Worldwide Climate Change Opinion: Limitations on Latent Trend Models

Abstract

In this paper, I test how well a group-level, ordinal item-response theory (IRT) model can summarize trends in concern about climate change at the country level. I fit the model to data from almost 2.8 million survey question responses. I show that the resulting indicator gives us a fair summary of countries' comparative climate concern at a given point in time. However, it does not do well at summarizing over-time variation in climate opinion within countries. These results show that even with a large amount of data, group-level IRT models are not guaranteed to retrieve time trends correctly, and that it is crucial to validate model outputs.

Keywords: climate, trend, model, latent, IRT, output, test