Does democracy reduce income inequality?

Boix (2003) argues that income inequality inhibits democratisation and can affect the consolidation of established democracies. Houle (2009), on the other hand, finds that income inequality has no systematic effect on democratisation, although it can destabilise the already established democracies.

While many factors contribute to a reduction in income inequality, one should intuitively expect that an extension of civil liberty and political rights to the underprivileged people who were previously excluded from these rights can lead to a more egalitarian distribution of income in a society. A continuing denial of these rights can trigger movements for democracy as evidenced by the break-up of the former Soviet Union, the Arab Spring, and regime changes in many Latin American countries.

The country-level Gini coefficients cannot be aggregated into a global coefficient, although a coefficient can be computed for the aggregate. This is because the type of data and methods used for different countries may not be identical. Gini coefficients are also not unique, as income inequality can be measured in many ways. An anonymous referee rightly notes that the Gini index can decline when redistribution of income helps some poor move up to become middle class and the elite loses, while inequality among the rest of the population may in fact be rising. Similarly, the Gini index for a developing country can rise due to increasing income inequality, while the number of people in absolute poverty decreases. Thus, the shape of the distribution of income inequality, as indicated by the coefficient of variation, provides a better picture of inequality among people than a uni-dimensional measure of inequality.

Redistribution is usually measured by government tax revenues (%GDP), but tax revenues do not include grants, an important channel through which income inequality can be minimised. Other alternative redistribution measures can be expenditure on social protection and welfare (%GDP), public spending on education, and health care. In this paper, I use total revenues including grants (%GDP) as a measure of redistribution. Note that there is a good deal of controversy regarding the redistributive effect of any particular measure, or for that matter the overall welfare state. Berg (2005) rightly concludes that the commonly used method of comparing the pre-fisc income distribution with the post-fisc distribution cannot correctly describe the effect of any redistribution scheme in a welfare state.

Another alternative source of information on political freedom is Polity IV (Political Regime Characteristics and Transition), where the level of democracy/autocracy is indexed on a discrete scale from \(-\)10 (strongly autocratic state) to +10 (fully democratic state). The authors of the project (Project IV 2000) describe ‘anocracy’ with a score from \(-\)5 to +5 or any of the three special values, \(-\)66, \(-\)77, and \(-\)88, referring to foreign interruption, interregnum or a collapse of the central authority, and a period of transition from anarchy to constitutional democracy, respectively. The special values are converted to standard polity scores by treating (\(-\)66) as a missing value, (\(-\)77) equal to a polity score of ‘0’, and (\(-\)88) prorated across the span of transition. There is a good deal of controversy regarding the quality of the existing freedom indices, and they are found to differ in terms of conceptualisation, measurement, and aggregation (Munck and Verkuilen 2002). Although Gastil and polity indices are highly correlated, I nevertheless use the Gastil index in the estimation of the model and perform sensitivity analysis of the results, with the Polity index as an alternative measure of freedom level. The index is normalised by (polity score + 10)/20 to measure freedom level on a continuous scale.

A turning point in the nonlinearity occurs at a freedom level \(F^{*}= 1/2 (7.719)/(6.102) = 0.6325\), which exceeds the average freedom level of 0.6267 for all countries in the sample.