1. Naturally, crossover voting occurs up and down the ballot, in well-publicized statewide contests as well as in local elections, such as those for the State Assembly and Senate. Cain (1997) has suggested that crossover voting may even be more prevalent in down-ballot races, and Cohen, Kousser, and Sides (1999) find comparable levels of crossover voting in 1998 California legislative races and Washington State Senate elections under its blanket primary. Though the Field Institute poll included questions about lower offices, such as Attorney General, in its pre-primary surveys, it did so only intermittently. Furthermore, such questions typically produce a great deal of missing data, since many respondents have no opinion about these races, especially months before election day. For example, in the April 1998 Field poll, 62.5 percent of respondents did not express a preference for Attorney General, even after the names and party affiliations of the candidates were read twice. [BACK]

2. In the closed primary, all voters could vote for propositions and nonpartisan offices, but only those registered with political parties could vote in their party's contest. [BACK]

3. In California, the official designation is "Decline-to-State." [BACK]

5. The Field poll uses the conventional measure of party identification with an initial question and then a two-stage partisan probe. The first question asks, "Generally speaking, do you usually think of yourself as a Republican, a Democrat, an Independent, or what?" The follow-up asks, "(If Republican or Democrat) Would you call yourself a strong (Rep/Dem) or a not very strong (Rep/Dem)? (If Independent, Other, or No Preference) Do you think of yourself as closer to the Republican or Democratic party?" One squabble in the literature on crossover voting centers on whether independence should include partisan "leaners" as well as pure independents. Hedlund and his colleagues (Hedlund, Watts, and Hedge 1982; Hedlund and Watts 1986) argue that they should not. However, given the compelling evidence presented in Keith et al. (1992), Wekkin (1988) argues that they should. We agree with the latter camp and thus define party identifiers as strong, weak, or leaning partisans. [BACK]

6. Obviously, since all of these polls occurred before the election, they are not measuring crossover voting but intended crossover voting. In discussing results from the Field polls, we will occasionally remind the reader that this analysis concerns

7. These estimates of the magnitude of crossover were generated using only those respondents who expressed a preference for a gubernatorial candidate. [BACK]

8. There was more of a discrepancy between the magnitude of crossover voting by identification and by registration when independents were included. For example, in the Governor's race, the difference was roughly 7 percent (21.8 percent vs. 28.6 percent). This larger discrepancy derives from the greater number of independents as defined by party registration (15.5 percent of the pooled Field poll sample) compared to the number of independents by party identification (11.2 percent). However, since we focus primarily on the measures of crossover that exclude independents, this discrepancy is not presently significant. [BACK]

9. We calculated the percent voting for candidates of the other party using the last pre-election Field poll in each election year from 1984 to 1996. In each of these years the last preelection Field poll deviated no more than two percentage points from the actual election results, and the outcome of each race was predicted correctly. [BACK]

10. In a general election, hedging as defined here cannot occur, since there is no future election in which the hedger can return to his or her own party. However, voters may not vote sincerely in a general election if they vote for their second choice because their first choice is likely to lose. In doing so, voters avoid "wasting" their votes. [BACK]

11. The composition of voters in the 1998 primary and in past general elections was also similar. Looking at likely voters in the last Field poll before the primary and general elections, we found nearly identical distributions of liberals and conservatives, as well as Democrats and Republicans (data not shown). This complements the earlier finding that primary voters are not much different from primary nonvoters (Ranney and Epstein 1966; Robeck 1984). [BACK]

12. This is a collapsed version of the standard, seven-point party identification scale, with leaners included as partisans. [BACK]

13. The results from the May 1998 Field poll are a bit problematic because the poll did not include Frank Riggs, who won roughly 10 percent of the vote, among the Republican candidates. Fong's actual winning margin was 45 percent to 40 percent. [BACK]

14. The numbers of both decline-to-states and minor-party registrants who participated in the Republican primary were too modest to affect the outcome, even if they had all voted for Issa (data not shown). [BACK]

15. Evidence from a Los Angeles Times poll of October 1997 suggests the possibility of policy-driven crossover. When asked which party could do a better job of handling issues ranging from the economy to education and immigration, between 7 and 27 percent of Democrats and Republicans felt that the other party would do a better job. [BACK]

16. However, since both the gubernatorial and senatorial races featured female candidates, we do not have variation by contest. Analysis of more offices and of other elections would provide a richer understanding of this issue. [BACK]

17. Union membership captures whether the respondent or someone in the respondent's family is a union member. [BACK]

18. When estimating these logit models, we analyze only partisans, and thus this measure takes on only three values in reality. [BACK]

19. However, Lungren nominally opposed Prop 227 and, according to his campaign staff, sought to run a campaign that would attract minority voters (see Lubenow 1999). [BACK]

20. We performed several replications to test the robustness of the results. The first includes only the last three Field polls in the sample, since the first poll, which mentioned potential candidates (e.g., Riordan and Wilson) who never ran, may have produced anomalous results. The next replication includes only the sample from the May 1998 Field poll, conducted just before the election when vote preferences presumably had solidified. The third replication includes only those respondents who identified themselves as likely voters, since they might have a different motivation for crossing over. The last replication employs a new dependent variable, the measure of crossover voting using respondent's party registration instead of party identification. In general, the results of all these replications are highly consistent with the reported results. We footnote any discrepancies below. [BACK]

21. For the most part, the replications confirm these results, although the effect of partisanship wanes in the first replication (March–May polls only). Interestingly, in the second replication, which includes only the poll closest to the primary election, a yes vote on Prop 227 is a significant predictor of crossover voting. As expected, its effect is negative: a vote for the proposition (i.e., against bilingual education) is associated with a declining probability of crossover voting among Republicans. [BACK]

22. The replications produce comparable results. One difference is that the effect of ideology increases when the May 1998 poll is analyzed separately, as does the effect of union membership, somewhat curiously. The effect of education wanes to a statistically insignificant level in this replication. [BACK]

23. The replications largely confirm these results, except that in the May 1998 version only ideology is statistically significant. [BACK]

24. These numbers come from the same Field poll dataset analyzed in the previous section. [BACK]

25. One reason for the observed recall bias is that the Field question only listed the major candidates. The senatorial recall question, for example, did not list Frank Riggs or "other Republican candidate" as one of the options. On election day, Riggs received 10 percent of the votes for Republican candidates. [BACK]

26. We rely on party registration here because the November Field poll did not include the standard questions about party identification. [BACK]

27. We described above how during the primary campaign similar trends in candidate preference emerged among crossover and loyal party voters, and we took this as evidence of sincere voting. There were also similar trends in recall bias among crossover and loyal voters, suggesting that the same factors shaped the memories of both types of voters. Seventy percent of Republicans who claimed to have voted for a Democratic candidate in the primary recalled voting for Davis, as did 76 percent of Democrats, and 90 percent of decline-to-states. (For both Republicans and Democrats,

28. Overall, 92 percent of voters who voted for a Democrat or Republican in the gubernatorial primary voted for the same party's nominee in the general election. In the senatorial race, 95 percent voted for a candidate of the same party in both elections. [BACK]

29. In making this assumption, we ignore the possibility that a voter may have changed his or her mind during the general election campaign and come to support Davis over Lungren. Given the large and constant lead Davis maintained over Lungren during the campaign, such a switch does not appear very common. [BACK]

30. To be crystal clear: this percentage comes from adding up the number of voters who returned home (44 + 40 + 7=91) and dividing it by the total number of Republicans who said they crossed over in the gubernatorial primary (91 + 157=248). [BACK]