Improving Balance in Survey Experiments with Ordinal Variables Through Sequential Blocking

Abstract

This paper improves balance in survey experiments by providing a new method of using ordinal variables to block in survey experiments. Ordinal variables, such as education, are highly important predictors of public opinion and behavior. It is currently not possible to block on ordinal covariates in a way that incorporates their unique ordinal nature. I apply an ordinal probit model approach borrowed from machine learning that fully utilizes the ordinal nature provided. The tool can be applied to in-person and online survey experiments, as it includes code that provides a finished web-based questionnaire interface. I demonstrate the benefits of this approach with simulations, external data, and original data.

Status