Social Capability, History and the Economies of Communist and Postcommunist States

On the economic system of the planned economy era, see Kornai (1992). On the economics of transition, see Roland (2000) and Campos and Coricelli (2002). For a recent analysis of the “initial conditions” issue, see Godoy and Stiglitz (2006).

An interesting analysis relying on archeological evidence to demonstrate the similarity of the progression from village to state to empire in the New World with that which occurred independently and earlier in the Old World is Sanders and Marino (1970). Recent anthropological discussions using a social evolutionary framework include Johnson and Earle (1987) and Shifferd (1987).

Hibbs and Olsson (2004) conducted a direct test of Diamond’s hypothesis. They constructed a measure of “biogeographical” initial conditions based on the identified existence in the wild, in 6 world regions, of (a) large-seeded grasses from which grains could be domesticated, and of (b) precursors of domesticated animals such as horses, sheep, and pigs. After demonstrating that the biogeography measure accurately predicts the known time of transition from hunting-and-gathering to agriculture in these regions, they showed that the transition year itself is capable of predicting 53% of the variance in log 1997 per capita income among 112 present-day countries and 43% of the variance of a much-used measure of “institutional quality” (the ICRG country risk index), and that the transition year, an index of geography, and the institutional quality measure together predict 80% of the variance in log 1997 income.

The term “socialist” is used here and in the remainder of this article in the sense applied by ruling Communist parties in the Soviet Union and similar states, where state or “social” ownership of the major means of production was a defining feature. Western “social democracy” is accordingly excluded.

In China’s case the proportion of urban and educated was not large, but absolute numbers were.

An extended discussion of how greater levels of preindustrial development might facilitate modern economic growth is in Putterman (2000).

For details, see Putterman (2004). In past work by the authors cited, results were robust over a wide range of discount rates.

The correlation between population density of 1500 and that of 1950 is 0.74 for the 47 countries in our full sample and 0.82 for our 31 core Eurasian Communist countries (see below). Both correlations are significant at better than the 0.0001 level. That the two are so highly correlated raises the possibility that population density favorably influences economic growth simply because it facilitates transportation, trade, and specialization, rather than because it is proxying for social capabilities built up in the course of long-term development. We use the very early density indicator partly to reinforce our interpretation of the variable as an indicator of early development, but that interpretation is most importantly buttressed by the parallel and often stronger results for state history.

The data can be found in Scherer 1984. The correlation is significant at the 0.1% level. Interestingly, the correlation between the 1897 literacy figures and the growth rates of the 15 former republics during the transition years 1991–2002 is positive and significant at the 5% level. The included former republics are Armenia, Azerbaijan, Belarus, Estonia, Georgia, Kazakhstan, Kyrgyzstan, Latvia, Lithuania, Moldova, Russian Federation, Tajikistan, Turkmenistan, Ukraine, and Uzbekistan.

Diamond 1998: 306. A countervailing principal of linguistic differentiation might work in the opposite direction: one usually finds more distinct members of a family of languages close to its place of origin than in places of its late dissemination (see Diamond’s discussions of the origins of the Austronesian language group and of the Bantu language family). And there have been numerous empires that lacked linguistic unity. Overall, however, correlations support the early development/linguistic unity conjecture. The correlation between statehist and the index of ethnolinguistic heterogeneity used by Easterly and Levine, for 98 developed and developing countries, is − 0.21, significant at the 5% level (Chanda and Putterman 2005).

Our variable ethnic is the ethnic fractionalization index taken from Alesina et al. 2003, defined as the probability that two randomly selected people in a country will not belong to the same ethnic group.

Although Germany is noted for its late aggregation into a modern nation-state in the nineteenth century, parts of its territory fell under Roman influence, and throughout medeival times, it was home to numerous kingdoms that lead to a “multiple kingdoms” rating and to a higher score on the state antiquity scale than in lands under strictly tribal rule, although a lower score than for lands under a unified and home-based kingdom or empire. For details on calculation of the state antiquity measure see Bockstette et al. (2002) and, for a more complete discussion, Putterman (2004).

