---

5. The Causes and Consequences of Crossover Voting in the 1998 California Elections

John Sides, Jonathan, Cohen, and Jack Citrin

INTRODUCTION

Both advocates and opponents of the blanket primary believed that the change in rules could affect voting behavior, candidate attributes, campaign strategies, and ultimately election outcomes. In this chapter, we explore voting behavior, in particular the much-discussed, much-anticipated, and, in some quarters, much-maligned phenomenon of crossover voting—the act of voting for a candidate outside one's own party. Drawing upon a series of pre-and post-primary surveys conducted by the Field Institute as well as the Los Angeles Times primary election exit poll, we examine California's 1998 gubernatorial and U.S. Senate races.^[1] Of course, with only one blanket primary having taken place in California, its full impact is impossible to determine. However, our analysis of the 1998 elections suggests that the blanket primary leads to neither the millennium envisaged by its advocates nor the apocalypse predicted by its detractors.

We first discuss crossover voting in relation to two manifestations of partisanship: party registration and party identification. Our analysis then addresses the following empirical questions: How much crossover voting occurred in the primary election? For whom did crossover voters vote? Did crossover voting affect the election outcome? Finally, what motivates crossover voting?

CONCEPTUALIZING CROSSOVER VOTING

Crossover voting is, on its face, a simple notion: voting for a candidate outside of one's political party. However, this basic definition masks several more complicated issues (see Wekkin 1988). First, how should the political

---

party to which a person belongs be defined? One possibility is party registration. Under this definition, voters cross over when they vote for a candidate from a party other than the one in which they are registered. The blanket primary, in doing away with the party primary, makes party registration electorally superfluous. There is effectively only one primary, and any registered voter can participate fully.^[2] The virtue of defining crossover voting based on party registration is that one can estimate what would likely have happened had the California primary still been closed. That is, one can "separate" the winning candidate's vote share received from minor-party, nonpartisan, and crossover voters, and then, given certain assumptions, gauge the chances of his or her winning this simulated primary. Defining crossover voting in terms of party registration therefore provides useful counterfactual comparisons across electoral regimes.

Another alternative is to define crossover voting based on party identification. Whereas party registration is a legal formality, party identification is a psychological construct, an enduring tie between a citizen and a particular political party. Party identification does not entirely determine one's vote, and therefore affiliating with a party is conceptually distinct from voting for that party's candidate (Miller and Shanks 1996). Nevertheless, if party identification constitutes, in V. O. Key's phrase, a "standing decision," then voting against one's normal affiliation in the absence of any strategic consideration implies a weakening of loyalty, either temporary or permanent. Because voters vary in the intensity of their party identification, defining crossover voting with this as the reference point allows for a deeper analysis of its underlying psychology, since one can compare strong and weak partisans.

Another definitional issue concerns the self-styled political independent. In the case of party registration, "independents" include both members of minor parties and those who do not register with any party ("nonpartisans").^[3] In the case of party identification, "independents" profess no attachment to any party. Strictly speaking, any independent (however defined) who votes for a Democrat or Republican engages in crossover voting (Adamany 1976). Under California's closed primary system, major-party primaries were restricted to voters registered in those parties, and minorparty members, nonpartisans, and members of the opposing major party could not participate. A blanket primary system effectively enfranchises these groups and thus creates a population of nonparty members who can vote for a major party's candidates.^[4] However, because Democratic and Republican party officials worry mostly about potential mischief by the major opposition party, not about the votes of minor-party members and nonpartisans, we focus primarily on crossover voting among Democrats and Republicans.

In California, party registration and party identification are closely interrelated.

---

When the respondents from the four Field polls conducted before the primary are pooled (N ? 4,060), 91 percent of both registered Republicans and registered Democrats had a consonant party identification.^[5] A similarly strong relationship emerges when we construct measures of crossover voting.^[6] These variables are simply dichotomous, coded 1 if a respondent crossed over (i.e., a registered Republican or a Republican party identifier voted for a Democratic candidate, and vice versa) and 0 if not. In the case of crossover voting for Governor, the correlation between these two measures was .75. In the U.S. Senate race, the correlation was .71. Crosstabulating the two measures demonstrates that 93 percent of party identifiers who intended to cross over in their vote for Governor also intended to cross over vis-à-vis their party registration (the comparable figure for the Senate race was 91 percent).

THE MAGNITUDE OF CROSSOVER VOTING

We calculated overall estimates of crossover voting in the 1998 primary election for both the Governor's and Senate races, again analyzing the pooled Field poll dataset. In light of the measurement issues raised in the previous section, we calculated these measures using both party registration and identification, and then both including and excluding independents.^[7] In the gubernatorial race, 15.5 percent of respondents who identified with a major party planned to cross over by the identification measure, compared to 16.6 percent by the registration measure. The magnitude of crossover voting was similar in the Senate race: 13.7 percent as defined by identification and 14.8 percent as defined by registration. When independents were counted as crossover voters, the magnitude of crossover voting naturally increased (to the 20 to 30 percent range).^[8]

However measured, the extent of crossover voting is quite substantial: in both the gubernatorial and senatorial races, about one in six voters said they would choose a candidate outside their own party. Whether crossover voting changed electoral outcomes obviously depended on the circumstances of individual races. Nevertheless, the observed level of crossover voting can be put in some perspective by comparing it with the level of crossover voting in past general elections. Blanket primaries and general elections possess similar structural constraints (or lack thereof) in that all registered voters can select any candidate for any office. It is therefore reasonable to expect primary elections to resemble general elections in some respects. This claim is born out in the aggregate level of crossover voting. The number of partisan crossover voters in the 1998 primary was quite similar to the traditional level of defection in general elections (DiCamillo 1998). In the twelve presidential, gubernatorial, and senatorial races from 1984 to 1996, 20.8 percent of registered Democrats voted for

---

Republican candidates and 14.3 percent of registered Republicans voted for Democratic candidates.^[9]

The comparability of crossover voting in the blanket primary and past general elections suggests something about the motivation for crossover voting. In general elections, raiding or hedging are not relevant strategies because the election determines who governs and thus most voters sincerely select the candidate they prefer to hold office.^[10] Because a segment of each party's registered voters regularly prefers candidates from the other party, the blanket primary, instead of creating partisan mischief, may simply allow these voters to express their true preferences earlier in the election season.^[11]

CROSSOVER VOTING IN THE PRIMARY CAMPAIGN

How did crossover voters cast their ballots in the 1998 California primary? Tables 5.1 and 5.2 present the distribution of votes for Governor and U.S. Senator, broken down by a three-point party identification scale.^[12] To portray the dynamics of voter preferences as the campaign unfolded, we present a separate distribution for each of the four pre-primary Field polls, beginning in February and ending in May.

