On the ambiguous economic freedom–inequality relationship

Section 2 provides a detailed review of this literature, which includes papers by Apergis and Cooray (2015), Berggren (1999), Scully (2002) Carter (2006), Clark and Lawson (2008), Bergh and Nilsson (2010), Sturm and De Haan (2015) and Bennett and Cebula (2015).

The Deininger and Squire (1996) database contains inequality measures representing different income and household concepts. These conceptual differences are believed to produce systemic differences, and considerable effort has been exerted to construct more comparable inequality datasets in recent years (e.g., Atkinson and Brandolini 2001; Deininger and Squire 1996; Solt 2014).

This approach common in the literature, but has also been subject to criticism (Atkinson and Brandolini 2001; Pyatt 2003).

Growth is predicted by a regression of actual growth on initial GDP and a set of geographic and investment variables.

The EFW index used by Clark and Lawson (2008) excludes regulation due to scarcity of data.

Carter (2006) also includes Gini coefficients derived from different income concepts and like Scully (2002), controls for a set of concept indicator variables.

Bennett and Vedder (2013) used US state-level data to estimate a quadratic inequality-economic freedom model and found the relationship to exhibit an inverted U-shape, opposite Carter’s findings; however, results from the two studies are not directly comparable due to differences in the samples and composition of the economic freedom indices. Differences between country and subnational economic freedom measures are discussed in more detail below. See footnote 12. In a forthcoming paper, Apergis and Cooray (2015) also find evidence of an inverted U-shaped economic freedom–inequality curve for a panel of 138 countries using cointegration techniques.

SWIID version not indicated in Bergh and Nilsson (2010), but confirmed in email correspondence with Therese Nilsson on June 23, 2015.

Bergh and Nilsson (2010) also find a positive and marginally significant relationship between EFW and gross income Gini coefficients. In addition, they obtain null results when employing a dynamic model to examine the effect of long-run changes in EFW on changes in inequality, as well as when utilizing GMM estimation.

The argument that government redistribution via the government sector and inflation does not impact market incomes assumes that redistribution is not distortionary and takes place without economic cost. Redistributionary policies often distort investment, labor-leisure and allocation decisions such that it is highly plausible that market incomes, and hence gross income inequality, would differ in the absence of such redistribution, although admittedly, it is not possible to measure incomes in such a counterfactual world. Regarding the use of his dataset, Solt (2014: 21) adds that “Market-income inequality, although accurately described as measuring the distribution of income before taxes and transfers are taken into account, cannot be considered ‘pre-government’: a wide range of non-redistributive government policies, from public education and job-training programs to capital-accounts regulations, also shape the income distribution. In addition to such market-conditioning policies, market-income inequality also includes the feedback effects of redistributive policies on household’s decisions regarding savings, employment, and retirement.”

Sturm and De Haan (2015) also found that the effect of economic freedom on income redistribution, as measured by the difference between net and gross income inequality, is conditional on fractionalization.

The EFNA index only accounts for heterogeneity among the US states in three areas: government expenditures, government revenues, and labor market policies (Stansel and McMahon 2013). National institutions measured by the EFW index such as the regulatory environment, monetary policy, international trade policy, and legal institutions are relatively homogenous across subnational regions. These macro-level institutions may nonetheless influence the distribution of income such that results pertaining to the relationship between subnational economic freedom and inequality are not directly comparable to those resulting from the study of county-level economic freedom and inequality because the margins by which institutions and policies are operating at the subnational and national level differ.

The Heritage Foundation provides an alternative economic freedom measure, the Index of Economic Freedom (IEF). The EFW measure is used in this study for several reasons. First, the EFW data are available for a large number of countries as far back as 1970, while the IEF data are only available since 1996. Second, one of the objectives of this study is to replicate the results of existing literature and all of the cross-country studies on economic freedom and inequality have utilized the EFW data. Third, the majority of the empirical economic freedom literature uses the EFW and not the IEF data. Hall and Lawson (2014), for instance, document that hundreds of scholarly articles have been published in academic journals that cite the EFW data. Analogous system GMM results using the Heritage data are presented, however, in Table 9 of Appendix.

