Data

The inequality of income is an important factor in study of numerous social phenomena. Since the seminal work of Klaus Deininger and Lyn Squire (Deininger and Squire 1996), the original authors and their followers at the World Bank and the UN University have been collecting literally thousands of studies. But when it comes to using these data, sample sizes shrink radically due to time differences and due to the large variability of measurement specifications. In the literature, researchers are more or less cautious regarding to these differences.

We follow the idea that both time trends and specification differences exhibit stabilities which allow from one measurement at a point in time to estimate inequality at another point in time and from inequality measured using one specification to estimate inequality as it had been measured using another specification.

(Upload of the complete paper and data follows.)

Data on economic differentiation related to the structure of production are not available for a global sample. The OECD covers its member countries, which show little variance in democracy levels; other data are scarce. Luckily, there is a helpful proxy available. Though there are no data on production structure, there are data on the export structure available in the trade database by Robert Feenstra. This database provides values for bilateral trade flows to and from roughly 200 territorial entities, disaggregated in 1263 4-digit Standard International Trade Codes (SITC4).

We aggregated these data on the SITC1-level, obtaining ten groups from food (0) over coal and fuels (3) to pre-products (6) and end-products (8). Heterogeneity between groups was calculated as the standard deviation of logs between export shares of these ten groups, where the share value was replaced with 0.01\% if it was below that value or zero. To obtain a measure with the right direction and interpretable scale, the sign was reversed and the scale adjusted to have 1 for the maximal value (Greece 1998) and 0 for the minimum value (Uganda 1991).

Though the SITC1-groups define well distinguishable categories of goods, the terms of production in the respective industries are different, leaving room for economic differentiation within the ten groups. We roughly grouped the ten SITC-1-groups into three classes (see column 3 in Table \ref{tabheterogeneity}): one containing four SITC-1-groups such as coal and fuel which were considered to be described by high concentration and low differentiation levels in qualification to make their impact on democratization negative; one containing three SITC-1-groups that were considered neutral; and one containing the SITC-1-groups of machinery, end-products, and `others', which were considered to be so strongly differentiated internally that we expected a positive effect from their relative share of total exports (as a proxy for total production) on the evolution of political institutions. The construction of the variables leads to a heteroskedastic relation: from perfect between-groups heterogeneity, a value of -.1 for within-groups heterogeneity would result.

(Upload of the complete paper and data follows.)