The Future of the Past: Protestant Missions Persistent Impact on Democracy

• In Event: Democratization and Regime Change

Abstract

In a widely cited APSR article (Woodberry 2012) I argued that Protestant missions fostered the global spread of democracy by moderating colonial abuses and by pioneering mass education, mass printing, newspapers, and voluntary associations. When other religious elites feared losing adherents to Protestantism, they copied these innovations – thus dispersing power and enabling peaceful political mobilization prior to independence. These factors made stable democratic transitions more likely.
I demonstrated my argument historically and statistically - using Bollen/Paxton democracy data from 1950-1994. However, much has changed since 1994. My arguments focused on missions prior to WWII, but missionaries no longer dominate printing, education, etc., and most colonized countries gained independent decades ago – thus the impact of historic missions may have dissipated. Moreover, there has been the “Catholic Wave” of democratization; the post-1989 fall of Communism, and so on. Vatican II and other movements have changed the Catholic Church. Various Pentecostal, Fundamentalist, and Evangelical movements have changed Protestantism. Any of these factors may have altered the association between Protestant missions and democracy. Moreover, many statistical associations are not causal, and in the past decade many new methods for isolating causation have emerged. Are the arguments in Woodberry (2012) robust to these changes and new methods?
In this paper I extend my analysis from 1994 to 2017, test its robustness using 7 different measures of democracy, and use multiple methods to test causality. The results suggest that societies that had greater exposure to Protestant missions are still more democratic (although the impact has weakened somewhat over time). There is some evidence that societies with greater exposure to Catholic missions have become more democratic since the 1990s – although the evidence for causation is weaker than for Protestant missions. The impact of missions is stronger using continuous and latent measures of democracy (rather than dichotomous measures), and stronger in samples that include small countries (many of which have the longest exposure to Protestant missions). However, the results are amazingly robust.
To test for causality I first used instrumental variable regression with the new measures of democracy (1950-2017) and instrumenting for Protestant missions, Catholic missions, and European settlers. If I control for Protestant missions (i.e., if it is an included instrument), then the instrumented versions of Catholic missions and of European settlers are never significant. However, if I control for both Catholic missions and European settlers (plus all the other included instruments I used in 2012), the instrumented version of “years of Protestant missions” remains significant. This suggests Protestant missions is causal.
Second, I used Baysian Model Averaging (BMA) for 20 models with 4 continuous measures of democracy (V-Dem polyarchy, V-Dem Liberal Democracy; SVMDI, ExUDS): for 1950-1994 the posterior inclusion probability (PIP) suggests that “years of Protestant missions” should be included in 100% of models and “number of Protestant missionaries per 10,000 in 1923” in 75%. For 1950-2017 they should be included 100% of the time and 30% respectively. The only other variables with high inclusion probabilities were “island”, “percent European” and occasionally “being a major oil producer”, but not Catholic missions. However, BMA is a weighted average of models based on model fit and thus can be influenced by a small number of models.
Thus, I next used mrobust (Young and Holsteen 2017). Mrobust runs every possible model using the variables in the analysis and reports the sign and significant of each variable in all models. In the more than 10 million models I ran, “years of Protestant missions” was positive in 100%, and significant in 97% for 1950-1994 and 92% for 1950-2017. “Number of Protestant missionaries per 10,000 in 1923” was also positive in 100%, and significant in 91% for 1950-1994 and in 57% for 1950-2017. The Catholic missionary variables were generally positive (although not always) and were less likely to be significant (especially prior to 1994 – 46% & 17% respectively). This suggests that regardless of the measure of democracy, the time frame, or the other variables in the model, the impact of Protestant missions is amazingly robust and the impact of Catholic missions is plausible, but less robust.
Finally, I used two techniques to estimate the size of omitted variable bias necessary to remove the impact of Protestant missions. Both suggested that the omitted variable bias would have to be implausibly large. These results suggest that future analyses should take the impact of historic missions seriously. Most statistical research about democracy, education, etc., does not measure missions and is thus likely to suffer from severe omitted variable bias.

Author

• Robert D. Woodberry, Baylor University