Publications  Income Distribution


 With the Gini coefficient and the share of the poorest 40 percent of the population as dependent variables, we tested the following hypotheses:

(1)  The Kuznets Curve does not exist. That is, the level of per capita income has no effect on income distribution, once other relevant factors are taken into account.

(2)  Even if the Kuznets Curve exists, the relationship between per capita income and income distribution is not stable over time.

(3)  Differences in socio-political systems are much more important than per capita income in explaining cross-country variations in income distribution. It will be more egalitarian in countries that are Communist, or suffer extensive government intervention in the economy, or have no dualistic socio-political structure.

(4)  Spread of education leads to greater income equality.

(5)  Rate of growth does not affect income distribution.

(6)  Structure of the economy, especially the relative importance of  primary and manufactured exports, is a major factor in income distribution.

(7)  There are no systematic differences in income distribution among the major regions of the world, once such explanatory variables as socio-political systems or education are included in the analysis.


 Gini coefficients of income distribution data were found for 83 countries. For 39 countries, observations for more than one year were available, resulting in 145 observations in total. For the share of the poorest 40 percent, the respective figures are 80 countries and 136 observations (see Appendix for list of countries). The data span the post-World War II years from 1952 to 1976, but are concentrated in the period from 1955 to 1971. Since the Kuznets Curve describes changes over time, it is reasonable to use several observations from a single country whenever income distri­bution data are available for several years. The basic source is Jain  [10], supple­mented by others listed in the Appendix.

 Income distribution data are notoriously unreliable. The data used here, drawn from a variety of sources, suffer from all the defects common to the breed. However, we have statistically tested the influence of outliers on the results (see the Metho­dology section below) and found only two sets of outlier data which have much influence (Taiwan and Pakistan) and only one which seems implausible. The share of the poorest 40 percent for Pakistan is quite inconsistent with Pakistan's Gini coeffi­cient and even more inconsistent with the shares reported for neighboring countries with similar characteristics and per capita income. Indeed, Pakistan's share is double that of comparable countries.3 Given our doubts about these particular figures, regressions in this paper for the share of the poorest 40 percent exclude Pakistan.4 Note that if we had included Pakistan's share data our hypothesis on the Kuznets Curve would have been more strongly supported - indeed the Curve would have completely disappeared. That the results including Pakistan are quite inconsistent for the Gini and for shares suggests that shares data on Pakistan represent a "bad"' outlier.

 Inclusion of Taiwan reverses the signs of the purely intertemporal Kuznets Curve (Table 1) but in the case of the combined cross-country / inter-temporal curve, it only weakens the Kuznets Curve effect. Moreover, while Taiwan is an outlier, the underlying data are plausible. They show a sharp improvement in income distribution as per capita income rose, but that is precisely what historical studies of Taiwan's experience have also shown. Therefore it seems reasonable to include Taiwanese data in the analysis. Results excluding Taiwan are available from the authors.

 Outliers and their influence on the results are discussed further below.

 Variables. For income distribution data, different sources use different definitions and differ in the populations covered, e.g. the whole country, rural or urban areas; population, households, income recipients or the economically active. Such differences in definition or coverage could influence the results. Ideally, separate regressions should be run for each definition, but there are not enough observations for some definitions. Moreover, the results of such independent regressions would hardly be comparable. We therefore made the simplifying assumption - not implausi­ble in our view - that the differences in definitions and coverage of income distribution data influence only the intercepts, not the slopes, of regression curves. This means, for instance, that while we allow for differences in inequality between rural and urban areas, we assume that this difference is identical at various levels of per capita income, or for different levels of education etc.5 The assumption allowed us to reduce possible bias from ignoring definitional differences, by introducing a set of corrective dummy variables. The coefficients and t-statistics for these definitional variables are quite stable and are not of great interest. They are therefore not reported.6

 Dependent variables are the Gini coefficient and the share of the poorest 40 percent, as measures of income inequality. Alternative indexes were chosen because interest in income distribution has focused on both the shape of distribution and the absolute income and income share of the poor. The main explanatory variable was per capita income in the 1964 U.S. dollars. The Kuznets Curve is defined as the quadratic function of the log of per capita income, perfectly standard for studies of the Kuznets Curve.7

 The time variables were introduced to capture any shift in the curve. The interaction of the time variable with the log of income and with the square of the log of income was to capture any changes in the slope (flattening) of the Kuznets Curve.

 Dummy variables distinguished the Communist countries of Eastern Europe and countries with a dualistic socio-political structure. To be defined as dualistic, the elite had to be a minority and ethnically different from the majority of the population. Iran is not classed as socio-politically dualistic, although the economy is dualistic, because the elite is indigenous, but Gabon is, because of the foreign (French) role in the society and economy for the year concerned. Judgements may differ on the classification of some countries (see Appendix for list).

 The extent of government intervention in non-communist countries is mea­sured by the share of public investment in total investment. We considered this a more suitable index than stated ideology since some governments call themselves socialistic but rely heavily on the market and private enterprise, while a few pro­claim their devotion to private enterprise but intervene massively in the economy. This index is flawed, since government can intervene as effectively by controlling private actions as by expanding the size of the government sector. But no index exists to measure the extent and effectiveness of controls. Other proxies are even more flawed than the one we used. For instance, the share of government expendi­tures in GNP is dominated in some countries by the size of military expenditures. By that measure, for instance, the U.S. appears far more interventionist than Japan, contrary to reality. The share of public in total investment also seems to be broadly correlated with the degree of control over the private economy. We therefore con­sidered it the most suitable quantitative index available.

 Education was measured by the proportion of children in primary and second­ary school, combined into a weighted index, the same variable used for other studies. To take account of lags we have used participation rates for roughly five years before the year of the income distribution data.

 To test the effect of economic structure, the share of primarily and manufac­tured exports in national income seemed more appropriate than the share of primary or manufactured exports in total exports. One would expect little effect on income distribution if total exports are 5 percent of GNP, even if the share of primary or manufactured exports is 90 percent of total exports.

 Economic growth was measured by the mean rate of growth in GDP for the five years preceding the year for which income distribution data are available, to take account of inevitable lags in the effects of growth-enhancing policies on income distribution.

 Since regional (dummy) variables presumably stand for a variety of not clearly defined historical, social, political and economic factors which groups of countries have in common, we attempted to define regions that were not only contiguous, but also showed some other attributes. So, for instance, North Africa was combined with West Asia on the assumption that shared ethnicity, religion, history and social characteristics were more important than geographic definition. The reference region included Western Europe and the developed areas of European settlement (North America, Australia and New Zealand).


OK economics

Last update Tuesday, August 24, 2004

Designed and maintained by Oldrich Kyn - Copyright © 1997-2004.  All rights reserved. Copyrights of signed articles, graphics, sounds and programs is held by the original authors.