be true. Further, it is expected that the effect of urbanization will significantly interact with the Kuznets curve since urbanization is expected
to alter the course of the Kuznets curve1. Significant interaction would imply that the effect of urbanization depends on the level of development. From this, the following three hypotheses follow: A. 𝐻J: 𝛽? > 0 vs. 𝐻?: 𝛽? < 0 B. 𝐻J: 𝛽N ≠ 0 vs. 𝐻?: 𝛽N = 0 C. 𝐻J: 𝛽Q < 0 ∪ 𝛽R > 0 vs. 𝐻?: 𝛽Q > 0 ∪ 𝛽R < 0 Aside from the main regression, a few side regressions will be run as robustness checks with small changes to the model. 4. Results The results of the main regression (1) and the two auxiliary regressions (2) and (3) are presented in table 1. Using the results of regressions (1) and (2), the null hypothesis of A is rejected in favor of the alternative hypothesis. The coefficient for the percentage urban is significant at the 0.1% level, showing a negative sign. Thus, urbanization itself decreases inequality in a country, ceteris paribus and without taking the interactive effect with the development level into account. In regression (2), this is only the case at the 1% significance level. This interaction effect is also significant at the 0.1%-level, which means the null hypothesis of B is rejected as well. The coefficient shows a negative sign. That is the case in both regressions (1) and (2). The higher the level of development, the more urbanization contributes to increasing inequality, i.e. urbanization decreases inequality less for higher levels of development and even increases it. This is the case both with a linear and a quadratic cross term. When all three terms that have been mentioned previously are included, the effects become insignificant at all significance levels. In both the main regression as well as the supplementary regressions, the linear HDI-term shows a positive, and the quadratic HDI-term a negative coefficient, all significant at the 0.1% level. This means the null hypothesis in C can be rejected, evidencing the Kuznets curve. In addition, all 1 An alternative specification of the interaction term of the urbanization level with a squared HDI term yields almost identical results. Hypothesis B continues to refer to the interaction
three regressions yield significant (0.1%) support for the inclusion of the control variable cpi.2 Furthermore, the regressions yield nearly identical values for the constant scaling factor. For the main regression, this entails a conditional variance of gini of approximately 0.05, implying a low dispersion generated by the regression. The variance is even smaller in regression (8), which is to be expected since the predicted mean for the dependent variable is 0.37 and thus further away from 0.5 (Paolino, 2001).
Table 1: Regression coefficients and average margins; dependent variable gini
(1) (2) (3) 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 -2.820*** -3.061*** -3.278*** (0.147) (0.168) (0.304) 𝑢𝑟𝑏𝑝𝑜𝑝 -0.955*** -0.473** 0.0526 (0.283) (0.158) (0.637) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖 1.627*** -1.619 (0.402) (1.877) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖N 1.302*** 2.504 (0.302) (1.411) ℎ𝑑𝑖 9.665*** 10.53*** 11.24*** (0.545) (0.673) (1.063) ℎ𝑑𝑖N -9.230*** -9.943*** -10.49*** (0.524) (0.641) (0.898) 𝑐𝑝𝑖 0.320*** 0.305*** 0.294*** (0.0576) (0.0582) (0.0596) 𝑆𝑐𝑎𝑙𝑒 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 4.013*** 4.014*** 4.015*** (0.0403) (0.0404) (0.0404) N 2256 2256 2256
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
The sizes of the coefficients imply that the lion’s share of the regression’s explanatory value lies in the two terms representative of the Kuznets relationship. This can be exemplified using table 2, where the main regression and the auxiliary regressions have been run with the variables centered to their means, resulting in regression (4) for the centered main regression, and (5) and (6) for the centered auxiliary regressions, respectively3. This makes the coefficients more descriptive for a country with an average share of urban population and level of development. Using the main regression, the predicted Gini coefficient for a country with average values for all independent variables is 0.44. For such a country, which would classify as being between medium and high development according the United Nations (n.d.) due to its HDI of just below 0.7, the dependent variable is predicted to increase for higher levels of urbanization and to decrease for lower levels, ceteris paribus. This relationship can be seen for all levels above that development threshold, where higher urbanization is associated with higher inequality. For countries with lower development, this relationship is reverse and a higher share of urban population is associated with lower inequality. The nature of this approximate turning point shows in the centered regression (4), where a change in the level of urbanization does not significantly affect inequality at mean levels, ceteris paribus. For development levels above this, the effect becomes increasingly positive, whereas it becomes negative for below. This can be seen by the highly significant negative effect in the uncentered regression (1) for a hypothetical (but unrealistic) country with zero values for development und urban population. It also demonstrates that the level of development is pivotal in determining the effect of urbanization on inequality by showing strong interaction. The effect of urbanization is thus clearly present, discernable through the high significance of its coefficients, but rather small. It is rather surprising that the interaction term is positive on such a high level. The implication is that for a great part of observed human development, and especially in highly developed countries, urbanization actually contributes to rising inequality, which stands in contrast to what has been expected.