The Gubernatorial Nomination Campaign

One dynamic of the campaign, as in most campaigns, was the winnowing of potential candidates. The earlier Field polls presented respondents with a broader array of candidates than did the later polls, when the candidate pool had narrowed. A second dynamic is the electoral fortunes of the candidates over time, as presented in the "Total within Party" column of table 5.1. Here, the major story was the come-from-behind victory of Lieutenant Governor Gray Davis in the Democratic primary. In the March poll, he lagged significantly behind both airline millionaire Al Checchi and Congresswoman Jane Harman, but by May he garnered more than half (51.3 percent) of the votes for Democratic candidates.

Tables 5.1 and 5.2 also track crossover intentions over time and by party identification. Though the large majority of partisan voters did not cross over, a notable fraction did, even in the earliest Field poll. In February 1998, 19.8 percent of Republicans and 7.6 percent of Democrats preferred a gubernatorial candidate in the other party. In the Governor's race, crossover voting was primarily a Republican phenomenon. This is not unexpected, since one might anticipate more crossover voters where there is electoral "action," that is, in races with several serious candidates. In the gubernatorial election, the Democratic race was contested, while the Republican, State Attorney General Dan Lungren, ran essentially unopposed.

---

This arguably created an incentive for Republicans to cross over, since the outcome of the Republican primary was predetermined. Whether this incentive stimulated raiding, hedging, or sincere voting is explored below.

The magnitude of intended crossover voting in the gubernatorial race was quite stable during the four months before the primary, hovering around 20 percent among Republicans and 5 percent among Democrats. This stability suggests that crossover voting was not much affected by campaign events, by the shifting electoral fortunes of various candidates, or by the reduction in undecided voters over time. There seemed to be some fraction of voters in each party who were ready and willing to cross over, even as the likely winner in each race changed.

In comparing the vote choice of crossover voters and noncrossover voters, two findings emerge. First, and most important, similar trends affected vote intention within each group. In the Democratic race, Davis gained support over time among Democrats, independents, and Republicans alike. By May 1998, he won pluralities of all three groups. While both Republican crossovers and independents demonstrated a greater and more durable preference for Checchi than did Democrats, the similarity of the trends across parties suggests that most crossover voting was a genuine response to the perceived qualities of the competing candidates.

The Senatorial Nomination Campaign

As table 5.2 shows, the U.S. Senate race mirrored the Governor's race in that it featured one competitive and one uncompetitive primary. In this case, the Democratic nomination was a foregone conclusion: the incumbent Barbara Boxer was the only serious candidate. All of the action was on the Republican side, where the race became a duel between State Treasurer Matt Fong and car alarm magnate Darrel Issa, which Fong ultimately won.^[13]

Just as in the gubernatorial race, this disparity in competition resulted in asymmetric levels of crossover voting. As expected, there were fewer Republicans than Democrats crossing over. By May the fraction of crossover voters had shrunk to 8.6 percent among Republicans but had grown to 15.8 percent among Democrats. As in the Governor's race, trends in candidate preference appeared in all partisan groups. At first, the vote choice of Republicans, independents, and Democrats was somewhat different, but as the campaign unfolded Fong gradually became the preferred Republican among each group.

In sum, the locus of crossover voting varies with the competitiveness of the contest both between parties and within each party. Crossover voting tends to increase in primaries with asymmetric competition, as voters desert their own party's uncompetitive race to participate in a competitive race in the

---

TABLE 5.1 Preference for Governor, by Party Identification

SOURCE: Field Institute, Field Polls (San Francisco: The Field Institute, 1998). NOTE: The total crossover rates differ from those presented earlier in the text because here they are expressed as a percentage of all voters, including those without a preference (undecideds).
	February 1998				March 1998
	Republican (N = 318)	Indepedent (N = 56)	Democrat (N = 355)	Total within Party	Republican (N =496)	Indepedent (N = 117)	Democrat (N = 565)	Total within Party
Republican	57.9	19.6	7.6	100.0	47.4	7.7	4.6	100.0
Lungren	36.5	7.1	1.7	56.8	45.0	6.8	3.4	92.6
Riodran	21.4	12.5	5.9	43.2
Peron					2.4	0.9	1.2	7.4
Democrat	19.8	32.2	65.6	100.0	20.1	28.1	65.0	100.0
Checchi	10.4	12.5	16.3	31.2	10.9	17.9	20.0	37.6
Davis	4.7	7.1	22.5	31.5	3.4	3.4	17.9	24.4
Harman	0.3	1.8	4.5	5.7	5.8	6.8	27.1	38.0
Panetta	3.8	5.4	18.6	25.8	—	—	—
Vasconcellos	0.6	5.4	3.7	5.7	—	—	—
Other	—	—	—		3.4	6.8	3.5
Undecided	22.3	48.2	26.8		29.0	57.3	26.9
	TOTAL CROSSOVER:9.0%				TOTAL CROSSOVER: 8.9%
― 83 ―
	April 1998				May 1998
	Republican (N = 353)	Indepedent (N = 99)	Democrat (N = 478)	Total within Party	Republican (N = 279)	Indepedent (N = 62)	Democrat (N = 373)	Total within Party
Republican	51.6	6.1	5.0	100.0	65.9	14.5	5.1	100.0
Lungren	51.6	6.1	5.0	100.0	65.9	14.5	5.1	100.0
Riordan
Peroin
Democrat	21.5	48.4	68.7	100.0	22.6	48.4	80.1	100.0
Checchi	11.9	23.2	24.7	40.5	7.5	16.1	15.0	22.2
Davis	6.8	14.1	26.2	36.0	9.7	19.4	43.4	51.3
Harman	2.8	11.1	17.8	23.5	5.4	12.9	21.7	26.5
Panetta	—	—	—		—	—	—
Vasconcellos	—	—	—		—	—	—
Other	3.1	7.1	2.7		3.6	11.3	4.0
Undecided	23.8	38.4	23.6		7.9	25.8	10.7
	TOTAL CROSOVER: 10.8%				TOTAL CROSSOVER: 11.5%

