When Good Predictions Go Bad: Vote Context, Win Margins, and Misclassified Votes in the 75th to 108th Congresses

Abstract

The logic of ideal point estimation requires that there be some errors in vote prediction, but not that these errors be random. In considering the distribution of misclassifications at the level of individual roll calls we draw from the literature on ideal point scores and parties to identify potential covariates at the vote and congress level and track these relationships in the House from the 83 rd to 108 th Congresses (1953-2004). At the individual vote level, we consider attributes related to salience and the conflictual nature of the vote, partisan divide, the vote type (i.e. procedural, amendment, final passage, etc…), issue area (appropriations, foreign policy, etc…). At the Congress level we include attributes such as majority size, whether the congress and/or government were under divided party control, and measures of party strength. Lastly we look for variation in these relationships across different periods during these 65 years. Ultimately we find support for the broad claim that context matters: procedural votes are subject to fewer misclassifications while amendments generally have more classification errors. Roll calls associated with either appropriations or foreign policy bills also have more misclassifications. Lastly, many of these relationships vary in magnitude (and in some cases, direction) across the pre-reform, postreform, and contemporary Congresses.