Question: I was wondering if you would mind explaining the stats terms below briefly (as the online sources I have found on this topic resort to rather convoluted language in their explanations) in the context of the Przeworski and Limongi (1993) paper (‘Political Regimes and Economic Growth’)?

• Simultaneity: the causal relationship between democracy and economic development is unclear because it could go both ways, which makes researchers’ jobs very difficult?
• Attrition bias: there are existing differences between the original state of treatment and control groups that are impossible to avoid in selecting case studies?
• Selection bias: the selection criteria of choosing regimes based on their growth level, dependent on survival (why specifically was this criteria chosen?) is problematic, and it’s relevant to exogenous factors (ones that are taken as a given and not explained in a model) and endogenous factors (ones that are explained by variables in a model)?

Answer: Thanks a lot for the question, which is a good one and particularly useful because I think a number of students found these terms slightly unclear.
Simultaneity: your definition here is correct. We observe a correlation in which economic development is related to regime type (i.e. democracy or dictatorship) but it could be that that the former leads to the latter, or the latter leads to the former. Or, indeed, it may be that reaching a certain level of economic development prompts transition to democracy and also affects subsequent growth rates (that are then attributed to democracy, but actually stem (at least to a degree) from passing a threshold of economic development). As you say, these competing plausible explanations for the correlation are not easy to distinguish between empirically.

Attrition bias: this is different from the definition that you suggest. Let’s start with an observed correlation between regime type and economic growth. It is easy to assume that such a correlation is because the regime type affects the growth. However, it may actually be the case that different regimes have a greater or lesser propensity to survive when confronted with positive or negative circumstances. So, (drawing on the article) let’s imagine a scenario in which dictatorships are more likely than democracies to collapse when faced with poor economic growth. If we measured growth and regime type in this scenario it would look like democracies lead to worse growth than dictatorships. In fact, what is happening is that dictatorships with bad economic growth collapse, and the democracies that emerge from them inherit poor growth. At the same time, democracies continue through periods of poor growth and have those figures included in their averages whilst, of course, dictatorships only have good growth figures in their averages (because they collapse when they have poor growth). Thus, overall it looks like dictatorships promote higher average growth but, in fact, they have higher attrition rates (hence attrition bias) when faced with poor growth.

Selection bias: I interpret this to be the overarching problem (encompassing the above two problems) that Przeworski and Limongi are concerned with, based on the following quote:

• ‘If democratic regimes are more likely to occur at a higher level of development or if democracies and dictatorships have a different chance of survival under various economic conditions, then regimes are endogenously selected [hence selection bias].’ p. 62.

The point here is that we cannot select an unbiased set of cases (i.e. one that is not beset by problems such as simultaneity or attrition bias). We cannot observe each country as it would be if it was the same but with a different regime type, and if we try to find an equivalent country we see that the similarity between the two countries means that they share regime type and economic growth as well. So we cannot observe the impact of changing regime type in otherwise identical circumstances. Another way of thinking about this is that the factors affecting regime type and economic growth are not distributed randomly across the world so every time we look at countries with a particular regime type we automatically look at countries with other shared characteristics that undermine our attempts to investigate the effect of regime type on economic growth.