What People Think: Advances in Public Opinion Measurement Using Ordinal Variables

Surveys are a central part of political science. Without surveys, we would not know what people think about political issues. Survey experiments also enable us to test how people react to given treatments. Surveys and survey experiments are only as good as the analytical techniques we as researchers use, though. This applies particularly to how we use and measure variables. For ordinal variables, some of our current measurements and techniques are insucient. Ordinal variables consist of ordered categories where the spacing between each category is uneven and not known. One example is education, one of the most important predictors of political behavior. The distances between education categories such as “Elementary School”, “Some High School”, and “High School Graduate” are not evenly spread. Current practice nonetheless often does not take this information into account. This could misrepresent the data and potentially distort results. My dissertation develops two methods to address this and applies them in original survey research. Chapter 2 develops a new method to improve the use of ordinal variables in blocking in survey experiments. Preliminary evidence suggests that the re-estimation of ordinal variable categories with an ordered probit approach might matter. Chapter 3 develops a new method to treat missing survey data with ordinal variables. The results show that the method performs worse than existing software, with exceptions in specific cases. Chapter 4 applies both methods in an online survey experiment that tests morality and self-interest in political framing. The results confirm the findings from the previous chapters and show tentative evidence for the importance of morality in issue-opposing frames.