Constructing Social Cleavage Indicators Using the Mixed-Effects Model

This study demonstrates how we can construct national indicators by analyzing comparative surveys using mixed-effects (ME) model. ME model, also known as multilevel analysis, is usually understood as a method to analyze nested data. Few studies pay attention to its unique ability to estimate individual (or random) effects.

Estimating individual effects, country-specific effects in the case of national comparison, using ME model has two distinct advantages that existing methods do not have.

One is that we can "estimate" indicators with statistical confidence, not just "compose" them by adding up or averaging relevant figures. This merit addresses the problem of composing indicators using unbalanced comparative data. A comparison of attitudes between a country with sample size of 1,000 and one with sample size of 10 would be questionable if we just calculate means of several attitude values for both countries, because the value of a country with small sample size lack reliability. Estimating individual effects using ME method can avoid this problem by introducing "shrinkage estimator".

Another merit of ME estimation is that we can eliminate the effects derived from differential composition among countries. An indicator value of attitudes or happiness could be different because of different distributions of social groups such as gender, age, and classes. Also, simple averaging of different level of attitude among social groups can suffer from bias, because one social attribute (such as educational background) correlates with others (such as age or gender). ME method can deal with this problem because it uses linear regression technique.

This study applies ME method to compose "Attitudinal Divide Indicator" (ADI). ADI can be interpreted as indicating how an attitude toward government roles differs by social groups. A country with high ADI can be understood as a country where there is a sharp divide of opinion among people.