10. Strategic Voting and Candidate Policy Positions
Elisabeth R. Gerber
One of the most debated questions about the blanket primary is whether it advantages candidates with ideologically moderate or ideologically extreme policy positions.[1] This question directly addresses the issue of political representation. If moderate candidates benefit from the blanket primary, in the sense of having a higher probability of being nominated and elected, then we expect the interests of citizens sharing their centrist or moderate views to be better represented in the policy process. If ideologically extreme candidates benefit from the blanket primary, we expect citizens with extreme views to benefit instead.
Two views regarding the effects of the blanket primary on candidate policy positions prevail. One view, which is central to the arguments made by proponents of the blanket primary, is that the blanket primary advantages moderate candidates (see California Secretary of State 1996). By opening participation in the primary to voters outside a candidate's party, the blanket primary system forces candidates to compete for the ideological center of the electorate. Candidates with more moderate positions, or who can move to the center, receive greater support and prevail. A second view, often advanced or implied by opponents of the blanket primary, is that it advantages extremist candidates (California Secretary of State 1996). By allowing participation by members of the other party, the blanket primary invites "raiding" or "sabotage" by ideologically motivated partisans. These voters nominate weak candidates in the other party and their own ideologically polar candidate prevails.
In this chapter, I assess these two alternative views of primary election dynamics in light of the experience of the June 1998 California primary. First, I sketch out the logic of each view. Second, I review existing evidence about the relationship between primary election systems, candidate strategies,
TWO VIEWS OF PRIMARY ELECTION DYNAMICS
The two views of primary election dynamics agree on the basic structure of primary elections. Both sides see primary elections as the first phase in a multistage election process. Primary election laws define the rules of participation; that is, they define which voters are permitted to participate in candidate nominations. Both sides agree that under closed primaries, participation in the primary is limited to registered members of that party. Under blanket primaries, participation is extended to all registered voters, including partisans, independents, and members of other parties.[2]
Where the two views differ is on their interpretation of the consequences of extending participation to all voters. Specifically, they differ in their expectations of how voters will behave in blanket primaries and how candidates will respond. As discussed in the introduction to this volume and the several chapters on voter behavior, voters in blanket primaries can engage in several forms of crossover voting—sincere crossover, hedging, and raiding. All three forms of crossover voting are difficult or impossible under closed primaries, since voters would need to anticipate their desire to cross over long before the election and re-register under a new party label. In some closed primary states, such as New York, voters must re-register a full year before the election. In California prior to 1998, the registration period was twenty-nine days—shorter but still potentially prohibitive for most voters (Bott 1991). Low rates of re-registration in closed primary states suggest that crossover voting in these systems rarely occurs. By contrast, voters in open primary systems (both traditional open and blanket) can decide to cross over at the polls, requiring no precommitment to one or the other party. Voters in blanket primaries are even less constrained by party and can cross over race by race.
The first view of primary election dynamics argues that of the three types of crossover behavior, sincere crossover and hedging will prevail. This is not to say that voters will never engage in raiding; rather, the argument is that most voters are not sufficiently informed to engage in such sophisticated behavior. Raiding requires voters to calculate (1) which candidate would pose the weakest challenge to their favored candidate; (2) whether that candidate can win the primary; and (3) whether that candidate will lose the general. Further, even if they are sufficiently informed to figure out how to raid the other party's primary, voters are likely to be risk-averse and may not vote for a candidate they dislike, even if it means increasing
From the candidates' perspective, extending participation in the primary to centrist independent and moderate voters creates advantages for centrist and moderate candidates. Extending participation means shifting the primary electorate median voter toward the ideological center. In other words, the blanket primary median is closer to the ideological center than the closed primary median. Hence, moderate candidates with policy positions close to the median will prevail, resulting in the nomination of more conservative Democrats and more liberal Republicans compared to their closed primary counterparts.[3] I refer to this prediction as the "moderation hypothesis."
The second view of primary election dynamics argues that of the three types of crossover behavior, the effects of raiding will prevail. This view recognizes the possibility that some voters, especially independents and moderates, may engage in sincere crossover voting and hedging, drawing candidates to the ideological center. However, they also argue that voters will sometimes face strong incentives to engage in raiding. The most favorable situation for raiding is when one party's primary is uncontested. Voters from that party waste their vote by casting it for their party's presumptive nominee. The question then becomes which candidate they should vote for from the other party. Voters ask if they are better off voting for the more preferred candidate in the other party (i.e., hedging), leading their top two candidates from the two parties to face off in the general election, or voting for the less preferred candidate in the other party (i.e., raiding), leading their most preferred candidate to face his or her weakest opponent. In the end, voters' choices will depend on their preferences relative to the candidates, their attitudes toward risk, and the several candidates' electoral prospects. As discussed above, critics of this view argue that voters are simply not well enough informed to behave in such a sophisticated manner. Defenders counter by pointing out the potential for candidates and campaign organizations to help voters coordinate their behavior.