Table 2: Regression coefficients and average margins; dependent variable gini; variables
centered to their respective means
(4) (5) (6) 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 -0.239*** -0.234*** -0.242*** (0.0133) (0.0132) (0.0133) 𝑢𝑟𝑏𝑝𝑜𝑝 0.0990 0.107 0.0543 (0.0536) (0.0568) (0.0590) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖 1.627*** 1.624*** (0.402) (0.401) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖N 2.537 2.504 (1.425) (1.411) ℎ𝑑𝑖 -1.376*** -1.538*** -1.436*** (0.102) (0.101) (0.108) ℎ𝑑𝑖N -9.230*** -7.567*** -9.079*** (0.524) (0.355) (0.529) 𝑐𝑝𝑖 0.320*** 0.326*** 0.294*** (0.0576) (0.0591) (0.0596) 𝑆𝑐𝑎𝑙𝑒 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 4.013*** 4.002*** 4.015*** (0.0403) (0.0396) (0.0404) N 2256 2256 2256
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 To further check for the robustness of the result above, it is reasonable to include the consideration that some countries have qualities that could distort the regression outcome. For the regressions in table 3, countries have been additionally assigned three potential identifiers that are suspected to play a role for how urbanization or development relate to a country’s income distribution. small indicates a country whose population was smaller than two million in 2016 (World Bank, n.d.), dev specifies data points with an HDI-value greater than 0.8 to classify as “very highly developed” according to the UN (n.d.), and trans refers to the economies in transition from socialist planning to market-based economies since the 1980s. These mostly include countries from the former Eastern Bloc and Yugoslavia, but also former planned economies in Asia and such economies not yet in transition.4 4 Specifically, the designated countries are: Albania, Armenia, Azerbaijan, Belarus, Bosnia
Small countries are often characterized by special features of the nation’s population and mostly pertain to island states in the sample. In most countries that small, the distinction between urban and rural becomes blurry and economies and diseconomies of agglomeration might fail to materialize in light of the lack of an extensive hinterland, thus possibly altering the urbanization effect. Considering regression (7) in table 3, this is indeed the case at a 0.1% significance level. Urbanization decreases inequality in small states, ceteris paribus. The other two factors are expected to interact with the way the level of development relates to inequality. Socialist planned economies are very equal by nature and thus follow a different development path. Even during transition, this may alter the Kuznets relationship. As the results show, however, there is no significant interaction effect despite the observation that being a transition economy does correspond to significantly lower inequality. It can be concluded that the outcome of regression (7) does not alter the conclusions from the main regression and presents the same relationships as above. Regression (8) considers the same beta regression for countries as identifiers, simulating fixed effects to capture time developments across nations. Taking the observed trend of a widening income gap within very highly developed nations into account, belonging to that group might range outside of the Kuznets process already (Piketty, 2015). Excluding all observations satisfying either small, trans, or dev does not yield a corroboration of the previous results. The observed effects regarding urbanization are flipped, and there is no significant result for the proposed Kuznets relationship. Hungary, Latvia, Lithuania, Kazakhstan, Kosovo, Kyrgyzstan, Laos, Macedonia, Mongolia, Montenegro, Moldova, Myanmar, North Korea, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Tajikistan, Turkmenistan, Ukraine, Uzbekistan, Vietnam; including a few observations of predecessor states. China is notably not included as the initiation of the transition process goes back as far as 1978. Otherwise, the specification used in this paper
Table 3: Regression coefficients and average margins; dependent
variable gini. (8) with fixed effects excluding observations satisfying any of the dummy categories
(7) (8) 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 -2.880*** -1.000*** (0.142) (0.232) 𝑢𝑟𝑏𝑝𝑜𝑝 -0.846** 3.900*** (0.273) (0.684) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖 1.221** -5.326*** (0.388) (1.068) ℎ𝑑𝑖 9.839*** -1.070 (0.533) (1.040) ℎ𝑑𝑖N -8.663*** 2.176 (0.514) (1.272) 𝑐𝑝𝑖 -0.0369 0.0544 (0.0552) (0.0917) 𝑡𝑟𝑎𝑛𝑠 -0.378** (0.130) 𝑠𝑚𝑎𝑙𝑙 0.229*** (0.0616) 𝑢𝑟𝑏𝑝𝑜𝑝×𝑠𝑚𝑎𝑙𝑙 -0.464*** (0.0955) ℎ𝑑𝑖×𝑡𝑟𝑎𝑛𝑠 0.0723 (0.176) 𝑆𝑐𝑎𝑙𝑒 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 4.234*** 6.624*** (0.0424) (0.0970) N 2256 1065
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 5. Discussion The regression in this paper can draw from rich data with a large number of observations. However, the panel data is highly unbalanced as it tries to bring together large amounts of data that are collected nationally and throughout time and thus miss some data points where this is not possible. This is especially true for the estimates of the Gini indices
