Dynamic ethnic fractionalization and economic growth

Abstract

In their survey of the literature on ethnic fractionalization and economic performance, Alesina and La Ferrara (2005) identify two main directions for future research. One is to improve the measurement of diversity and the other to treat diversity as an endogenous variable. This paper tries to address these two issues: it investigates the effects of ethnic fractionalization on economic growth across countries using unique time-varying measures. We first replicate the finding of a weak effect of exogenous diversity on growth and then we show that accounting for how diversity changes over time and treating it as an endogenous variable makes a difference. Once diversity is instrumented (with lagged diversity and latitude), it shows a significant negative impact on economic growth which is robust to different specifications, polarization measures, econometric estimators, as well as to the use of an index of ethnic-religious-linguistic fractionalization.

Keywords:

• ethnic diversity
• fractionalization
• polarization
• growth

JEL Classifications:

• O11
• Z12
• O55
• H1

Acknowledgements

The authors would like to thank Jan Fidrmuc, Julius Horvath, Gabor Kezdi, Branko Milanovic, Jeffrey Nugent, James Robinson, Colin Rowat, seminar participants at Oxford University (CSAE) and two anonymous referees for valuable comments on previous versions, the Central European University for the hospitality and the Economic and Social Research Council for financial support (ESRC, Grant RES–000–22–0550.). The usual disclaimer applies. The ethnic, religious, and linguistic fractionalization (and polarization) indexes constructed for this paper are available on-line: http://www.naurocampos.net/delf19892007.xls

Notes

 1. These two features are related, as the secondary data used to measure diversity refers to the early 1960s. The huge popularity of the index constructed by Soviet researchers and published in the Atlas Narodov Mira (Bruk and Apenchenko 1964) is due in large part to its inclusion in Taylor and Hudson's (1972) World Handbook of Political and Social Indicators. For studies that use this index, see Mauro (1995), Easterly and Levine (1997), Collier (2001), La Porta et al. (1999) and Woo (2003a, 2003b).

 2. Campos and Kuzeyev (2007) examine the relationship between growth and diversity between 1989 and 2002 within an endogenous growth framework, while this paper uses the traditional Solow model to study the growth-diversity nexus between 1989 and 2007.

 3. We divided the sample into five groups for exposition purposes (Figure 1). The transition countries in ASIA are Armenia, Azerbaijan, Georgia, Kazakhstan, Kyrgyztan, Mongolia, Tajikistan, Turkmenistan and Uzbekistan. The BALKAN countries are Albania, Bulgaria, Croatia, Macedonia, Moldova and Romania. The BALTIC countries are Estonia, Latvia and Lithuania. The group called BUR comprises Belarus, Ukraine and Russia. The VISEGRAD countries are the Czech Republic, Hungary, Poland, Slovakia and Slovenia. CEEB stands for Central and Eastern European and Baltic countries and which is the sum of the BALTIC, BALKAN and VISEGRAD sub-groups.

 4. Although it is difficult to objectively judge the quality of these different sources of data, note that in each country, collection of these data was done by the same agency, with comparable methodologies. They differ in that censuses cover the entire population and micro-censuses cover a representative sample. These figures were checked against various additional sources, including Rosenko (1999) Nasii I Etnosi V Sovremennom Mire (Nations and Ethnicity in Today's World) and Natsionalniy Sostav Naseleniya SSSR (Ethnic Composition in The USSR, Finansi I Statistika 1991).

 5. For example, we found data on the ethnic composition of the population in Latvia for the years 1994, 1995 and 1996 from different sources, whose indices were identical up to third decimal place. Hence, according to our rule, the time series 1989–1994–2000 was preferred to 1989–1996–2000.

 6. The average number of ethnic groups in the restricted sample was 5.19 and in the unrestricted sample 7.04. Alesina et al.' s and Fearon's analogous figures for Eastern Europe and former Soviet Union countries are 6.48 (27 observations) and 4.55 (31 observations), respectively. The lowest number of groups in our data, including ‘others’, is 3 (in several cases), while the largest is 8 (12 for Mongolia in the unrestricted sample). In addition to data quality, we must also be concerned with data comparability. In this respect, there are few dimensions over which researchers can exert some control. One of the few, however, refers to the number of groups used in the computation of the diversity indexes. Here, we explore different ways of using this information across countries and over time. We find that these variations do not affect our main conclusions.

 7. For instance, the mean of this ethnic fractionalization index declines from 0.3726 (0.3768) in the first period to 0.345 (0.3538) in the second period to 0.3147 (0.3154) in third period to 0.30145 (0.30314) in the fourth period (values using the unrestricted number of groups are in parenthesis). For comparison, Alesina et al.'s value for the early 1990s is 0.3696, while Fearon's is 0.3723.

 8. Fractionalization measures an increase in the number of groups, while polarization maximum is reached with two groups of equal size.

 9. To be more precise, the ADEKW index of polarization is the original index of polarization of Esteban and Ray (1994). The Alesina et al. (2003) index is obtained using different values of α and under the assumption that distance is constant and equal to 1.

10. We thank an anonymous referee for this point.

11. In the fractionalization and growth literature, this approach is used by, for instance, Montalvo and Reynal-Queyrol (2003).

12. We follow MRW in assuming that the sum of g and δ is constant. Although they assume it is constant at 5%, here we report results assuming that the sum of rates is 7.5% so as to reflect the larger depreciation observed in the capital stocks inherited from the socialist period.

13. In previous versions of this paper we also reported specifications for the level of per capita GDP in the left-hand side, instead of the growth rate. Our main results are unaffected by this change. In other words, we still find that fractionalization is important vis-à-vis growth only when treated as an endogenous variable and its dynamics is taken into account.

14. There are important data quality issues that should be kept in mind when interpreting these results (for a discussion see Campos and Coricelli 2002).

15. The results from a standard Granger-causality test show that there is no evidence supporting the notion that growth (Granger-) causes diversity. These results are available from the authors upon request.

16. As our panel covers only four periods and we use the one-period lagged diversity as an instrument, we are unable us to run the AR(2) test.