---

TABLE 5.2 Preference for Governor Senator, by Party Identification

SOURCE: Field Institute, Field Polls (San Francisco: The Field Institute, 1998). NOTE: The total crossover rates differ from those presented earlier in the text because here they are expressed as a percentage of all voters, including those without a preference (undecideds).
	February 1998				March 1998
	Republican (N = 318)	Indepedent (N = 56)	Democrat (N = 355)	Total within Party	Republican (N = 496)	Indepedent (N = 117)	Democrat (N = 565)	Total within Party
Democrat	9.4	35.7	67.6	100.0	47.4	7.7	4.6	100.0
Boxer	9.4	35.7	67.6	100.0	47.4	7.7	4.6	100.0
Republican	73.8	32.1	17.4	100.0	51.7	14.5	8.8	100.0
Fong	11.6	7.1	3.7	17.1	19.2	3.4	4.2	38.1
Issa	13.8	7.1	2.8	18.4	25.8	9.4	3.4	48.9
Riggs	2.5	0.0	0.8	3.5	6.7	1.7	1.2	13.0
Wilson	45.9	17.9	10.1	61.0	—	—	—
Other	—	—	—		5.4	7.7	3.4	13.0
Undecided	16.7	32.1	14.9		32.7	47.0	115.6
	TOTAL CORSSOVER: 9.2%				TOTAL CROSSOVER: 7.2%
― 85 ―
	April 1998				May 1998
	Republican (N = 353)	Indepedent (N = 99)	Democrat (N = 478)	Total within Party	Republican (N = 279)	Indepedent (N = 62)	Democrat (N = 373)	Total within Party
Democrat	11.6	30.3	69.7	100.0	8.6	38.7	73.7	100.0
Boxer	11.6	30.3	69.7	100.0	8.6	38.7	73.7	100.0
Republican	53.9	23.2	10.9	100.0	73.8	33.9	15.8	100.0
Fong	24.4	12.1	6.1	47.7	38.0	21.0	9.4	53.8
Issa	29.5	11.1	4.8	52.3	35.8	12.9	6.4	46.2
Riggs	—	—	—		—	—	—
Wilson	—	—	—		—	—	—
Other	4.2	6.1	5.0		2.9	8.1	2.9
Undecided	30.3	40.4	14.4		14.7	19.4	7.5
	TOTAL CROSSOVER: 10.0%				TOTAL CROSSOVER: 11.6%

---

opposite party. Furthermore, within a given party's primary, the candidate preferences of crossover voters ebb and flow with candidate fortunes much as do party members' preferences. Since the trends in the preferences of loyal, presumably sincere partisans and crossover voters were very similar, it seems unlikely that many crossover voters were raiding the other party's primary in 1998.

THE IMPACT OF CROSSOVER VOTING ON THE PRIMARY ELECTION'S OUTCOME

Several studies of crossover voting (e.g., Hedlund, Watts, and Hedge 1982; Hedlund and Watts 1986) conclude that crossover voting rarely changes the outcome of a primary election. Others (Adamany 1976; Wekkin 1988) argue that since crossover voters' preferences usually differ significantly from those of same-party voters, they may have an indirect impact on electoral outcomes by influencing factors such as a candidate's momentum and fund-raising capacity. The present analysis examines the election tally itself. It draws upon the Los Angeles Times primary election exit poll to address the counterfactual question of whether Gray Davis and Matt Fong would have won under a closed primary, where crossover voting is by definition impossible.

The Governor's Race

The top panel of table 5.3 shows that Davis won among exit poll respondents with 57.4 percent of the votes for Democratic candidates, a result quite close to the final tabulation (57.6 percent). If the old closed primary rules had been in effect, all Democratic candidates would have lost the votes of registered Republicans, decline-to-states, and members of minor parties. It appears that Davis would have won even with this restriction, other things equal. First, looking only at Democratic voters, Davis beat out his opponents with 59.3 percent of the vote. Second, even if all of the registered Democrats who crossed over and voted for Lungren had voted for either Checchi or Harman, Davis still would have won. Among Democrats in the exit poll, Davis's margin of victory over Checchi, his nearest challenger, was 994 respondents. There were only 224 Democratic crossovers in this sample, so even if every one of them had voted for Checchi rather than Lungren, Davis would have won handily.

Another indication that the new rules did not affect the outcome is the pattern of preferences among voters who were not registered Democrats. Davis won a majority or near-majority of the votes of Republican crossovers, decline-to-states, and minor-party members, so it seems quite certain that these voters did not help elect a different candidate than the one Democrats themselves preferred.

---

TABLE 5.3 Primary Vote Choice, by Party Registration

SOURCE: Los Angeles Times Primary Election Exit Poll, June 1998 (Los Angeles: Los Angeles Times Poll).
Vote for Democratic Gubernatorial Candidates
	Vote Choice
Party Registration	Checchi	Davis	Harman	Other Democrat	Total
Democrat	18.5%	59.3%	20.8%	1.4%	100%
Decline-to-state	18.0	59.3	21.7	1.0	100
Republican	29.5	49.7	19.1	1.7	100
Other party	29.7	51.7	16.0	2.5	100
TOTAL	20.8	57.4	20.4	1.4	100
Vote for Republican Senate Candidates
		Vote Choice
Party Registration		Fong	Issa	Other Republican	Total
Democrat		56.0%	31.5%	12.5%	100%
Decline-to-state		59.8	35.0	5.2	100
Republican		45.0	42.0	13.0	100
Other party		39.8	44.1	16.1	100
TOTAL		47.3	40.0	12.7	100

The U.S. Senate Race

According to the Los Angeles Times poll, Fong beat Issa in the Republican primary 47 percent to 40 percent, a result that overestimated Fong's actual margin of victory by about 2 percent. The lower panel of table 5.3 shows that Fong probably would have won under California's previous closed primary regime as well. First, Fong won among Republicans (i.e., the electorate in the hypothetical closed primary). At the same time, Fong's margin of victory among Republican voters (45 percent vs. 42 percent for Issa) was markedly smaller than it was among Democrats or decline-to-states. In this case, Republican crossovers, had they been limited to voting in the Republican race as a closed primary dictates, theoretically could have changed the outcome. In the exit poll sample, Fong beat Issa by only 156 votes among Republican voters; Republican crossovers numbered 219.

