Predicting Congressional Elections From Expert Ratings, 1964–2022

Objective Predicting election results has become increasingly common in political science and in data‐based journalism, where polls, macroeconomic factors, and campaign finance totals are used to develop models that translate statistical patterns into vote share forecasts. Before the adoption of “big data,” though, political experts used qualitative assessments to rate elections as toss‐ups or as leaning or likely to break in a party's direction. Congressional Quarterly, for example, issued election forecasts for House and Senate races for over 60 years. These ratings were based on qualitative judgements of the likely outcome of federal races, using past election results and cycle‐specific developments. The objective in this article is to assess the accuracy of these predictions over six decades. Methods We examine House and Senate ratings between 1964 and 2022, comparing the final rating with the vote results. We also compare ratings across sources for the period between 2006 and 2022, to assess the level of agreement across sources. Results We show that qualitative assessments of races (which requires considerable knowledge of the candidates and district or state idiosyncrasies) are quite accurate in not only predicting the outcome of elections but the vote share of the candidates. Conclusion Citizens are likely to learn a lot about the likely outcome of congressional elections from predictions based on qualitative analyses of available data. Congressional Quarterly did this work reliably for many decades, and several academics and journalists have added their own predictions in recent years. By and large, they collectively hit the mark.