Growth, Ethnic Diversity and Political Rights in Sub Saharan Africa
Andrew Duguay December 4 2008 EB 313
2 (a)
Before doing any research, I initially had a question in my mind, “Does ethnic
diversity have a negative effect on economic growth?” This was a partly unfounded hypothetical question, but also an inquisitive interrogation of claims recently made that the reason the United States does not have universal healthcare is due to the rules of incentives. A more ethnically diverse country such as the U.S. would perceive fewer gains from tax dollars spent on social services than an ethnically homogenous European country. This would be due to the fact that the perceived benefits from the tax dollars would be lessened due to more of ones money being shared among “other” people (i.e. another ethnic group that one does not normally associate with). In the same way, ethnic diversity could effect overall growth if both the majority and minority groups find it difficult to efficiently and effectively work together. I was curious if this applied to Sub-Saharan Africa at all. What if the forced colonization in years past that split socioeconomic groups into countries made it more difficult for members of the country to pursue economic ends in a way that a society normally could? The consequences in later years could certainly be war, oppression of minority groups and consequently less overall economic growth due to the underutilization of all people groups. In my research, I set out to find if ethnic fractionalization negatively effected long term economic growth in Sub-Saharan Africa. There has been a significant amount of research done on this topic in recent years. Most of the studies can be traced back to the empirical work of Easterly and Levine in 1997 on the effect of ethnic diversity on national outcomes (1). They created an index (commonly known as ELF) for ethnolinguistic fractionalization measuring the diversity level of a country by languages
3 spoken and difference ethnic backgrounds (2). Easterly and Levine then showed this index to be a statistically significant variable when determining long term economic growth on a country wide level. Since then, many studies have been done using ELF; including one study in 2001 by economist Paul Collier. In his research, Collier showed that ethnic diversity is predicted to be damaging to the economic growth of a country only in particular circumstances, namely dominance and dictatorship (3). Theoretically, this would be due to the fact that in a democracy, minority groups can effectively be heard and influential in governance. Similarly in a country that is not dominated by one ethnic group, there is greater chance and opportunity for minority groups to have effective say in political matters. In Collier’s research, he controlled for the effect of democracy on ethnicity by multiplying the variables together. Because of the importance showed in Collier’s work of considering the political rights of country, I did the same for all of my regressions. In Paul Collier’s regression (seen listed as regression 1 in table 1) he used cross section data over a period of 30 years from 1960-1990. He used the ELF data produced by Easterly and Levine in his regression, but since that time, Alberto Alesina et al. have published an updated ELF data set on ethnic fractionalization using more detailed and up to date findings (4). In my regressions, I attempt to recreate and update Collier’s important findings on ethnic diversity, democracy and growth. Specifically, I wanted to see if his methods were relevant to Sub Saharan Africa in a more recent time period, say from 1990 to 2007.
4 (b)
The data set I use for my results was built from scratch (for the variables I use and
the definitions for all the variables, see the appendix). When looking at the data from my regression and the subsequent results, it is important to note four key differences in the data between my new regression and Collier’s.
1. ELF index. I replaced the ELF statistics with new “Ethnic” data produced in 2003. Ethnic uses the same methods of measurement and scale (0 to 1). However, the new data divides language effects from ethnic effects and I use Ethnic (though the two are highly correlated and produce similar results). 2. Democracy Index. Collier uses the Gastil index of political freedom, since 1990, Freedom House (producers of the Gastil index) has slightly altered their methods but use the same scale and concept. 3. Time Period. Collier uses data from the period 1960-1990. I will use the latest data from 1990-2007. 4. Countries. Collier runs his regression using data from every country. I will just use Sub-Saharan countries to see how relevant these variables are to their particular case.
The data used for my research was pulled primarily from the World Development Indicators on the World Bank website. There are two exceptions: the democracy measurements come from Freedom House and the ethnic diversity measurements come from work done by Alberto Alesina et al. I believe all data in my panel to be the best available to me at the time. The World Bank is a primary resource for data for many
5 econometricians. I found that for the variables and time frames I was seeking, the World Bank’s WDI had the most available data. Other databases searched include: UNDP HDI indicators, WHO, WTO and Harvard University. Concerning the ethnic fractionalization variable, I chose the Alesina et al measurements based on the widespread use and acceptance I found in reading other research articles. It also made sense to use a diversity index based on the same ELF index that Collier used in his work. There are other indices of diversity that measure effects such as polarization and religious diversity but these have been found to be less effective predictors (5, 6). The democracy indicator came from Freedom House, a non-profit, nonpartisan organization. Their indices of political rights and civil liberties have been the most widely used in the research on this topic that I have found. There are other measures of democracy out there, but I chose this index based on both the widespread acceptance among top economists and its similarity to the Gastil index used by Collier for his work.
