Democracy, Geography and Model Uncertainty

AuthorDoris A. Oberdabernig,Jesus Crespo Cuaresma,Stefan Humer
Publication Date01 May 2018
Doris A. Oberdabernig*, Stefan Humer** and
Jesus Crespo Cuaresma**
We analyse the nature of robust determinants of differences in democracy levels
across countries taking explicitly into account uncertainty in the choice of
covariates and spatial spillovers. We make use of recent developments in Baye-
sian model averaging to assess the effect of a large number of potential factors
affecting democratisation processes and account for several specifications of spa-
tial linkages. Our results indicate that spatial spillovers are present in the data
even after controlling for a large number of geographical covariates. Addressing
the determinants of democracy without modelling such spillovers may lead to
flawed inference about the nature of the determinants of democratisation pro-
cesses. In particular, our results emphasise the role played by Muslim religion,
population size, trade volumes, English language, natural resource rents, GDP
per capita, being a MENA country and the incidence of armed conflicts as fac-
tors affecting democracy robustly.
Why do democracies emerge, survive, or fail and become autocracies? Social
scientists have often drawn their attention to the quantitative assessment of
driving factors of democratisation processes. Based on the large variety of
democratisation theories, they have tested the impact of a vast number of
covariates that were argued to be conducive to more democratic forms of
political organisation, resulting in a variety of (sometimes contradicting)
empirical findings.
In a seminal contribution, Lipset (1959) discussed the social prerequisites
for democracy and concentrates on the role of economic development, wealth,
education and religion. Since the inception of Lipset’s hypothesis, the theoreti-
cal prerequisites for democratisation have been extended and refined in
*University of Bern, World Trade Institute (WTI)
**Vienna University of Economics and Business (WU)
***Austrian Institute of Economic Research (WIFO)
****International Institute of Applied System Analysis (IIASA)
*****Wittgenstein Centre for Demography and Global Human Capital (WIC)
Scottish Journal of Political Economy, DOI: 10.1111/sjpe.12140, Vol. 65, No. 2, May 2018
©2017 Scottish Economic Society.
various dimensions. The abundance of natural resources is believed to hinder
democratisation as autocratic rulers face high resource rents which they could
use to sustain their power. It has often been argued that income generated by
natural resources generates less pressure for democratisation than income gen-
erated through human capital accumulation.
The effect of education and human capital on democratisation has been
examined in more detail and hypotheses concerning the distribution of educa-
tion among different groups of the population have been added to the theoret-
ical framework linking human development and democracy. Lutz et al. (2010)
reasoned that female education affects fertility rates and has an additional
impact on democratisation through this channel. The induced change in the
proportion of young people in the population has been often argued to impact
on regime stability in what has become known as the youth bulge theory (see
Cincotta, 2008 2009). Inequality and heterogeneity of a country’s population
are also seen as a factor potentially affecting political outcomes. Studies have
often focused on income inequality, and also on ethnic, linguistic and religious
In addition, a country’s colonial history is seen as influential for the rise of
democracy, due to the role played by early institutions and legal traditions
brought to colonies by Western settlers. A range of historical events and geo-
graphic variables have been used in empirical studies to explain and predict
probabilities of democratic transitions. More recently, the attention has
shifted towards taking into account also international factors when assessing
the determinants of political regimes. In this respect, the influence of geogra-
phy and international organisations has been reasoned to impact on the emer-
gence and survival of democracy.
The variety of theories linking geographic, historical, demographic and
socioeconomic developments with the democratisation process, together with
the lack of an overarching theoretical framework, implies that empirical
assessments of the determinants of democracy should explicitly address the
problem of model uncertainty when performing inference. Neglecting the
uncertainty associated to the choice of covariates within linear regression
models results in an overestimation of the precision of estimates and thus
potentially in an overconfident interpretation of the importance of particular
predictors of democratisation (Fern
andez et al., 2001).
Gassebner et al. (2012) and Hegre et al. (2012) provided evidence on the
empirical drivers of democratisation based on methods that take into account
model uncertainty. Gassebner et al. (2012) performed extreme bounds analysis
(EBA) in order to unveil the robustness of democracy determinants and con-
clude that GDP growth, past transitions, and OECD membership, as well as
fuel exports, and the share of Muslims in the population are significant drivers
of a transition to democracy. On the other hand, GDP per capita, past transi-
tions, having a former military leader as chief executive, and having other
democracies as neighbours are variables that have a robust significant effect
on the survival of democracies. Hegre et al. (2012) applied a less stringent ver-
sion of EBA based on considering the entire distribution of parameter
Scottish Journal of Political Economy
©2017 Scottish Economic Society
estimates to determine the level of confidence in each of the explanatory vari-
ables (see Sala-I-Martin, 1997). In contrast to Gassebner et al. (2012), the
authors find that more than half of the 85 variables included in the analysis
are robust determinants of democratisation, while considerably fewer variables
are robust determinants of democratic stability.
Our study builds on the work of Gassebner et al. (2012) and Hegre et al.
(2012) and expands it in several respects. First, we move away from methods
related to EBA and implement a fully Bayesian approach to model uncer-
tainty, relying on recent advances of Bayesian model averaging (BMA) in the
presence of spatially correlated data (Crespo Cuaresma and Feldkircher, 2013;
Crespo Cuaresma et al., 2014). By basing inference on the posterior distribu-
tion across all possible model specifications, Bayesian methods, in contrast to
EBA, present a natural framework to deal with the uncertainty in model spec-
ification, not holding neither model size nor a particular subset of the
explanatory variables fixed. This allows not only to evaluate the statistical sig-
nificance of the coefficients but also to quantify the uncertainty of belonging
to the true data generating process for each covariate.
In addition, using spatial filtering methods, the recent developments put
forward by Crespo Cuaresma and Feldkircher (2013) provide tools for the
assessment of spatially correlated data in potentially very large model spaces.
A further novelty of this study, thus, is that we expand the set of model speci-
fications assessed hitherto in the literature that deals with robust correlates of
democracy by explicitly taking into account spatial autocorrelation in democ-
racy data. This is particularly important as previous studies controlling for
such spillovers either indirectly by including information on democratic neigh-
bours among their regressors (see e.g. Pevehouse, 2002a, b; Li and Reuveny,
2003; Gleditsch and Ward, 2006; Eichengreen and Leblang, 2008; Csord
and Ludwig, 2011; Gassebner et al., 2012) or directly by estimating spatial
models (see Leeson and Dean, 2009; Seldadyo et al., 2010; Kelejian et al.,
2013) find such domino effects to be statistically relevant.
In a more recent
contribution, Bonhomme and Manresa (2015) allow for group-specific time-
varying patterns of democratisation without imposing a particular spatial
structure on group-membership. Their results suggest that group-membership
in the detected waves of democratisation is geographically correlated, provid-
ing a further indication for potential spatial contagion. Not controlling for
geographical spillovers in the presence of spatial dependence results in omitted
variable bias and inconsistent parameter estimates (see LeSage and Pace,
2009). As spillovers in democracy might not only be based on geographical
proximity (Leeson and Dean, 2009) we also consider models with spatial
weighting matrices that are based on religious proximity additionally to differ-
ent definitions of geographical linkages. In order to minimise potential reverse
causality concerns, we move away from the panel structure used by Gassebner
Acemoglu et al. (2008) included a democracy index of a country’s trading partners
among the regressors and do not find a statistically significant effect on democracy levels.
For an extensive overview of potential mechanisms that cause democratic spillovers, see
Kelejian et al. (2013) and Leeson and Dean (2009).
Scottish Journal of Political Economy
©2017 Scottish Economic Society

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT