1We have used the post-election rather than pre-election survey because of the ambiguity of the relevant questions in the earlier survey, which was conducted, in part, during the time when Perot had taken himself out of the race. The question asking for the candidate for whom the respondent intends to vote listed Bush and Clinton, but required that Perot be volunteered by the respondent. Respondents were asked if Perot was ever their first choice for president, but no comparable questions were asked about Clinton or Bush.

2The following assignments are made if there are ties: if u(A) = u(B) > u(C), both A and B receive an approval vote; if u(A) > u(B) = u(C), only A receives an approval vote; and if u(A) = u(B) = u(C), only the candidate reported voted for receives an approval vote. For respondents who fail to provide thermometer scores for all three candidates, approval votes in addition to the reported vote are assigned only to a candidate rated strictly higher than another rated candidate.

3Three respondents (0.2 percent), who did not provide thermometer scores for both Clinton and Bush, could not be assigned a vote for either.

4Unlike for Perot, rule (3) gives different sources of approval for Clinton and Bush than does rule (1). More specifically, because rule (3) assumes that neither Clinton nor Bush voters would approve of the other, the extra approval of these candidates can come only from Perot voters, who are assumed always to support the major-party candidate with the higher thermometer rating. Although we think this rule is more realistic for third-party candidates--like George Wallace in 1968--who have little chance of winning under AV because of their ideological extremeness and the consequent `need' for strategic voting, it is useful to include rule (3) in races with more centrist candidates like Perot. Among other things, it tends to provide a lower bound on the likely AV support of the major-party candidates. In 1992, for example, while rule (3) surely attributes too much approval to Clinton and Bush from Perot voters--who are assumed to vote for one or the other major-party candidate regardless of his thermometer rating--this inflation is more than balanced by the deflation caused by attributing no approval to the major-party candidates from each other.

5These percentages are more than doubled (to 14.4% for Clinton, 19.1% for Bush, and 36.4% for Perot) if we include all respondents who did not vote for a candidate to whom, nevertheless, they gave their highest rating. Under this broader definition of insincerity, our conclusion does not change--Perot voters remain decidedly more insincere than Clinton or Bush voters. Still other respondents either failed to rate the candidate for whom they voted, or gave more than one candidate their highest rating. For the latter reason, the total number of highest-rating responses, 1,882, exceeds the total number of voters in the survey (1,658).

6Perot was perceived to have `won' the first debate October 11, Clinton the second on October 15, and Perot and Clinton the third (ending in a tie) on October 19. Bush came in last on all except the second debate, in which he came in second (Francovic, 1993, p. 120).

7Black and Black (1993) report that enough voters might have switched to Perot--if they thought he had a chance to win--to put him in the lead. This observation is, however, based on the responses to a single question in an exit poll, which in isolation are hard to interpret, as is pointed out in a rebuttal by Hugick (1993).