However, the interplay of two factors reduces the likelihood that these Republican crossovers would have changed the outcome: these voters were quite liberal (38 percent classified themselves as such), and Fong appeared more attractive than Issa to liberals. Among Fong voters, 17 percent classified

---

themselves as liberals, 12 percent as conservatives, and the vast majority as moderates. The comparable percentages among Issa voters were 12 percent liberal and 40 percent conservative, with many fewer moderates. Thus, it appears that Republicans who voted for Boxer would have been more likely to vote for Fong than Issa in a closed primary (assuming constant turnout). Similarly, the 378 Democrats who voted for a Republican candidate could have changed the outcome of the Republican primary if they had favored Issa over Fong by a significant amount. However, Democrats, like Republicans, favored Fong and did so by even larger margins than Republicans did.^[14]

In sum, while crossover voting was substantial, it does not appear to have altered the outcome of either the gubernatorial or the senatorial race in California in 1998. The distribution of candidate preferences among crossover voters in each primary was similar to the distribution among all voters. Moreover, because both Davis and Fong garnered the support of their respective party faithful, they probably would have won in a closed primary as well.

THE MOTIVATION BEHIND CROSSOVER VOTING

We now analyze the motivational basis of crossover voting, using the tripartite typology introduced in chapter 1. One motivation is sincerity, which means that voters select the candidate they like best. A second motivation is hedging. In this case, voters actually prefer a candidate in their own party, but cross over to select their favorite candidate in the other party, thereby hedging their bets in the general election. If their preferred candidate in each party wins the nomination, then no matter who wins the general election, hedgers will find the outcome acceptable. A third motivation is raiding. Raiders vote for the putatively weakest candidate in the opposing party in hopes of helping their own party's candidate win the general election. The incentive for organized raiding—by which outsiders could determine the nominee—is what political party organizations fear most about the blanket primary.

Previous studies have found little evidence of hedging and raiding in American primary elections (Hedlund, Watts, and Hedge 1982; Hedlund and Watts 1986; Abramowitz, McGlennon, and Rapoport 1981; Southwell 1991; Wekkin 1991). One hypothesized reason is that most voters lack the political sophistication to vote strategically, particularly in a state like California, where the length and complexity of the ballot challenges the interest and capacity of most citizens. Knowing how to hedge or raid is thus not always obvious, and organized efforts to mobilize voters to act in these ways are likely to be difficult and costly. If so, then crossover voting should be mostly sincere.

---

A Los Angeles Times Poll conducted in October 1997 provides initial support for this conclusion. Among the respondents, 90 percent of registered Democrats and 82 percent of registered Republicans said the most likely reason that they might vote for a candidate of another party was simply that they favored that person. Only 7 percent of Democrats and 5 percent of Republicans said they would be likely to raid. Eighty-seven percent of the decline-to-state respondents also said they would choose their most preferred candidate.

Theoretical Expectations

The analysis undertaken here assumes that most crossover voting is sincere or possibly hedging. Essentially, we propose that crossover voting will be more likely among respondents with attitudinal or demographic attributes that deviate from the core constituencies of their party—for example, a conservative Democrat or a female Republican. However, it is important to point out that crossover voting under these circumstances could reflect either sincere voting or hedging. A conservative Democrat could cross over because he or she genuinely prefers the views of the Republican candidate. Likewise, because this conservative Democrat sits somewhere "in between" the two parties, he or she could also hedge to ensure that if the Democrat lost the general election, the Republican victor would be entirely palatable. Given these preliminaries, we can then ask, What are the likely characteristics of crossover voters?

One obvious hypothesis is that the strength of party identification should influence the likelihood of crossover voting. Strong partisans should be less likely to cross over than weak partisans. Similarly, crossover voting should be more likely among those whose ideology is out-of-step with the dominant outlook of their party. Thus, conservative Democrats should cross over to the Republican party at a higher rate than do liberal Democrats. In the same vein, crossover voting might also have some specific issue content. For example, pro-choice Republicans might have an incentive to vote for a Democratic candidate if the Republican candidates are explicitly pro-life (as Dan Lungren was in the gubernatorial primary).^[15]

It is quite likely that ideology, partisanship, and policy preferences do not exhaust the reasons for crossover voting. Because demographic variables may function as proxies for political values and interests, they may affect crossover voting as well. In the 1998 California elections, gender may have played a role, as numerous pundits speculated about the attractiveness of Democrats like Jane Harman and Barbara Boxer to female Republicans. The presence of female candidates thus may cue gender considerations otherwise absent in an all-male race.^[16] However, the effects of gender might vary across party, "pushing" Republican women to cross over more than

---

men without exerting a comparable "pull" on Democratic women to keep them in the party. Gender issues might also push Democratic men to defect.

Race and ethnicity could also have impacts on crossover voting. Blacks and, to a lesser extent, Latinos are predominantly Democratic constituencies. Black or Latino Democrats should be less likely to cross over than their white counterparts, especially because the California Republican party is often identified with conservative stances on affirmative action and immigration that tend to alienate racial minorities. The ethnic background of candidates themselves might make voters' ethnic ties even more salient. However, in June 1998, only one nonwhite, Matt Fong, was a major candidate. It is nevertheless possible that Asian voters in particular were attracted to Fong.

Similarly, given that high income is generally associated with a Republican vote, rich Democrats may be more likely to cross over to the Republican party, and rich Republicans less likely to cross over to the Democratic party. In the current electoral climate, religiosity may also benefit Republican candidates. Union membership could function the same way for Democrats, especially in an election featuring the anti-union Proposition 26.

Two variables, age and education, do not generate clear directional hypotheses. Age could have a consistently negative impact on the propensity to cross over regardless of party, if one assumes that partisan loyalties, whether Democratic or Republican, ossify with age. However, to the extent that age is associated with increasing conservatism, as is the conventional wisdom, it should associate negatively with crossover voting among Republicans, but positively with crossover voting among Democrats. As for the influence of formal education, one hypothesis is that it will have a consistently negative impact on crossover voting, since educated voters could be more dedicated partisans. However, it is probably also true that educated voters more diligently consume political information, and thus are likely to learn about and perhaps support candidates in the opposing party. Furthermore, educated voters' greater cognitive capacities may provide the wherewithal to vote strategically. Education could also have a varying impact on crossover voting among Democrats and Republicans if it is generally associated with a loyalty to one party or another—e.g., if more education leads to a stronger Democratic party identification.