(c)
Table 1 lists four separate regressions. Regression 1 is Paul Collier’s original
regression for his paper Implications of Ethnic Diversity, published in 2001. The three following columns are regressions produced from my created data set and variables. In my panel data, I divided the time variable into three equal periods of six years each (1990-1995, 1996-2001, and 2002-2007). This allowed for more observations and to control for fixed effects. It is important to note that in controlling for fixed effects, my following three regressions dropped the variables landlocked and dom65 because for all three time periods these variables are constant over time.
6 Table 1. Dependent variable is growth of per capita real GDP Variable (1) (2) (3) (4) lnpop
-0.97 (-3.56)
2.654 (1.85)
-0.645 (-0.55)
pop ethnicXdem lngdp
-0.005 (-3.37) -0.87 (-3.35)
-0.384 (-0.70) -6.195 (-2.14)
-0.990 (-2.12) -8.906 (-5.03) 0.136 (5.66)
-0.369 (5.65) -0.989 (-2.00) -8.896 (-5.06) 0.137 (5.65)
39.16 (2.15)
57.11 (4.94)
57.37 (5.16)
20.13 0.71 118 Yes
26.24 0.72 119 Yes
exports landlocked dom65
_cons
-0.93 (-2.02) -0.55 (-1.33) 9.60 (4.46)
F R^2 No. Observations Fixed Effects
7.18 .23 102 No
8.82 0.30 132 Yes
In Regression 2, I attempted to update Collier’s data set using data from 19902007 instead of 1960-1990. I also updated the Ethnic variable from ELF to Ethnic as described above. In running this regression, the sole purpose was to compare it to Collier’s results by making the regression as similar as possible. There was no attempt to specify the functional form for regression 2; I merely imitated what Collier perceived as correct for his regression. However, I did hypothesize the signs below before running the regression.
•
Lnpop – negative – Because an increased rate of population growth, all else equal, will spread thin per capita GDP growth.
7 •
Lngdp – negative – Because country starting at a lower incomer per capita level, all else equal, has the ability to grow percentage wise at a higher rate due to diminishing returns at higher levels of income.
•
ethnicXdem – negative- Because both higher levels of ethnic diversity and lower levels of democracy within a country could be deterrents of growth.
The results from regression 2 showed a surprisingly weak connection between the variables and economic growth. My negative hypothesized sign on logged population growth showed to in conflict with the results (positive). The log of initial GDP was the only variable that showed to be significant at the 5% level, which indicated something was wrong. Together the variables were relevant according to the F-Test so there is not much ground to say that the regression included irrelevant variables. However it appeared that the regression suffers from incorrect functional form and/or omitted variable bias. In light of the results from my ‘Collier mimicking’ regression, there appeared to be a strong case for omitted variable bias. In hypothesizing possible omitted variables to long term growth per capita, I came up with…
•
Female literacy rate – a measure of educational attainment in a country but also a reflection of that country’s willingness and ability to have both female and male participation in production in the economy.
•
HIV prevalence rate – a country that is suffering from a widespread epidemic can certainly see the effects in worker productivity which can show a strong negative effect on long term economic growth
8 •
Export growth per year- a trade type variable is especially important for developing or natural resource scarce countries such as many of those in SubSaharan Africa. Export growth can be positively correlated with economic growth due to significant gains to trade such as diversification, comparative advantage and increased incentives for peaceful interconnectedness with neighboring countries.
In adding these variables from my own created dataset, I found that my HIV and literacy rate data proved not comprehensive enough. Due to not having adequate data for all of my three time periods per country, these variables dropped my number of observations down to 72 and 51 respectively. My dataset often had only one figure per country which would not effectively reflect any changes in these variables with changes in growth. My export growth rate data was more comprehensive and so I add it to my regression, as seen in regression 3. In regression 3, we see a substantial improvement on many levels. The R^2 rises to 0.72, and all the variables (minus lnpop) are statistically significant at the 5% level. Importantly, all variables now reflected the hypothesized signs in the coefficients. To formally test whether regression 2 had omitted variable bias, I ran a correlation matrix in stata between exports and the other variables and then completed an omitted bias matrix.