and the levels of urbanization despite the data being quite comprehensive. For the latter, a more accurate representation could have been an agglomeration index to account for national differences in measurement. All variables involved might suffer from inaccuracies as they can only be seen as estimates. This is best possibly accounted for in the dependent variable due to the combination and standardization of multiple databases and the incorporation of the different estimates and their standard errors by Solt (2016), but the other variables possibly lack such great comparability. The Human Development Index and the Corruption Perceptions Index are both indices constructed to give a realistic, comprehensive and comparable measure of human development and corruption in a country, respectively. Their composition, however, is by definition imperfect and cannot be taken as exact either, making them possibly prone to systematic errors in measurement. It is also safe to assume that the missing data points do not occur at random. Especially for less developed countries and older measurements, data on all variables included are less available. This might imply a sample selection bias. The results of auxiliary regression (3) are striking in the sense that the effects hypothesized in A and B are highly insignificant, which stands in great contrast to the results of the main regression and regression (2). It is very likely that this is caused by excessive multicollinearity. The variable inflation factors for the explanatory variables must naturally be high due to their multiple inclusion in the regression, which renders them insignificant in the case of regression (3). Furthermore, the regression’s goal of identifying the urbanization effect that is not part of the Kuznets curve is only controlled for by one variable, as with the CPI the literature is unambiguous and the variable proves easy to include. There is, however, much research on breaking down the different causes and factors affecting a country’s inequality with an array of possible influencing variables that could not be included in the regression. In regression (7) however, this has partially been accounted for, yielding the same results. The Kuznets curve, on the other hand, makes no pretense of the fact that it does not describe a direct causal relationship but in fact combines different effects on inequality that change with the stage of development a country is in. Accordingly, this regression has made an attempt at showing that the curve does not sufficiently include urbanization effects. The
results offered thus open up room for a deeper analysis of the causes that takes urbanization into account. Nonetheless, the effect itself needs to be interpreted with caution: It is not entirely clear if causality only runs in one direction. Castells’ and Royuela’s (2012) analysis shows that within the development and urbanization processes, the income distribution itself can be a moderating factor. If one argues that the inequality in a state, that is greatly caused by the urban-rural gap, prompts rural-urban migration and thus influences urbanization, the causality is reversed and would alter the observed effect. Consequently, as often happens with analyses of data in social sciences, the regression might suffer from simultaneous causality bias. Another potential difficulty of panel data is the presence of cross-sectional dependence. This would primarily pose a problem for auxiliary regression (8) with fixed effects. In this case, it would imply that certain changes in the dependent variable for one country would be caused by an external effect that would likely induce an effect on other countries as well, hence implying a dependency throughout observations across countries. However, the external effect variable included in this paper’s regression, the CPI, is not likely to undergo sudden changes that could resonate across countries. Further, it has been concluded that a country’s income distribution is mostly very specific to that country, and is unlikely to directly depend on other countries’ income distributions. Despite the conceivably large amount of data and despite its panel structure, the data do not suffice to adequately analyze time developments. As could be seen in the discussion of the data, no country goes even through half of the proposed Kuznets process, and taking the large differences for each state into account, this could be a likely reason why regression (8) does not deliver satisfactory results. It might even be the case that especially in developed countries, inequality has been increasing in recent decades, as recent research suggests (Piketty, 2015). Thus, more research regarding the use and truthfulness of the Kuznets curve is necessary. Indeed, here a larger time horizon would be of use. It can be said, however, that despite the above limitations urbanization is likely to be a relevant factor for inequality in a country. Further research should look into how exactly this relationship works. Due to lack of data and scope limitations for this paper, the observed
