Income and democracy: dynamic misspecification due to the presence of serial correlation

ABSTRACT

We address the issue of dynamic misspecification arising from the possible existence of serial correlation when examining the relationship between income and democracy by employing a panel count data estimation method. We find evidence that income per capita systematically predicts democracy, but it does not so in relation to the colonization experience of each country. Our results are robust across model specification and samples.

KEYWORDS:

• Income
• democracy
• panel count data
• serial correlation

JEL CLASSIFICATION:

• C23
• O10
• P16

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ We drop all non-independent territories and very small states, newly established countries that followed the collapse of the Soviet Union, Yugoslavia and Czechoslovakia and some African countries.

² In particular, the democracy index (DEMOC), from the Polity IV data set, and the Political Rights index, from the Freedom House (FHPR) data set, are originally measured on a scale or Likert basis of 0 to +10 and +1 to +7, respectively. This means that by construction they are positive integers. The two conversions most often used for this purpose are that of a single continuous measure (e.g. Acemoglu et al. 2008; Heid, Langer, and Larch 2012; Moral-Benito and Bartolucci 2012; Benhabib, Corvalan, and Spiegel 2013; Cervellati et al. 2014; Posso and Feeny 2015) and that of a dichotomous measure by means of a dummy variable (Przeworski et al. 2000). However, in the aforementioned studies, DEMOC was converted into a continuous variable by subtracting autocracy scores (ranging from 0 to −10) from democracy scores (ranging from 0 to +10) to create a composite index of the political regime with values ranging from +10 (strongly democratic) to −10 (strongly autocratic). On the other hand, the transformed FHPR used by Acemoglu and Robinson (2006) was obtained, following Barro (1999) and supplemented with data from Bollen (2001), by assigning a value of 0 to the fewest political rights (FHPR rank +7) and a value of +1 to the most political rights (FHPR rank +1). Lastly, Przeworski et al. (2000) assumed that the true nature of democracy lies on a J-point scale where the dummy variable takes the value of 1 if J ≤4 and 0 otherwise.

³ We do not supplement the FHPR index with Bollen (2001) data, as the latter is not a discrete count.

⁴ Using the colony coding in the CEPII data (2011). Posso and Feeny (2015) have accounted for the identity of the colonizer.

⁵ In our sample, the conditional mean and the conditional variance of the DEMOC index are, respectively, equal to 4.5 and 18.0 for the 5-year data set and 4.6 and 17.5 for the 10-year data set. The conditional mean and the conditional variance of the FHPR index are, respectively, equal to 3.5 and 4.5 for the 5-year data set and 3.4 and 4.3 for the 10-year data set. Clearly, the above comparisons indicate the possible presence of binomial variation in the marginal distribution of the data.

⁶ Due to space restrictions, we omit the results of this test but are available on request.

⁷ See Lee and Paine (2017) for a detailed analysis on this.