Most Recent Posts

Wednesday, November 26, 2014

Resource Wealth and Political Regimes in Africa

Another exam entry.

Jensen, N. and Wantchekon, L. (2004), “Resource Wealth and Political Regimes in Africa,” Comparative Political Studies 37(7): 816-841.

Question 1: What is the central hypothesis of the article? How strong is the theoretical justification for that hypothesis? Explain.

Jensen and Wantchekon are testing four hypotheses, all based on the relationship between natural resource endowment and regime type and action. First, they examine the relationship between natural resource endowment and likelihood of authoritarian regime type. Second, they test the relationship between natural resource endowment and government spending. Next, they consider the relationship between natural resource endowment and quality of governance. Finally, they examine the likelihood that greater natural resource endowment leads to a breakdown in democracy in the 1990s. They limit their analysis to Africa, which justifies the 1990s as a critical break point. This approach is justified because while their first three hypotheses are all well-supported in the literature, the prior literature was based on a specific, narrow group of states – oil-rich OPEC states. Thus, the variance in type of natural resource endowment throughout Africa allows for more robust testing of the hypotheses. Specifically, qualitative research supports their argument by showing three paths for resource-rich states to develop authoritarian regimes.

Question 2: What dependent variables are used and how are they measured? Are these valid and reliable measures? Why or why not?

Each hypothesis uses different dependent variables (as is appropriate). For the first hypothesis, Jensen and Wantchekon operationalize democracy through the standard derivation of Polity III scores, and then adding 10 to eliminate negative values. For government spending, they use government spending expressed as a percentage of GDP. For governance, they use six aggregate measures of governance developed by the World Bank. For the final hypothesis, they use democracy measures again as with the first hypothesis, but use a later derivative of the Polity dataset (Polity 98). These variables have internal validity throughout: using Polity scores is a standard measure of regime type, and the manipulation they employ renders the data amenable to regression analysis without eliminating any meaning; and using government spending to measure government spending is, well, intuitive. The most interesting variable is the governance measures, which they justify by reference to the literature developing it. Presumably the aggregate data used to generate the measures are internally valid; however, since they are only World Bank working papers, and not subject to formal peer review, I would not object to someone questioning the validity of these measures. For the same reasons, I would regard all of these measures as reliable, since they can presumably be reconstructed; but would not complain if someone else reached a different conclusion, particularly if the World Bank measures had inter-coder issues.

Question 3: What are the primary independent variables and how are they measured? Are these valid and reliable measures? Why or why not?

Jensen and Wantchekon use a constructed variable for their tested independent variable. This variable was assigned an ordinal value based on the percentage of merchandise exports derived from fuel, minerals, and metals. I am concerned about the internal validity of this measure, as it excludes exports based on renewable natural resources such as agricultural products and timber. Jensen and Wantchekon do not justify this exclusion; as it could reduce the values of many cases’ independent variables, the exclusion could lead to Type I error. Because the measure is invalid, its reliability is irrelevant; however, I note in passing that the data is easily recoverable, and the way that they reduce the granularity of the data to allow them to fill in missing data is actually pretty ingenious.

For control variables, they rely exclusively on control variables found in the data sets they use for their panel data, to insure internal validity (since all the data therefore comes from the same source). Log GDP per capita eliminates GDP fluctuations caused by population shifts, and is therefore externally valid, and GDP growth controls for democratization driven by economic growth – the “rising tide lifts all boats” hypothesis. Both measures are easily reconstructed either from existing data sources or by minimal mathematical manipulation, and thus are reliable.

Question 4: Are there any unnecessary control variables in the analyses? What are they and why are they unnecessary?

For their final hypothesis, Jensen and Wantchekon are replicating a test done in 1994 by Bratton and Van de Walle, and thus must use their control variables. So even though they find that none of Bratton and Van de Walle’s control variables significantly account for democratic sustainability between 1994 and 1998, their inclusion is necessary to make it a valid replication. The use of colonial dummies baffles me. It clearly matters, as it is strongly significant at even a higher correlation coefficient than the test IV, but Jensen and Wantchekon say that it “had little effect on our dependence on political regimes.” This statement seems either to be saying something different than its facial content, or to contradict the data altogether. While I can’t say this variable was unnecessary, I would say that without a better explanation than what was offered, that the paper should have been unpublishable.

Question 5: Does the study not control for something that should be controlled for? What is it and why should it be controlled for?

Because African politics can frequently be driven by ethnic divisions, I would like to see that variable accounted for somehow. In particular, I question whether differential resource endowment for different ethnic groups could be driving authoritarian regimes, or whether it merely heightens existing autocratic tendencies that are driven by ethnic divisions.