relationship could not be put in econometrical context to the urban-rural gap, which the literature suggests bears the largest share of spatial inequality. Here, it would be particularly interesting to further investigate the role of within-region inequality along the process of urbanization. 6. Conclusion This paper analyses the effect of urbanization on the income inequality in a country using panel data across time and countries in the time period 1995-2015. The literature shows that a nation’s inequality can largely be predicted by the stage of development a country is in. Kuznets describes this relationship as an inverted U, commonly known as the Kuznets curve. Due to agglomeration effects and positive feedback, urban areas are shown to consistently fare better in terms of welfare and output than rural areas. The World Development Report 2009 and similar research investigate the effect of urbanization and urban concentration on spatial inequalities and the urban-rural income and consumption gaps. Young (2013) shows that this is the main way urbanization relates to inequality as most of a country’s spatial inequality, whose relevance is described by Williamson (1965), can be attributed to the urban-rural gap. The aim of this paper was to show the effect urbanization has on inequality that cannot be described by the Kuznets process. In order to do so, a beta regression was run with global panel data in the time period from 1995 to 2015. The results for urbanization were significant at the 0.1%-level in the main regression, implying a negative effect of urbanization on income inequality for countries with low development that diminishes and subsequently reverses with higher levels of development. The Kuznets curve could be corroborated for most specifications, and it was shown that the stage of development is decisive for the effect urbanization has on inequality. Despite having controlled for corruption and having employed robustness checks with both alternative regression specifications and the inclusion of the elements transition economies and small countries, the paper does not incorporate all potential factors influencing inequality in a country that are not described by the Kuznets process, of which there are many by nature. Furthermore, there are econometric limitations to this research including potential measurement errors, endogeneity, and sample selection bias.
This research shows that urbanization is a highly relevant factor when assessing a country’s income distribution. In spite of the fact that the underlying factors vary greatly among states and potential policies will have to be devised taking factors specific to the country into account, some general policy implications can be derived. It is advisable to take inequality, which research shows comes with many negative effects, into consideration when devising urban and spatial policies as those will have an effect on inequality. It can further be seen that higher urbanization is associated with lower inequality in less developed countries, implying that urbanization strategies for those countries may yield desirable results in terms of the distribution of income. Thus, despite the perception of large slums and urban pauperism in many of the countries classified as having a low level of development, urbanization potentially comes to the overall benefit of society, and it should even be considered to encourage it. On the other hand, the results also show the opposite relationship for developed and high-income countries, suggesting that in order to achieve low inequality urbanization should not be stimulated. One implication could be, for instance, to prevent rural flight by investing in rural development and connecting the countryside with agglomerations through infrastructure. Of course, other factors such as growth perspectives and case-specific dynamics need to be taken into consideration as well. Building up on this rather broad insight, further research will be necessary to address the specific nature of the effects urbanization has on inequality, especially as regards within-region inequality and its relationship with both the observed urbanization effect as well as with between-region inequality. In addition, it has also become apparent through the results of this research that despite the Kuznets curve being a strong statistical relationship, its shape is alterable and the Kuznets process can be altered by policies, e.g. those that influence factors such as urbanization. References Aguiar, M., & Bils, M. (2015). Has Consumption Inequality Mirrored Income Inequality? The American Economic Review, 105(9), 2725-2756. Arthur, W. (1990, February). Positive Feedbacks in the Economy. Scientific American, 262(2), 92-99. Billger, S. M., & Goel, R. K. (2009, November). Do existing corruption levels matter in controlling corruption?: Cross-country quantile regression estimates. Journal of Development Economics, 9(2), 299-305.