From the candidates' perspective, the possibility of strategic crossover greatly complicates matters. On the one hand, extending participation to independents and members of other parties means shifting the primary electorate median voter toward the ideological center, as noted above. On the other hand, if some voters engage in raiding, it is no longer clear whose preferences become decisive. Hence, the question candidates must ask is: What types of voters are likely to cross over? To the extent that ideologically extreme voters cross over strategically, benefiting extremist candidates, this view results in what I refer to as the "polarization hypothesis."
To illustrate the logic of voter choice in blanket primaries, consider a
Figure 10.1. Example of crossover voting.
In figure 10.1, since the D party primary is uncontested, voters whose ideal points are closest to d1 need not "waste" their votes in the uncontested D party primary, as they would under closed primaries. Rather, they can vote for one of the R party candidates. The question is, which one? Voters know that if r1 wins the primary, d1 and r1 will meet in the general. Since d1 and r1 are equidistant from the general election median (at .5), r1 will win the general election with prob=.50. If r2 wins the primary, r2 will beat d1 with certainty. Therefore, all voters who prefer a lottery between d1 and r1 to a win by r2 will vote for r1. This will be the case for voters with ideal points from 0.00 up to .375. These voters have ideal points that are closer to d1 than to r2 and so prefer a primary outcome that gives d1 some chance of victory in the general election. Voters to the right of .375 receive higher utility from r2 than from the lottery, and so they vote for r2. Since in this example the number of r2 voters is greater than the number of r1 voters, r2 wins the primary and then the general.
This example is constructed to illustrate all three types of crossover voting. Voters with ideal points between .4375 and .5 engage in sincere crossover, voting for their most preferred candidate, r2. Voters with ideal points between .375 and .4375 engage in hedging—they most prefer d1, but prefer
Note that a number of the example's assumptions, such as the frontrunner candidates' positions being equidistant from the general election median voter's ideal point, are critical to generate the precise predictions about r2 winning the primary and the general. If we relax some of these highly restrictive assumptions—for example, if we allow candidates to take more extreme or more moderate positions, or for more than one challenger to enter, or for the number of D and R voters to be different, or for voters to be risk-averse, or for turnout to be probabilistic or uncertain—then optimal voter strategies and candidate responses become much more difficult to determine.[6] The point of the example is not to make specific predictions about winning candidate positions, but rather to illustrate how all of these variables interact.
EVIDENCE OF CROSSOVER VOTING, CANDIDATE ENTRY, AND POLICY POSITIONS
Several recent studies have analyzed the effects of primary election laws on candidate strategies. Gerber and Morton 1998 compare the ADA scores of U.S. Representatives elected under different primary systems.[7] They show that winning House candidates elected from open, blanket, and nonpartisan primaries are more moderate (that is, Democrats are more conservative and Republicans are more liberal) than otherwise similar candidates elected from closed primaries. They also find that the greatest share of moderates are not elected from the most open systems, but rather from semiclosed systems. They conjecture that in semiclosed systems, where new voters and independents can participate, but not members of other parties, incentives for sincere crossover and hedging clearly dominate. In other words, Gerber and Morton's evidence suggests that while sincere crossover, hedging, and raiding all are important in blanket primaries, the moderating effects of sincere crossover and hedging prevail. In a related study, Grofman and Brunell (2001) show a higher proportion of mixed-party U.S. Senate delegations in open primary states, suggesting that voters are trying to moderate the net ideology of the delegation by electing Senators of different parties.
Together, these studies provide evidence that is consistent with the moderation hypothesis and contrary to the polarization hypothesis. They imply that, on average, the moderating effects of sincere crossover voting and hedging dominate the polarizing effects of raiding. As a result, moderate candidates tend to prevail in open primaries. These results are consistent with the findings in the five chapters in this volume on voter behavior
There is, however, some compelling evidence to the contrary. King (1998) shows that party organizations are less cohesive and candidate positions show greater variance in open primary states. He argues that under open primaries, candidates have an incentive to "specialize," focusing on one or a few issues and appealing to ideological subgroups in the electorate. In other words, he finds that candidates may be more moderate or more extreme under open primaries, depending on the issues they promote. Further empirical research is required to assess the generality of this claim.
CANDIDATE MODER ATION IN THE CALIFORNIA BLANKET PRIMARY
I now consider the candidate moderation and polarization hypotheses in light of evidence drawn from the 1998 California primary and general election. Did primary and general election winners hold moderate or extreme policy positions relative to those of their challengers? Did they hold more moderate or more extreme policy positions than their counterparts previously elected under closed primaries? Did the competitive circumstances of the race matter? That is, were more moderate candidates nominated and elected in contested races? Uncontested races? Open seats? Was the moderation effect more important in some races than in others? Was the polarization effect more important in some races?
Governor's Race
The Governor's race represented a prime opportunity for crossover voting. