The Accuracy of Self-Reported Data and Prediction of Political Activity

Abstract

Misreported information, whether deliberately falsified or not, constitutes a source of error in most, if not all, surveys. Methodological procedures have been developed to handle the difficulties created by misinformation, and the sophisticated researcher has at his disposal a variety of techniques designed to minimize the adverse effect of invalid responses. In a gross sense, these procedures and techniques have been designed with one or both of the following problems in mind: (1) preventing the occurrence of misinformation, or (2) assessing the influence of misinformation after it has occurred. The present note is addressed to the second of these general problems. A problem frequently facing the survey researcher is that of employing a set of independent variables in an effort to describe or predict variation on some dependent variable. While error can have a number of important consequences and can enter into the situation in a variety of ways, one fundamental instance involves the case where misinformation influences the classification of respondents in terms of the dependent variable. For example, in determining the correlates of voting behavior, self-reports of such behavior have been utilized. In such instances, voters are composed of two classes of respondents: (1) those who report they have voted and who actually have voted, and (2) those who report they have voted but who actually have not voted. Under such circumstances, three questions are of particular interest. First, what is the character of the sub-population of misinformants? Second, how does such measurement error with respect to the dependent variable affect the ability to make predictions concerning that variable? Third, what is the utility of using a single item to identify inaccurate informants? The data employed in addressing these questions were derived from a larger study of the use of attitude measures as predictors of student political behavior. Attitude toward personal participation in student politics was measured using five different measurement approaches.' In addition, five indexes of political behavior were con-