Problematic polling projections in the 2016 election have generated substantial interest in alternative methodologies for increasing accuracy of election forecasting. Langer Research Associates is offering half-day private workshops on a promising strategy, multilevel regression with poststratification, or MRP.

MRP has grown in popularity in recent years among researchers seeking to improve survey weighting techniques, particularly in making state-level election forecasts. By combining pre-election polls with Census data and other variables, MRP provides a powerful approach for improving opinion estimates by pooling information across groups. The technique offers a robust alternative to traditional weighting techniques, whose weaknesses may have contributed to polling errors in the 2016 elections.

Our workshops will be based on internal research that employed MRP to predict state-level election results using national tracking poll data. This approach proved highly accurate for the 2016 election, exactly predicting the estimated 2-point national popular vote margin and correctly forecasting the outcome in 49 of 50 states, all but Michigan. This approach has been tested in the last five presidential elections, correctly predicting the outcome in 46 states in 2000, 47 in 2004 and 48 in 2008 and 2012 alike.

These tutorials will be geared toward applied survey researchers and campaign professionals interested in employing this technique in their future work. Familiarity with R is recommended though not required. Presentations can be customized to clients’ needs. See a more detailed summary and please contact us for details at info@langerresearch.com