Since Converse (1964), researchers have worked to distinguish between political attitudes which are consequential for political behavior and those that are not. This has motivated scholars to ask not whether attitudes predict behavior, but which attitudes predict behavior? Unfortunately, much of the information needed to answer this question is lost by the way survey responses are quantified, meaning what researchers often analyze are rough approximations of fully articulated opinions. Using a multi-input neural network and the audio from three different surveys, this study not only effectively measures attitude strength from open-ended responses, but when this is done — new insights are gained regarding when and where strong attitudes are articulated. Simply put, by not differentiating between different levels of attitude strength, researchers wrongly assume all responses of are of equal quality. The measure introduced in this study not only avoids making such an assumption, but it is also embedded within software which will improve the quality of survey research by allowing scholars to automatically quantifying real-time audio streams.
Assessing Attitude Strength Using the Audio from Open-Ended Questions From Telephone, In-Person, and Online Surveys
Bryce J. Dietrich, Assistant Professor of Social Science Informatics