Data and Measures

To test the hypotheses spelled out above, we estimate multivariate models of crossover voting. To ensure an adequate number of cases for analysis, we rely on the pooled Field poll dataset. The choice of dependent variable again depends on whether we define crossover voting as voting against

---

one's party identification or party registration. Since an analysis of motivations is in essence an examination of the psychology of voters, crossover voting is measured as defecting from one's party identification. Thus, the dependent variable is coded one if the respondent intends to vote for a candidate in a party other than the one with which he or she identifies, and zero otherwise. Among our predictors, gender; ethnic identification as black, Latino, or Asian; and union membership are dichotomous variables.^[17] Religiosity is measured with two dummy variables, one for Protestants and one for Catholics. Age is the number of years, education a ten-point scale, and income a five-point scale. (Detailed descriptions of variable coding are provided in the appendix to this chapter.)

Ideology is measured by a seven-point self-identification scale, with a score of one representing a strong conservative and seven a strong liberal (thus we refer to the measure as "liberalism"). Strength of partisanship is measured by "folding" the standard seven-point party identification scale at the mid-point to create a four-category measure, where one indicates independents, and four indicates strong partisans.^[18] The only consistently available indicators of issue positions in our dataset are questions about two highly contested ballot propositions on the June ballot, Propositions 226 and 227. Proposition 226 would have mandated that unions obtain the permission of all members before spending their dues for political purposes. Proposition 227 radically limited bilingual education programs in California public schools. Proposition 226 failed by a narrow margin, while Proposition 227 passed easily. Although this interpretation is somewhat crude, we construe a vote for either of these propositions as conservative, and thus more congruent with a vote for a Republican candidate.^[19]

Multivariate Results: The Governor's Race

The multivariate analysis is comprised of a series of logit models, one for each party-contest combination.^[20] Table 5.4 presents the results for the gubernatorial race. The logit model for Republican crossover voting in the first column confirms several of our expectations. For one, the coefficient for liberalism is statistically significant and positive. As Republicans became more liberal, the probability of a crossover vote increased. Furthermore, the coefficient for strength of partisanship is negative and statistically significant, indicating that increased partisanship was associated with a declining probability of crossing over. The coefficient for gender is also significant and positive; other things equal, Republican women were more likely than their male counterparts to cross over in the gubernatorial race. By contrast, income is negatively associated with a crossover vote. The wealthier the Republican voter, the less likely he or she was to vote for a Democrat in the primary. Votes for both Prop 226 and Prop 227 are

---

TABLE 5.4 Logit Models of Crossover Voting for Governor

	Republicans		Democrats
	Coefficient	Probability	Coefficient	Probability
SOURCE: Field Institute, Field Polls (San Francisco: The Field Institute, 1998). NOTE: Table entries are logit coefficients, with standard errors in parentheses. The dependent variable is coded 1 for a crossover vote and 0 otherwise. The rate of Republican crossover was 27.8 percent, and the rate of Democratic crossover was 7.1 percent. Change in probability is the change in the probability of crossover voting associated with a shift from the minimum to the maximum value of the independent variable, holding all other variables at their mean values. *p < .05 **p < .01 ***p < .0001
Liberalism	0.18*	.27	−0.27**	−.07
	(0.07)		(0.11)
Strength of partisanship	−0.63***	−.30	−0.14
	(0.14)		(0.20)
Prop 226 vote	−0.32		1.03**	.05
	(0.22)		(0.34)
Prop 227 vote	−0.12		0.14
	(0.25)		(0.34)
Gender	0.41*	.10	−0.34
	(0.20)		(0.31)
Protestant	−0.19		−0.44
	(0.23)		(0.38)
Catholic	−0.06		−0.24
	(0.27)		(0.39)
Income	−0.22**	−.18	0.23*	.04
	(0.08)		(0.12)
Age	−0.005		0.008
	(0.006)		(0.01)
Education	−0.04		−0.16*	−.07
	(0.05)		(0.08)
Black	—		−0.59
			(0.56)
Latino	—		−0.79
			(0.54)
Union member	—		0.66*	.03
			(0.31)
Constant	1.98**		−1.55
	(0.73)		(1.11)
−2 x log-likelihood	664.3		339.6
Percentage correctly predicted	74.6		92.9
No crossover	94.9		100.0
Crossover	21.8		0.0
Pseudo-R 2	.11		.11
N	627		743

---

insignificant: opinions about these two initiatives were not related to the probability of a crossover vote, net of the other variables in the model. The rest of the variables in the model are also insignificant at the .05 level, though the coefficients for Protestantism and Catholicism are in the expected direction.

Table 5.4 also presents the change in the predicted probability of crossing over associated with a shift from the minimum to the maximum value of the statistically significant variables. The results for Republican crossover voting demonstrate the power of partisanship and ideology. With all other variables held at their means, the probability that a strong liberal Republican crossed over was .27 greater than that of a strong conservative Republican. A comparable probability obtains for strength of partisanship. Income also had a strong effect: the highest income group is predicted to have a probability of crossing over .18 lower than that of the poorest group. Gender has a weaker effect.^[21]

Comparable models of Democratic crossover for Governor, presented in column two of table 5.4, produce generally similar results. While liberalism has the hypothesized effect—the probability of crossing over declines with increasing liberalism—the strength of partisanship variable is insignificant. However, the Prop 226 vote is statistically significant. Support for Prop 226, which is in essence an anti-union vote, is associated with a higher probability of crossover voting among Democrats.

That gender is not significant in this or any model of Democratic crossover demonstrates that the "pull" effect of gender—its ability to keep Democrats within the party—is not as strong as its "push" effect—its ability to drive Republicans toward Democratic candidates. In other words, Democratic men are less likely to defect than Republican women are. Of the remaining demographic variables in this model, income, education, and union membership have significant effects. As income increases, the propensity to cross over among Democrats increases. Education has the opposite effect. As Democratic respondents become more educated, ceteris paribus, their propensity to cross over in the gubernatorial primary decreases. Curiously, the effect of union membership is in the opposite direction than expected: being a union member (or having one in the family) was associated with a greater likelihood to cross over and vote for Lungren. However, the associated change in predicted probability shows that this unanticipated effect was relatively weak.^[22]

Multivariate Results: The U.S. Senate Race

Table 5.5 presents two comparable models of crossover voting for Senator. By and large the results for Republican crossover generally conform to those for the gubernatorial election, despite the differences in the size of