Bourguignona, F., & Morrisson, C. (1998). Inequality and development: the role of dualism. Journal of Development Economics, 57(2), 233-257. Brown, M., & Uslaner, E. M. (2002). Inequality, Trust, and Political Engagement. Annual Meeting of American Political Science, 29. Boston. Calì, M. (2008, June). Urbanisation, inequality and economic growth: Evidence from Indian states and towns. Background note for the World Development Report 2009. London. Castells, D., & Royuela, V. (2012). Agglomeration, Inequality and Economic Growth: Cross-section and panel data analysis. Regions in Motion - Breaking the Path (pp. 21-25). Bratislava: 52nd Congress of the European Regional Science Association. Crenshaw, E. (1993, March 1). Polity, Economy and Technoecology: Alternative Explanations for Income Inequality. Social Forces, 71(3), 807–816. Davis, J. C., & Henderson, J. (2003, January). Evidence on the political economy of the urbanization process. Journal of Urban Economics, 53(1), 98-125. Ferrari, S., & Cribari-Neto, F. (2004). Beta Regression for Modelling Rates and Proportions. Journal of Applied Statistics, 31(7), 799-815. Gupta, S., Davoodi, H., & Alonso-Terme, R. (2002, March). Does corruption affect income inequality and poverty? Economics of Governance, 3(1), 23–45. Gustafsson, B., & Shi, L. (2002, October 1). Income inequality within and across counties in rural China 1988 and 1995. Journal of Development Economics, 69(1), 179-204. International Monetary Fund. (2000). World Economic Outlook: Focus on Transition Economies. Kanbur, R., & Zhuang, J. (2013, March). Urbanization and Inequality in Asia. Asian Development Review, 30(1), 131-147. Kawachi, I., Kennedy, B. P., Lochner, K., & Prothrow-Stith, D. (1997, September 1). Social capital, income inequality, and mortality. American Journal of Public Health, 87(9), 1491-1498. Kuznets, S. (1955, March). Economic Growth and Income Inequality. The American Economic Review, 45(1), 1-28. Lewis, W. (1954). Economic Development with Unlimited Supplies of Labour. The Manchester School, 22(2), 139-191. Li, H., Squire, L., & Zou, H.-f. (1998, January). Explaining International and Intertemporal Variations in Income Inequality. The Economic Journal, 108(446), 26–43.
Li, S., & Wang, W. (1994). The difference in fertility between urban and rural areas and its impact on the process of urbanization. Chinese Journal of Population Science, 6(2), 201-210. Martin, D., Wilson, M., & Vasdev, S. (2001, April). Income inequality and homicide rates in Canada and the United States. Canadian Journal of Criminology, 43, 219-236. Moomaw, R. L., & Shatter, A. M. (1996, July). Urbanization and Economic Development: A Bias toward Large Cities? Journal of Urban Economics, 40(1), 13-37. Mpando, L. R. (2000). Fertility levels and trends. National Statistical Office. Malawi Demographic and Health Survey. Nielsen, F., & Alderson, A. S. (1997, February). The Kuznets Curve and the Great U-Turn: Income Inequality in U.S. Counties, 1970 to 1990. American Sociological Review, 62(1), 12-33. Oishi, S., Kesebir, S., & Diener, E. (2011, August 12). Income Inequality and Happiness. Psychological Science, 22(9), 1095-1100. Paolino, P. (2001). Maximum Likelihood Estimation of Models with Beta-Distributed Dependent Variables. Political Analysis, 9(4), 325-346. Piketty, T. (2015, May). About Capital in the Twenty-First Century. American Economic Review, 105(5), 48-53. Rohwer, A. (2009). Rohwer, Anja. "Measuring corruption: a comparison between the transparency international's corruption perceptions index and the World Bank's worldwide governance indicators. CESifo DICE Report, 7(3), 42-52. Simler, K., & Dudwick, N. (2010). Urbanization and Rural-Urban Welfare Inequalities. World Bank, Poverty Reduction and Equity Unit. Washington, DC: http://siteresources.worldbank.org/INTPOVERTY/Resources/060310_Simler_Dudwic k_Rural-Urban_Welfare_Inequalities.pdf. Solt, F. (2016, May 31). The Standardized World Income Inequality Database*. Social Science Quarterly, 97(5), 1267–1281. Transparency International. (n.d.). Corruption Perceptions Index. Retrieved from https://www.transparency.org/research/cpi/overview United Nations. (n.d.). Human Development Data (1990-2015). Retrieved January 2018, from United Nations Development Programme Human Development Reports: http://hdr.undp.org/en/data United Nations, Department of Economic and Social Affairs, Population Division. (2014). World Urbanization Prospects: The 2014 Revision, custom data acquired via website.
