As we argued in our previous paper, when information is costly and imperfect, individuals can draw upon a limited, but powerful, set of inferential cues to guide their understanding of their representative's legislative actions. So, despite the fact that people are often poorly informed about politics, they can draw upon a series of individual and contextual elements which help them place candidates on issue scales, or which help them state how they believe their representative voted on a particular issue (Alvarez 1996; Alvarez and Franklin 1994; Alvarez and Gronke 1996; Conover and Feldman 1989; Mondak 1993a, 1993b; Powell 1989).
The individual-level components of this inferential process focus on how politically informed and how media savvy the individual is. Respondents who are more politically informed and who are more media attentive are more likely to make accurate inferences (Alvarez 1996; Alvarez and Franklin 1994; Alvarez and Gronke 1996) and are less likely to use source cues in developing their inferences (Mondak 1993a).
But individuals do use source cues, and do draw upon their environment for information. This happens directly, when information is made available to them by either the representative herself --- whether in constituent contact or during a reelection campaign --- or by the mass media. But we know that information is costly and difficult to obtain, that the ways in which political information is transmitted from elite to citizen are complex and sometimes error-prone, and last, that the abilities of citizens to process information is limited by their cognitive skills. Thus, the literature now posits that the ``citizen is a cognitive miser", which really means that individuals rely on various inferential mechanisms and source cues to understand the political world (Popkin 1991).
We argue that the sort of source cues which ought to be employed by people when called upon to infer the vote of one of their representatives on a salient issue should be cues about the representative which are simple for respondents to understand --- the representative's party affiliation and ideological stance (Alvarez and Gronke 1996). From these two cues citizens may be able to draw reasonable inferences.
Thus, in our efforts to understand what might drive individual respondents to state they do know the vote of their Senators on these two salient issues, or more importantly, how accurate they are in their recall or guesses, we include variables in our statistical models which test both the individual and and contextual influences on accuracy of respondent perceptions. For individual characteristics we include two variables, media attentiveness and political information. Our second set of variables should measure contextual political information, or the sort of source cues which citizens might use to make inferences about the political world. As we argued earlier, respondents might use two readily well-known and well-understood traits of their representative --- his or her ideology and partisan affiliation --- as source cues. Also, as rough measures of the availability of political information, we included measures for whether the Senator in question faced reelection in 1992 and for the year in which the Senator first took office.
Below we use three sets of dependent variables, taken from
Tables 
and : constituent knowledge (whether they
stated they knew the vote of a particular Senator or not); the
accuracy of vote recall; and the accuracy of vote guessing. We
provide binary probit estimates for each of these dependent variables
in Tables , , and . Each of
these tables provides maximum-likelihood probit estimates and standard
errors for each vote first. We then provide estimates of the marginal
effects of each probit coefficient.
Table Goes Here
Table examines the determinants of constituent willingness
to state they knew how the Senator had voted on each issue. We see
first that both the individual-level indicators have statistically
significant, and positively-signed estimated impacts. This means that
for both the Gulf War Resolution and the Clarence Thomas vote, individuals
who paid more attention to the the mass media, and whom were more politically
informed, were significantly more likely to be willing to state that
they knew the vote of their Senators on both issues. Notice, furthermore,
that the estimated magnitude of these effects is similar for both
roll call votes.
As to source cues, we do not find strong support for their utility in this case. The coefficients on partisan agreement between constituent and representative and for the representative's ideological extremity are both positively signed, but only the partisan agreement coefficient is significant, and in only the case of the Clarence Thomas vote. There we see that if the constituent and Senator are from the same party, that the constituent is somewhat more likely to be willing to state that he or she know the Senator's vote on the Clarence Thomas issue.
For the environmental information variables, we see consistent evidence that whether the Senator was up for reelection in 1992 positively influenced the probability that an individual would claim knowledge of both roll call votes. Here, the effect of a Senator being up for reelection in 1992 is very similar for the two roll call votes. We find that tenure had a statistically significant impact in only the Gulf War case, where the negative coefficient implies that constituents are more likely to believe they know the vote of their Senators the longer the latter have been in office.
Next, in Table we present our probit results for our
model of accuracy of respondent recall of both these roll call
votes. In this case, we see surprisingly little evidence that the
individual-level variables had much effect in determining whether
a respondent could accurately recall each vote. The only statistically
significant coefficient here is that of political information in the
Gulf War vote model.
Table Goes Here
Turning to the source cue indicators, we see that the partisan agreement coefficients are positively signed, but that they are statistically insignificant. However, the coefficients on representative ideological extremity are statistically significant in both models, and are negatively signed. Thus, constituents were more accurate in their recall of the votes of Senators who are more extreme in their ideological stances. The estimated magnitude of the effect of ideological extremity is roughly similar in both vote models.
The environmental information variables also have some influence on the accuracy of constituent recall of both these votes. The coefficients on the tenure variable are statistically significant and negative in both models; this implies that constituents are about to make more accurate inferences about the voting behavior of Senators who have been in office for longer periods of time. We see a negative, and statistically significant estimate for the reelection variable in the Clarence Thomas model. This indicates that constituents were less accurate in their inferences about their Senators' vote on the Clarence Thomas appointment, controlling for everything else in this model, if the Senator were up for reelection in 1992.
Last, in Table we provide similar estimates for models
which predict the accuracy of respondent guessing about these two
roll call votes. One surprise in Table is the insignificance
of the individual-level indicators for political information and
media attentiveness. Controlling for the rest of the variables in
this model, we conclude that neither of these two individual-level
variables influenced the accuracy of respondent guessing.
Table Goes Here
We do find, though, that one of the two source cue variables --- partisan agreement --- did significantly influence the ability of respondents to guess accurately about the votes of their Senators on both these issues. The effects we estimate are positive and consistent across the two votes, which implies that being of the same party as a Senator led respondents to be more accurate in their guesses about the votes of the Senator on each issue.
The environmental information variables, though, do not have much effect on the accuracy of respondent guessing. Only in one case do we find a statistically significant coefficient. That is for the effects of tenure in the Persian Gulf War Resolution model, which is positively signed. This leads us to conclude, contrary to our earlier results, that respondents were more accurate in their guesses of the Gulf War votes of Senators who had been more recently elected.
We draw two generalizations from these probit results. The first is that individuals use each of the different inferential strategies to develop their responses to these survey questions. We found that individual-level stores of political information and attentiveness to the mass media were important determinants of both willingness to state knowledge of both these roll call votes and the accuracy of the subsequent recall of these votes. But these same individual-level variables did not influence the accuracy of guessing. Yet, both source cues and environmental information were important in each of our three models.
Second, it is informative to see that there is little difference in the inferential processes we see for these two different roll call votes. One of these roll call votes was a serious matter of foreign policy, about which the public is often thought to be quite ill-informed (but see Holsti 1992 and Aldrich, Sullivan and Borgida 1989). But the other was about a contentious struggle to appoint a new member of the Supreme Court, a struggle which involved not just the individual's qualifications for this position but also serious allegations of sexual harassment. Given the wide differences in these two issues, and our findings that the inferential process is roughly the same in each for these respondents, it seems that citizens have general tools which they use to reason about politics and that they do not develop specific inferential tools to deal with different types of issues.