---

TABLE 5.5 Logit Models of Crossover Voting for Senator

	Republicans		Democrats
	Coefficient	Change in Probability	Coefficient	Change in Probability
SOURCE: Field Institute, Field Polls (San Francisco: The Field Institute, 1998). NOTE: Table entries are logit coefficients, with standard errors in parentheses. The dependent variable is coded 1 for a crossover vote and 0 otherwise. The rate of Republican crossover was 13.7 percent, and the rate of Democratic crossover was 17.6 percent. Change in probability is the change in the probability of crossover voting associated with a shift from the minimum to the maximum value of the independent variable, holding all other variables at their mean values. *p < .05 **p < .01 ***p < .001
Liberalism	0.45***	0.34	−0.40***	−.30
	(0.10)		(0.08)
Strength of partisanship	−0.40*	−.06	−0.76***	−.16
	(0.18)		(0.14)
Prop 226 vote	−0.74**	−.06	0.40*	.04
	(0.28)		(0.21)
Prop 227 vote	0.13		0.31
	(0.33)		(0.23)
Gender	0.44*	.03	−0.33
	(0.26)		(0.21)
Protestant	−0.14		−0.35
	(0.30)		(0.25)
Catholic	0.08		−0.37
	(0.35)		(0.30)
Income	−0.07		−.04
	(0.11)		(0.09)
Age	−0.002		0.005
	(0.008)		(0.007)
Education	0.02		−0.16**	−.14
	(0.06)		(0.05)
Asian	0.55		—
	(0.52)		—
Black	—		−0.18
			(0.32)
Latino	—		0.07
			(0.32)
Union member	—		0.11
			(0.22)
Constant	−1.50		2.94***
	(0.98)		(0.77)
−2 x log-likelihood	414.2		621.0
Percentage correctly predicted	86.7		82.9
No crossover	99.4		98.0
Crossover	6.1		12.2
Pseudo-R 2	.13		.16
N	600		791

---

the Democratic field (there were three Democratic gubernatorial candidates but only one in the Senate race). As hypothesized, greater liberalism is associated with an increased probability of crossover; greater partisanship is associated with a declining probability of crossover. The effect of liberalism was quite strong: the associated change in predicted probability is .34. Voting intention on Prop 226 is also significant and in the expected direction: a pro–Prop 226 stance, meaning a vote against unions, was negatively associated with crossover voting. In general, Republicans are more likely to remain loyal when they are more conservative, more partisan, and antagonistic to unions, a traditional Democratic constituency. Interestingly, gender is again significant: even after controlling for "political" variables, women were more likely than men to cross over and vote for Boxer.^[23]

The results for Democratic crossover voting for Senator, presented in column two of table 5.5, again confirm our theoretical expectations. Liberalism and partisanship have significant negative effects on the probability of a crossover vote. The coefficient for the vote on Prop 226 is again significant and in the hypothesized direction: a vote against unions was associated with a greater likelihood of a Democratic crossover vote to Fong or Issa. The coefficient for education, as in the analysis of Democratic crossover for Governor, is significant and negative. Among Democrats, a higher level of formal education was associated with a declining propensity to cross over and vote for a Republican candidate for Senator. The changes in predicted probability again underscore the power of liberalism and partisanship as predictors of defection.

Crossover voting thus derives from relatively weak party ties, sympathy with the ideological orientation of the other party, and membership in demographic groups generally linked to support for the opposition. This pattern of associations suggests that most crossover voting is sincere voting or hedging. If raiding were prevalent, crossover voting should be concentrated among strong partisans and others with an interest in sabotaging the other party's nomination process. Instead, just the opposite holds true.

Crossing Over Consistently? Synthesizing Electoral and Individual Motivations

Thus far, two sorts of motivations for crossover voting have emerged. First, the aggregate level of crossover voting tends to increase when voters confront an uncompetitive race in their own party but a competitive race in the other party. Thus, the percentage of Republicans crossing over into the Democratic gubernatorial primary was much higher than the percentage of Democrats crossing over to Lungren. Second, the multivariate analysis of individual behavior shows that the strength of partisanship and ideological self-identification consistently influence the decision to cross over. Taken together, these two sets of findings suggest that crossover voting

---

depends first on the choice structure. Weak partisanship or ideological distance from the party's core beliefs does not necessarily lead to crossover voting. Likewise, the mere presence of asymmetric competition stimulates crossover voting mainly among those voters who are already predisposed to defect from their party or, at the least, who find it wise to hedge their bets.

Further evidence for the confluence of electoral and individual attributes comes when we analyze the "consistency" of crossover voting—whether respondents crossed over in either the gubernatorial or senate race, or whether they crossed over in both races. We thus cease treating crossover voting in the gubernatorial race and crossover voting in the senatorial race as independent acts.

The incidence of consistent crossover voting—that is, in both the Governor's and U.S. Senate races—is quite small. For example, among Republicans who crossed over in the gubernatorial race, only 30 percent crossed over to vote for Boxer in the Senate race. Among Democrats who crossed over in the Senate race, only 18 percent crossed over to Lungren in the gubernatorial race.^[24] Altogether, only 4.3 percent of Republican identifiers and 1.7 percent of Democratic identifiers were consistent crossover voters, demonstrating the important role of factors such as the competitiveness of each party's contest. Confronted with a competitive Democratic gubernatorial primary, a sizable number of Republican voters crossed over to Davis, Checchi, or Harman; by contrast, confronted with an uncompetitive Democratic senatorial primary, very few of these gubernatorial crossover voters preferred Boxer as well.

However, a simple descriptive analysis of the demographic and political attributes of consistent crossover voters demonstrates something more: as one would expect, these voters are even further out of step with their party than are those who crossed over in just one of the races. Table 5.6 shows that consistent crossover voters in both parties tend to be weak partisans and ideological misfits when compared to voters who either did not cross over or who crossed over in only one race. For example, only 11.6 percent of Republicans who crossed over in both races were strong partisans, as compared to 26.7 percent of Republicans who crossed over in one race and 43.6 percent of those who did not cross over at all. Those who crossed over twice were also less conservative and less supportive of Prop 226. Similarly, the Democrats who crossed over in both races were weaker partisans, less liberal, and more supportive of Prop 226.

