State Methodology

While the EdChoice's Public Opinion Tracker is conducted among a nationally representative sample of respondents, we also set out to provide response estimates at the state level. Estimation methods can always improve based on the latest advances in the survey research industry. For this reason, starting in February 2022, Morning Consult and EdChoice employed a method known as multilevel regression and poststratification (MRP) to the Public Opinion Tracker's state data to provide even more reliable state estimates.

MRP is a method to generate sub-national estimates from national level polls. Polls conducted for the United States as a whole do not reliably allow to disaggregate results provide estimates on the state level, especially for small states. To address this methodological challenge, MRP combines different statistical techniques to generate more reliable state-level results from data collected through national surveys.

A multilevel regression estimates the responses of different demographic groups, allowing each group's response to vary by state. In this framework, small groups can also "borrow" information from larger groups through partial pooling, and additional state-level information can be incorporated as predictors. The poststratification part of the estimation process then incorporates information about the population by adjusting each group to be proportional to its population share to generate state-level estimates of the outcome variable of the regression model.

Combining these two approaches means state-level outcomes are now estimated by modeling the group-level relationships, incorporating state-level information, and projecting the results to match state-level demographics, usually via hierarchical Bayesian estimation.