Whereas there was a kingdom in Cambodia by the first century and an indigenous kingdom in Vietnam, following 2 centuries of Chinese control by the end of the second century, Laos had no state until it came under Khmer rule in the fifth century and developed indigenous kingdoms in the twelfth century.

One Asian country, Afghanistan, is also included in the non-core sample (see the Appendix). Although under Marxist rule for too short a period to enter any of our growth regressions, it is included in some reported correlations for the full sample. A second Asian Marxist state, South Yemen, had insufficient data from its Marxist period and subsequently merged with non-Marxist North Yemen; it appears in none of our computations.

In our sample, only Guinea Bissau exhibits substantial pre-European state influence, exerted by the ancient Mali empire. States existed in Africa (Ghana, Mali, the Buganda kingdom, etc.) and the Americas (e.g., the Aztec and Inca empires) before European contact, but not on territories later classed as Marxist states).

Island and coastal countries are disproportionately represented in the African and Latin American sample, and former Portuguese colonies in Africa are disproportionately represented among states experimenting with Marxist economic models. But these facts play no direct role in our analysis.

We also do not distinguish between more orthodox socialist economies and those countries, especially Yugoslavia and Hungary, which had more market-oriented economies before 1989.

The cost of focusing on this period is that it constitutes a small part of the socialist history of some of the countries involved, and may be unrepresentative in some cases due to the stagnation some experienced in the 1980s.

With respect to the observations that we use, Maddison’s data is identical to that in Groningen Growth and Development Centre Total Economy Database 2004.

Different adjustments are required for the different sources. The overall Soviet growth rate assumed by DeBroek and Koen differs from Soviet growth as reported in Maddison, so we rescale their republic growth rates by the ratio between Maddison’s and their growth rate for the USSR as a whole. Dedec (1996) has data for the percentage of national income generated in Czech lands and Slovac lands in 1970 and 1989. Combining this with population data for both republics (from Maddison), we split Maddison’s Czechoslovakia per capita GDP data for 1970 and 1989 into Czech and Slovak per capita GDP. Based on these estimates, we calculate separate per capita GDP growth rates for Czech and Slovak lands. For Yugoslavia Plestina (1992) shows the former republics’ and total Yugoslavia’s average Gross Material Product annual growth rates for 1966 to 1988. Using republic population data, we turn this into per capita growth rates and rescale them by the ratio of Maddison’s 1970 to 1990 per capita GDP growth rate estimate to Plestina’s 1966 to 1988 Yugoslav per capita product growth rate to achieve a comparable proxy for the individual republics’ 1970 to 1990 per capita GDP growth rates.

In place of the observations by former republic, this sample includes Maddison’s observations for the Soviet Union, Yugoslavia, and Czechoslovakia.

If den1500 is replaced by den1950, the correlation is significant at the 10% level for the full sample and at just short of the 10% level (p = 0.109) for the core country sample.

A more restricted estimate, not shown, for the 11 core socialist countries only, without republics, has roughly twice the absolute magnitude and is significant at almost the 5% level.

That is, the annual rate of growth is predicted to have been 2.34% in the Czech Republic versus 1.01% in Russia. If den1950 is used in place of den1500, similar results are obtained, with the coefficient on den1950 being significant at the 10% level in the full sample and at the 5% level in the core sample.

This result is not surprising because we are adding Third World socialist states with notoriously poor economic performance and substantial distances from Washington, Berlin, and Tokyo to the core sample for which the finding has already been ascertained. We also checked robustness by estimating the equation with only core countries, not republics, but the sample is small, with only 11 observations. The result is a still negative coefficient on distance, with a p-value a little shy of the 10% level.

Recall that the column 6 result uses data on a republic-by-republic basis. This makes the result all the more striking, as many of the observations are for fairly heterogeneous individual republics of the former Soviet Union and Yugoslavia.