This suggests that individual attributes and electoral circumstances combine to encourage crossover voting. Within each party, there exists a subset of voters with a predisposition to defect. In a sense, the party identification of consistent crossover voters in particular was an error in judgment. The blanket primary allowed them to find their true homes. Thus, these voters

---

TABLE 5.6 Characteristics of Consistent and Inconsistent Crossover Voters

	Strong Partisan	Prop 226	Supported Conservative
SOURCE: Field Institute, Field Polls, February–May 1998 (San Francisco: The Field Institute, 1998).
Republican crossover
None	43.6%	64.2%	79.3%
One race	26.7	48.1	71.2
Both races	11.6	33.3	56.6
Democratic crossover
None	39.5	12.9	48.3
One race	24.6	20.2	58.3
Both races	20.6	26.5	77.4

were almost surely acting sincerely. However, defection is not a given; instead, it is inspired by the nature of the race itself. In particular, competitive races tend to attract crossover voters, especially when their own party's race is a foregone conclusion.

THE EFFECT OF THE PRIMARY VOTE ON THE GENERAL ELECTION VOTE: DO CROSSOVERS RETURN HOME?

The previous analysis clearly indicates a limited incidence of raiding in the 1998 California blanket primary. However, that finding does not speak to the relative incidence of sincere voting and hedging. One way of disentangling these motivations is to examine how crossover voters in the primary election behaved in the November general election. Sincere voting in the primary means that one chose one's favorite candidate, regardless of his or her party affiliation. Such voters should stick with this choice in the general election rather than return to their partisan home. By contrast, hedging is motivated by risk aversion. Hedgers vote for their second choice in the primary in order to reduce the chances of the worst-case scenario—a general election victory for their least-liked opposition candidate.

Accordingly, crossover voters who persisted in their partisan disloyalty, such as Republicans who voted for Davis in both the primary and general elections, are deemed sincere, while the primary crossover voters who returned to their own party in November may have been hedging in the primary. We say "may have" because the phenomenon of returning home could also indicate sincere voting. Suppose a Republican crossed over to vote for Checchi in the gubernatorial primary but then cast a vote for Lungren in the general election. The primary vote would be hedging, if Checchi

---

were his or her most favored Democratic candidate but Lungren the first preference overall, but it would be an act of sincere voting if this Republican actually preferred Checchi above everyone else, and Lungren was simply his or her second choice. With the present data, it is difficult to distinguish hedging from sincere voting among crossover voters who voted for Checchi or Harman rather than Davis and Issa rather than Fong. But we can provide a reasonable, if inexact, estimate of the relative prevalence of these two voting strategies.

The data in this analysis come from the last preelection Field poll in October 1998. This poll not only asked voters their intended vote in the general election, but also their vote in the primary, if they stated that they had in fact voted in the primary. Naturally, such a recall measure is problematic. People tend to overstate their participation and may not accurately remember their primary vote, which occurred five months before. In this sample, more voters reported that they voted for the candidate who won the primary election than reasonably could have done so—evidence of the familiar bandwagon effect. According to this poll, Davis won the "recall" primary with 76 percent of the vote, instead of the 51 percent he actually garnered. Similarly, Matt Fong's vote share increased to 74 percent of the Republican "recall" electorate, whereas he actually won 45 percent.^[25]

While both margins of victory were significantly inflated, there was a much more modest bias in respondents' recollections of crossover voting. Sixteen percent of the gubernatorial voters recalled crossing over against their party registration, as did 14 percent of senatorial voters.^[26] These rates are similar to those from both the pre-primary Field polls and the Los Angeles Times exit poll. For example, according to the exit poll, 17.5 percent of voters crossed over in the gubernatorial race and 15 percent crossed over in the Senate race. The recall questions can thus help identify the motivations behind crossover voting, despite the uncertainty about how the aggregate recall bias toward Davis and Fong affects the individual results.^[27]

If crossover voting was largely sincere, most Republican crossovers in the gubernatorial primary should have supported Davis in the general election, and Democratic crossovers in the Senate race should have supported Fong in the general election. Table 5.7 shows just that. The top panel shows that 63.3 percent of Republican crossovers stuck with Davis in November. The bottom panel presents a comparable result: 64.6 percent of Democratic crossovers stuck with Fong in November. This tendency indicates another similarity between the blanket primary and the general election. In a sense, the blanket primary is a preview, or even a first stage, of the general election campaign.^[28]

General election preferences among crossover voters depended somewhat on their primary vote choice. Table 5.7 shows that, while only 25.4 percent of Republicans who voted for Davis in the primary returned to

---

TABLE 5.7 General Election Preferences

SOURCE: Field Institute, Field Poll, October 1998 (San Francisco: The Field Institute, 1998). NOTE: Weighted N s appear in parentheses.
Among Republican Crossovers for Governor
	Percentage of Crossovers Who …
Primary Vote Choice	"Returned Home" in General Election	Also Crossed Over in General Election
Davis	25.4%	74.6%
	(44)	(129)
Checchi	66.7	33.3
	(40)	(20)
Harman	46.7	53.3
	(7)	(8)
TOTAL	36.7	63.3
	(91)	(157)
Among Democratic Crossovers for Senator
	Percentage of Crossovers Who …
Primary Vote Choice	"Returned Home" in General Election	Also Crossed Over in General Election
Fong	22.5%	77.5%
	(27)	(93)
Issa	70.5	29.5
	(31)	(13)
TOTAL	34.5	64.6
	58)	(106)

their party, 46.7 percent of Harman voters and 66.7 percent of Checchi voters did so. A similar pattern emerges in the Senate race: Democratic voters who supported Fong in the primary were much more likely to stick with him in the general (77.5 percent did so) than were Issa voters, most of whom (70.5 percent) returned to the Democratic Party and supported Boxer in the general. Whether this pattern of voting represents hedging or sincere voting cannot be stated with precision.

However, it is possible to generate some minimum and maximum estimates of sincere voting and hedging. In the Governor's race, the hedgers include, at the minimum, everyone who supported Davis in the primary but not in the general.^[29] Let us assume, for the moment, that Checchi and Harman voters who returned home were expressing their sincere preference

---

in both the primary and the general; that is, these Republicans preferred a (losing) Democratic candidate in the primary, but Lungren in the general election. If this assumption is true, these voters should not be counted as hedgers. Thus, hedging is limited to those voters who voted for Davis in the primary but Lungren in November (i.e., 17.7 percent [44 of 248] of the Republican crossover voters analyzed in table 5.6). The maximum proportion of sincere crossover voters is thus 82.3 percent. At the maximum, the hedgers include every Republican who returned home in the general election—36.7 percent of the Republican crossover voters (91 of 248).^[30] This leaves a minimum of 63.3 percent sincere crossover voters. In the Senate race, similar computations among Democratic crossover voters generate nearly identical results: the minimum rate of hedging is 16.5 percent, and the maximum is 35.4 percent.