9 . corr exports lnpop ethnicXdem lngdp (obs=118) exports exports lnpop ethnicXdem lngdp
lnpop ethnicXdem lngdp
1.0000 0.1518 -0.0143 0.0037
lnpop ethnic~m 1.0000 0.3723 -0.4669
B w/ exports omitted 2.654 -0.384 -6.195
bias + + +
1.0000 -0.4551
B w/ exports -0.645 -0.99 -8.906
lngdp
1.0000
expected sign + + +
sign of corr between exports, X var + +
It appeared the regression did indeed suffer from omitted variable bias. The predicted bias was correct in two out of the three variables. The ethnicXdem variable showed incorrect bias but, given the negative correlation between exports and ethnicXdem is very small (-0.01), this does not appear to be of any big concern. A last improvement I made to my regression concerned the population variable. While Collier included population growth in log form, it appeared, from the scatter plot, that there is no clear evidence that population growth has to be logged. Theoretically, it can also be argued that a diminishing effect of population growth on per capita GDP growth might only substantially take place at population growth rates that are much higher than the normal growth rate experienced in Sub Saharan Africa, thus negating a reason for logging the variable.
20 -40
-40
-20
0
Growth
0 -20
Growth
20
40
40
10
-1
0
1
lnpop
2
-4
-2
0
2
4
6
Population
After changing the functional form of the population variable, the regression appears to make a better fit. In regression 4, the population variable becomes statistically significant while not altering the coefficients of the other variables drastically.
(d)
The results imply that Collier’s regressions do not entirely hold up on their own
when looking at results from recent years in Sub Saharan Africa. It appears that when looking at the effects of ethnic diversity and democracy on growth, controlling for trade liberalization is essential. While I am disappointed that my results did not hold up for the variables HIV and female literacy, I expect that with more data, these variables can also be good predictors of economic growth and show that an ‘ethnic diversity’ and a ‘democracy’ variable are relevant to long term economic growth. From this study we can conclude that ethnic diversity, when properly controlled for with political freedoms, can be a strong indicator of economic growth in sub Saharan Africa in recent years. Regressions 3 and 4 predict that for every 1 unit increase in the ethnicXdem variable, per capita GDP growth decreases by 1 percentage point, all else equal. While the statement is hard to conceptually quantify, the variable shows us that
11 democracy is a very real need for diverse African countries. With a democratic government, minority groups can be heard and participate in the economy fairly, leading to more economic productivity and leading more people out of poverty.
12 Appendix: Definition of Variables Growth – Mean GDP per capita growth (annual %) for period t • Annual percentage growth rate of GDP per capita based on constant local currency. GDP per capita is gross domestic product divided by midyear population Lnpop – Mean annual population growth for period t • Annual population growth rate. Population is based on the de facto definition of population, which counts all residents regardless of legal status or citizenship--except for refugees not permanently settled in the country of asylum Dem – Mean of annual democracy index over period t • Political Rights and Civil Liberties are measured on a one-to-seven scale, with one representing the highest degree of Freedom and seven the lowest Ethnic – Ethnic fractionalization within a country • An index from 0 to1 (zero being pure homogeneity) measuring the fractionalization of ethnicity within a country using the formula
Lngdp – log of GDP per capita for initial year in period t (constant 2000 US$) • GDP per capita is gross domestic product divided by midyear population. GDP at purchaser's prices is the sum of gross value added by all resident producers in the economy plus any product taxes and minus any subsidies not included in the value of the product Exports – Mean of exports of goods and services (annual % growth) for period t • Annual growth rate of exports of goods and services based on constant local currency. Exports of goods and services represent the value of all goods and other market services provided to the rest of the world Landlocked – A dummy that measures 1 if the country has no access to sea ports Dom60 – Dominance by a single ethnic group in a country • A dummy taking the value 1 when the largest ethnic group constitutes between 45% and 60% of the population. This is based on the range used in Collier’s regressions
13 Bibliography (1) Jackson, Ken (2007), “Why Does Ethnic Diversity Affect Public Good Provision? An Empirical Analysis of Water Provision in Africa” (2) Easterly W. and R. Levine (1997), “Africa’s Growth Tragedy: Policies and Ethnic Divisions”, Quarterly Journal of Economics, 111(4), 1203-1250. (3) Collier P. (2001)’ Implications of Ethnic Diversity’ Economic Policy 32 129-66. (4) Alesina, A., A. Devleschawuer, W. Easterly, S. Kurlat and R. Wacziarg (2003), “Fractionalization”, Journal of Economic Growth, 8, 155-94 (5) Reynal-Querol, M. (2002), “Ethnicity, Political Systems and Civil Wars”, Journal of Conflict Resolution, 46(1), 29-54. (6) Alesina, A., E La Derrara (2004), “Ethnic Diversity and Economic performance”, Journal of Economic Literature