As mentioned earlier, regressions paralleling those in Table 3 were also estimated with inclusion of a secondary school enrollment ratio measure for 1970. Unfortunately, this reduces the overall sample size to only 18 countries for the full and no republics samples (in this case identical, because republic-level data on the enrollment measure are not included in our sources) and only 7 for the core sample. The coefficient on enrollment is always positive but insignificant. The coefficients on distance, statehist, den1500, and ethnic have the same signs as in Table 2, and significance levels are similar but in several cases lower, dropping below the 10% level in the case of den1500.

Under robust regression, statehist becomes significant at the 5% level in all three samples.

All variants of the regressions are available at <http://​watsoninstitute.​org/​ped/​scid/​>.

The sample does not include observations for what were then republics of the Soviet Union and Yugoslavia, but instead it has observations for those countries as wholes.

The rate of economic growth presents us with a slight difficulty in the transition period, because virtually all of the former Communist countries in Eastern Europe and the former Soviet Union experienced deep declines in income followed by faster or slower recovery. An unusually deep trough, e.g., Bosnia, may occasion very rapid rates of subsequent growth that give a misleading impression of strong performance. We tested and found it best to use as our measure of economic performance the rate of growth of per capita income estimated for each country by linear regression. We leave out of the sample the two most problematic cases, Bosnia and Serbia. Estimates that included those two countries are not dramatically different.

Den1950 has similar positive correlations with the transition growth rate with p-values in the 15–20% range for both samples.

Using den1950 instead also leads to positive but insignificant coefficients. The p-value for the core countries (0.117) is in this case not very far from 10%.

An earlier study of the link between ethnic heterogeneity and the performance of transition economies in Eastern Europe was carried out as the senior thesis of Brown University student Lee Sabow (2001). Sabow’s finding that ethnic heterogeneity was associated with poorer economic performance encouraged us to include the variable in this study.

The full regression results are available at <http://​watsoninstitute.​org/​ped/​scid>.

While country size might influence the degree of inefficiency of a centrally planned economy, controls for country size are almost never included in empirical growth studies for market economies. We nonetheless checked and found that population of 1970, the proxy for country size that we used for our tests of Table 3’s regressions, did not have a statistically significant coefficient when added to most of the regressions in Table 5, and in no case was it significant in both the OLS and the robust version of the same regression.

Those two are den1500 and distance, both cases being in the core country sample regression for corruption. As before, the regressions are available at < http://​watsoninstitute.​org/​ped/​scid/​>.

With corruption added to the regression, the p-value of the coefficient on statehist is 0.106.

Over-identification restriction tests cannot be conducted because there is only one instrument included in each IV regression. The regressions reported in the previous paragraph provide an indication of the excludability of the instruments from the second-stage (growth) regression.

The main difference is that exogeneity is rejected more forcefully by that test in all cases involving ethnic, and in the core country cases when statehist is used, the rejection being below the 10% level for both the corruption and the law models, using that test.

Because the robust regression method is not easily implemented in 2SLS, we checked the robustness of these results by dropping one country at a time from our samples. We find that the signs and approximate magnitudes of the coefficients change little, but seven of the significant coefficients in Table 8 became insignificant and the Wu-Hausman test statistic failed to reject exogeneity in one additional case when we dropped China.

Simple convergence (poorer countries growing faster than rich) might have accounted for the correlation over time of growth rates, but recall that convergence does not hold for our sample during 1970–1990 (see Table 3).

We include all observations for countries in which the required data are available for both the earlier and the later period.

Since we did not find a procedure to implement robust regressions with GLS, we performed robustness tests by estimating each regression in the table dropping one country at a time. The 3 significant coefficients on statehist and den1500 remained significant in all cases except when China was dropped, in which case, the coefficients in the full sample remained qualitatively the same, but their significance levels drop to between 10 and 15%. In the case of statehist in the core country sample, the significance level without China was exactly 10%. Den1500 remains insignificant when dropping the observations for any country except Moldova; when that country’s observations are dropped, the coefficient on den1500 becomes significant at a little under 9%.