Thus, it appears that crossover voting in both of these races was largely sincere. First, a variety of evidence suggests that raiding was scarce. Crossover voters tended to support the winning candidate in each party's primary, and their preferences followed the overall trend as the primary campaign wore on. Furthermore, multivariate analyses show that crossover voting is most prevalent among voters with a predisposition to support the other party, not among the strong partisans and ideologues who would most likely perpetrate electoral mischief.

Second, the relationship between the primary and general election vote demonstrates that, at the most, hedging occurred among just over a third of crossover voters, meaning the majority of these voters were sincere. This estimate must be treated with some caution, given the vagaries of how respondents remembered their primary vote choices. Nevertheless, it seems reasonable to conclude that California's electoral reform did not radically affect voting behavior. It merely allowed voters to express their true preferences earlier in the campaign.

CONCLUSION

A broad array of evidence indicates that crossover voting in the June 1998 blanket primary was largely sincere. Party mischief was minimal; moreover, crossover voting did not affect the outcome of the gubernatorial and U.S. Senate races. Therefore, this election was no apocalypse for the parties. Neither was it an Eden for the blanket primary's proponents: 1998 did not produce moderate, centrist candidates in these races. The blanket primary's most important impact may be how it prefigures the general election. Those involved in the 1998 campaign expressed a similar sentiment. Davis's campaign manager, Garry South, has said, "For all practical purposes, the general was over on primary night, barring some cataclysmic

---

event or huge mistake on our part" (Lubenow 1999, 147). Doing well overall, not just among the registered voters of one's own party, thus can be critical. If this is so, an additional set of research questions emerges. How do the results of the primary election affect voters' perceptions of the viability of the candidates who emerge? Moreover, at an institutional level, how does the need to fuse partisan and more general appeals even before the primary shape the nature of campaigns and campaigning?

APPENDIX: VARIABLE CODING

Dependent Variable
Crossover Voting	Coded 1 if a respondent intends to vote for (or stated that they voted for) a candidate of the opposing major party, as defined by the respondent's party identification (or registration), and 0 otherwise.
Independent Variables
Strength of Partisanship	Coded 1 (independent) to 4 (strong partisan). This was constructed by "folding" the standard, seven-point party identification scale.
Liberalism	Coded 1 (strong conservative) to 7 (strong liberal).
Prop 226 Vote	Coded 1 (yes) and 0 (no). A yes vote indicates, support for Prop 226, which mandated that unions get the permission of members before spending their dues on political purposes, such as advertisements during campaigns.
Prop 227 Vote	Coded 1 (yes) and 0 (no). A yes vote indicates support for Prop 227, which was designed to restructure, and largely do away with, bilingual education in California public schools.
Gender	Coded 1 (female) and 0 (male).
Protestant, Catholic, Latino, Black, Asian	Coded 1 if respondent identifies as a member of religious/ethnic group, and 0 otherwise.
Age	Coded 1 (18–24), 2 (25–29), 3 (30–39), 4 (40–49), 5 (50–59), 6 (60 and older).
Education	Coded 1 (eighth grade or less), 2 (some high school), 3 (high school graduate), 4 (trade/ vocational school), 5 (1–2 years of college), 6 (3–4 years of college, no degree), 7 (college graduate), 8 (5–6 years of college), 9 (master's degree), 10 (graduate work past master's).
― 102 ―
Income	Coded 1 (under $20,000), 2 ($20–40,000), 3 ($40–60,000), 4 ($60–80,000), and 5 (more than $80,000).
Union Member	Coded 1 if respondent or a member of respondent's family is a union member, and coded 0 otherwise.

NOTES

We thank Liz Gerber and Bruce Cain for their helpful comments. Mark DiCamillo of the Field Institute and Susan Pinkus of the Los Angeles Times graciously made available the data analyzed here.

---

REFERENCES

Abramowitz, Alan; John McGlennon; and Ronald Rapoport. 1981. “A Note on Strategic Voting in a Primary Election.” Journal of Politics43: 899–904.

Adamany, David. 1976. “Crossover Voting and the Democratic Party's Reform Rules.” American Political Science Review70: 36–41.

Cain, Bruce E. 1997. “Report on Blanket and Open Primaries.” Expert testimony in California Democratic Party et al. v. Jones 984 F. Supp. 1288 (1997).

Cohen, Jonathan; Thad Kousser; and John Sides. 1999. “Sincere Voting, Hedging, and Raiding: Testing a Formal Model of Crossover Voting in Blanket Primaries.” Paper presented at the 1999 Meeting of the American Political Science Association, Atlanta, GA.

DiCamillo, Mark. 1998. “Californians Vote in the State's First Blanket Primary.” Public Affairs Report39, no. 5: 1, 4–5.

Hedlund, Ronald, and Meredith Watts. 1986. “The Wisconsin Open Primary, 1968 to 1984.” American Political Quarterly14: 55–73.

Hedlund, Ronald; Meredith Watts; and David Hedge. 1982. “Voting in an Open Primary.” American Political Quarterly10: 197–218.

Keith, Bruce E., et al. 1992. The Myth of the Independent Voter. Berkeley: University of California Press.

Lubenow, Gerald C., ed. 1999. California Votes—The 1998 Governor's Race: An Inside Look at the Candidates and Their Campaigns by the People Who Managed Them.Berkeley: Institute of Governmental Studies Press.

Miller, Warren E., and J. Merrill Shanks. 1996. The New American Voter. Cambridge, MA: Harvard University Press.

Ranney, Austin, and Leon Epstein. 1966. “The Two Electorates: Voters and Non-Voters in a Wisconsin Primary.” Journal of Politics66: 598–616.

Robeck, Bruce. 1984. “The Representativeness of Congressional Primary Voters.” Congress and the Presidency11: 59–67.

---

Southwell, Patricia. 1991. “Open versus Closed Primaries: The Effect on Strategic Voting and Candidate Fortunes.” Social Science Quarterly72: 789–96.

Wekkin, Gary. 1988. “The Conceptualization and Measurement of Crossover Voting.” Western Political Quarterly41: 105–14.

Wekkin, Gary. 1991. “Why Crossover Voters Are Not ‘Mischievous Voters': The Segmented Partisanship Hypothesis.” American Politics Quarterly19: 229–47.