Elkins, Zachary. "Gradations of Democracy? Empirical Tests of Alternative Conceptualizations". American Journal of Political Science, Vol.44, No.2 (2000): 293-300.

Elkins, Zachary. "Gradations of Democracy? Empirical Tests of Alternative Conceptualizations". American Journal of Political Science, Vol.44, No.2 (2000): 293-300.

• Apparently there is a dispute between political scientists, including a source mine of opinions on the issue, over whether intermediate categories should exist between democracy and dictatorship. The author argues that it makes sense in the light of increasingly specific methods of data collection (293).
• Dr. Przeworski supports a dichotomy between democracy and dictatorship on the grounds that democracy cannot exist by degree, but rather exists or does not exist. The question of degrees is therefore invalid (293).
• A measure of democracy has 'construct validity' if it is a good predictor of phenomenon associated with democracy. Since there is political science research about the behavior of democracies -- where indices of democracy can be used as a variable -- it would make sense to measure degrees of democracy if that degree is a good predictor of behavior associated with 'democracy-ness' (294).
• Drawing on the notion that democracies do not fight other democracies, the author tests whether graduated scales of democracy or dichotomies do a better job predicting wars by democracies. A test shows that a graded measure is predictive, whereas a dichotomy is totally unhelpful (294-296).
• Based on theories of regime stability being high with strict authoritarian regimes and democracies and weak for hybrid governments, the author runs a number of complex statistical tests, which demonstrate that a dichotomy is a good measure for this, although then a graded scale still has more practical predictive uses (296-297).
• Dr. Przeworski makes a claim that even if graded scales are better representations of 'actual democracy', they are more likely to contain error than a dichotomy because it is difficult for error to be introduced if the only possible values are 1 and 0. This, however, is only because a graded scale provides greater specificity than a dichotomy cannot (298-299).