Assuming that income is progressively redistributed and welfare programs are designed to not discourage work effort from transfer recipients. If the latter condition does not hold, then the net effect on an individual’s income could be neutral. Although the transfer would still reduce income inequality, it would do so by further increasing IO if the transfer recipient reduced his/her labor hours.

As discussed by Bennett and Vedder (2015), the relationship between educational attainment and income inequality is theoretically ambiguous and empirical evidence is mixed.

The theoretical result that redistribution reduces inequality is not completely generalizable either. Berggren’s framework hinges on income being redistributed from the rich to the poor at no economic cost, assumptions which may not hold in practice. “Director’s law,” for instance, suggests that the median voter will choose to redistribute resources to the middle class and not the poor (Stigler 1970). Additionally, the rent-seeking literature suggests that well-organized interest groups have an incentive to lobby for subsidies that result in regressive redistribution. Tax-and-transfer policies can also potentially distort labor market decisions on both the supply and demand sides, altering individual incomes.

The R\(^{2}\) values reported in Sect. 4 for estimates based on the SWIID version 5 data are computed using the Rubin (1987) combination rule and utilize Fisher’s z transformation over the imputed data.

Gini coefficients take a value ranging from zero (resources equally distributed over population) to one (one person or household possess all resources). It is a relative inequality measure and there is no direct mapping between the underlying income distribution and the Gini coefficient.

SWIID author Frederick Solt confirmed in an email on June 4, 2015 that this was the appropriate way to compute longer term averages of the data series for multiply imputed estimation.

Sensitivity of the results from the cross-sectional models used by Berggren (1999) and Clark and Lawson (2008), as well as the results from the 2SLS model used by Scully, is not considered here, primarily because of methodological issues with the models they utilize, as described in Sect. 3, but also because of space limitations. In results not reported, we find that the Berggren cross-sectional and Scully 2SLS models are also quite sensitive to the time period, inequality measure, and sample of countries examined. Sturm and De Haan (2015) and Bennett and Cebula (2015) both found a null result, so the sensitivity of their models are not considered here either. The study by Apergis and Cooray (2015) was released after the current study was completed and it includes a number of sensitivity tests.

Covariates include the following and their square: real GDP per capita, political rights, civil liberties, share of population living in urban area, average years of schooling of adult population, shares of population under age 15 and above age 65, shares of labor force employed in the industrial and service sectors.

The other inequality datasets employed in this analysis do not have observations for some of the countries used by Carter (2006).

It is acknowledged that some p values from the Hansen specification test in the results reported below are close to 1.0, suggesting potential over-identification. See Roodman (2009) for a discussion of over-identification in the context of System GMM. It is also acknowledged that the p value of the ar(2) test in some estimates is less than 0.1, suggestive that instruments lagged t−2 are invalid and an additional lag should be introduced with t−3 lagged instruments used. Because the results are generally null, there are a limited number of time periods, and the objective of this analysis is to estimate comparable coefficients for each of the various coefficients, additional results addressing these issues are not reported.

The analysis for this paper executed in Stata. To the best of our knowledge, there is not a known method to apply system GMM analysis using MI data. For the regressions utilizing the SWIID version 5 inequality measures, we use the average of the 100 imputations for each observation.

The control variables include: log of real GDP per capita (LRGDPL), the shares of labor employed in the industrial (INDUSTRY) and service (SERVICE) sectors, the price of investment goods relative to the US (PDISTORT), the fertility rate (FERTILITY), the average years of secondary education of males (EDUCF) and females (EDUCF), the dependent to labor force ratio (DEP2LABOR), and the share of population residing in an urban area (URBAN). See Table 7 of appendix for more information.

Because the Heritage data are only available since 1995, there are fewer periods in the dataset. To preserve sample size, contemporaneous values of the independent variables are used, whereas Tables 5 and 6 use lagged values of the independent variables.

ATG data available at http://​data.​worldbank.​org/​data-catalog/​all-the-ginis.