Question 6: Using only Table 1 of the article, answer the following:
a)    What is the likely substantive effect of each independent variable on democracy in column 2, when moving the variable from one value to another (using your choice of values)?
b)    Explain why and how the addition of dummy variables for four colonial legacies changes the statistical results from column 1 to column 2. Be as specific as possible in your answer.

a.    The relationships between the variables are assumed to be unstandardized. Thus, I will speak in terms of actual values instead of standard deviations. However, I will express the values of changes in terms of standard errors. Thus, for example, for an increase in log GDP per capita of 3.141 (the standard error), I would expect an increase in democracy score of 1.995 (the correlation coefficient). A quick review of the log tables suggests that this is a remarkably difficult achievement, since this would require a GDP growth of 23.127 percent. A log GDP per capita increase of 1 (still substantively unbelievable for a single country to accomplish without an exogenous variable not captured in the data), would result in an increase in democracy score of .635 – or an 11.3% increase in democracy scores for the mean country, ceteris paribus. Wealth thus clearly plays a tremendous role in the health and longevity of democratic institutions. A more reasonable increase in log GDP per capita of .5 would increase democracy scores by .318, a nearly 6% increase.

The same cannot be said for economic growth. It appears that a 28-point increase in GDP growth rates only increases democracy scores by .076. Increasing GDP growth rates by one point thus increases democracy scores by .003 – an increase of less than 1/10 of one percent.

Looking at natural-resource dependence changes the analysis, because unlike the two prior independent variables, Jensen and Wantchekon have converted their measure of resource dependence to an ordinal variable. This means that while we can still talk about the relationship, we can’t give it the kind of mathematical precision that we could before. We can assume with ordinal variables that a one-step change in resource-dependence equals the standard deviation (instead of assuming, as we did with the continuous IVs, that the standard error equaled the standard deviation). Thus, a one-step increase in resource-dependence (which could be a one-percent increase in the share of merchandise exports from the extraction sector) should decrease democracy scores by .364 – or a 6.5% decrease from the mean.

The remaining variables are dummies, and thus are treated differently. It appears that the 1970s were more authoritarian than the other twenty-five years of the study’s period. If a regime was measured during the 1970s, it averaged a 15.8% decrease in democracy scores. The 1980s were the same, but less so, with a 12.4% decrease in democracy scores – but the p-value cautions us that this may be noise. This is probably because of a tremendous increase in democratization in the 1990s. The remaining dummy variables are based on the colonial heritage of the regimes studied. Because there is no reference category, we can assume that these dummy variables do not interact. In summary, being a former British colony would be expected to increase democracy scores by 40.9% over the mean, or 2.305, while being any other nation’s former colony would be expected to decrease democracy scores by anywhere from 27.8% below the mean (-1.566) to 32.4% below the mean (-1.825).

b.    The addition of colonial heritage dummies reduces correlation values for log GDP per capita, GDP growth, and resource dependence, while increasing correlation coefficients for decade dummies. This suggests that the philosophy of the “white man’s burden” actually had relatively beneficial effects on a colonized population. Setting aside the question of whether colonialism was an absolute benefit to colonized peoples, it is clear that the British, who saw their institutions and systems as vital to the governance of colonies, tended to create wealthier and more diverse economies that had more robust democracies, while Continental colonizers, who saw the colonies as resource-extraction outposts and nothing more, did not have a lasting effect in creating institutions to provide just and secure government to the colonized after their departure. Perhaps they thought they’d never leave. It also is clear that the effect of colonialism diminishes over time, as other variables come into play to drive democratization and stability.

Question 7: How would you go about either a) extending this study or b) challenging this study with a separate research design? Discuss.


Frankly, given that I’ve demonstrated that the paper should have never been published in its current state, I object to being asked to extend or challenge a paper that should be retracted. However, this is clearly a place for comparative institutional design approaches. It seems clear to me that while resource dependence may have a role to play in democratization, colonial heritage – and thus, institutional design – has a far vaster role to play. It is thus vital to test this by examining the institutional design factors that are the legacy of colonialism, and to test Westminster-descended institutions against those of other types. I would start with a fine-grained comparative approach – paired case studies. I would use a successful Westminster democracy and a successful Continental democracy, and a similar pair of authoritarian regimes. Once that was done, and I had identified what I thought were the salient factors, then a quantitative method could robustly test that model throughout Africa. Controlling for resource dependence as Jensen and Wantchekon measure it would allow me to directly test the different models.