Williamson, J. G. (1965, July). Regional Inequality and the Process of National Development: A Description of the Patterns. Economic Development and Cultural Change, 13(4), 1-84. World Bank. (2009). World Development Report 2009: Reshaping Economic Geography. World Bank. World Bank. (n.d.). Population, total. Retrieved January 2018, from World Bank Indicators: https://data.worldbank.org/indicator/SP.POP.TOTL?end=2016&start=2016 World Bank. (n.d.). Urban population (% of total). Retrieved January 2018, from World Bank Indicators: https://data.worldbank.org/indicator/SP.URB.TOTL.IN.ZS?end=2016&start=1960&typ e=shaded&view=chart&year=2016 Young, A. (2013, November 1). Inequality, the Urban-Rural Gap, and Migration. The Quarterly Journal of Economics, 128(4), 1727–1785. Zhang, W., & Bao, S. (2015, July). Created unequal: China's regional pay inequality and its relationship with mega-trend urbanization. Applied Geography, 61(Spatial Inequality), 81-93. Appendix A. Table 1: Means and standard deviations of the independent variables
Mean Standard deviation
𝑢𝑟𝑏𝑝𝑜𝑝 56.43677 25.7166 ℎ𝑑𝑖 64.78399 16.77971 𝑐𝑝𝑖 43.09039 21.84638 N 6241
B.
Table 2: Regression coefficients and average margins; dependent variable gini
(1) (2) (3) 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 -1.770*** -1.956*** -2.057*** (0.105) (0.121) (0.203) 𝑢𝑟𝑏𝑝𝑜𝑝 -0.936*** -0.542*** -0.277 (0.224) (0.127) (0.451) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖 1.391*** -0.846 (0.326) (1.351) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖N 1.155*** 1.799 (0.252) (1.045) ℎ𝑑𝑖 6.425*** 7.132*** 7.479*** (0.405) (0.509) (0.751) ℎ𝑑𝑖N -6.421*** -7.044*** -7.325*** (0.403) (0.506) (0.672) 𝑆𝑐𝑎𝑙𝑒 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 3.829*** 3.830*** 3.830*** (0.0368) (0.0368) (0.0368) N 3393 3393 3393
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
C.
Table 3: Regression coefficients and average margins; dependent variable gini; variables
centered to their respective means
(4) (5) (6) 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 -0.322*** -0.320*** -0.325*** (0.00992) (0.00994) (0.00994) 𝑢𝑟𝑏𝑝𝑜𝑝 -0.0349 -0.0494 -0.0695 (0.0479) (0.0510) (0.0515) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖 1.391*** 1.486*** (0.326) (0.327) 𝑢𝑟𝑏𝑝𝑜𝑝×ℎ𝑑𝑖N 0.902 1.799 (1.043) (1.045) ℎ𝑑𝑖 -1.109*** -1.147*** -1.173*** (0.0745) (0.0848) (0.0847) ℎ𝑑𝑖N -6.421*** -4.985*** -6.309*** (0.403) (0.253) (0.406) 𝑆𝑐𝑎𝑙𝑒 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 3.829*** 3.819*** 3.830*** (0.0368) (0.0366) (0.0368) N 3393 3393 3393
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
