This paper speaks to a series of theoretical and empirical questions about representation, political information, and political psychology. We proceed on the assumption that some knowledge of roll call voting is a desirable feature in a democratic polity. Many models of representation and congressional action assume such a link. We provide some empirical meat to this assumption. Under conditions of highly salient roll call votes, only a quarter (or less) or a national sample is confident enough to say they can remember a particular vote, although nearly all of the remaining respondents will make a guess. Accuracy rates are much lower than we found in a similar study of the 1991 Gulf War vote, using 1991 data, and are not basically no higher than if respondents guessed ``all yea.''
We are not so bold as to claim that these two cases establish that citizens know nothing about roll call voting, and are only taking a wild guess. In fact, our inferential models demonstrate that there are predictable patterns to the probability that a respondent will claim to remember, and whether the response, whether a ``know'' or a ``guess,'' is accurate. It is not surprising that broad, contextual cues such as ideology, party, and tenure serve as cues for respondents. Only the most politically aware, or someone concerned with a specific issue, will attend to particular roll call votes. More likely, representation occurs via broad patterns of ideological agreement between the representatives and the represented (e.g. Erikson and Wright 1993; Stimson, MacKuen, and Erikson 1995).
In order to confirm these claims, there are a series of future tasks that we need to complete. Most importantly, we need to control for the nature of the campaign environment for 1992. If the Thomas vote was an issue in a state such as Pennsylvania, where Senator Arlen Specter's role as chief interlocutor with Anita Hill might have generated controversy, then we need to control for these effects in our individual level models. The differences in information flow between non--campaign and campaign environments is obviously central to any study of citizen/legislator relations, and must be delved into more seriously than we do here.
Second, there are unique features to the Clarence Thomas vote that deserve further scrutiny. In particular, the intersection of race and gender that made this such a fascinating vote should be incorporated into our analysis. Since we know from public accounts that African Americans and whites and women and men perceived Thomas's and Hill's credibility differently, it may be that they also differed in their knowledge about the Senate vote. Race, gender, interest in the issue, and ideological position should be interacted with evaluations of the member and ideological position of the member, as we did in our 1991 study.
Third, we speculate about the decline in accuracy since 1991, but have no real evidence for this decline. Better evidence is available. First, we should compare the structural parameters estimated from the 1991 data with those from the 1992 data. Since the 1991 study was part of a panel, we can estimate declines directly by taking advantage of the panel design for 1992. We hope to turn to each of these issues in subsequent versions of this work.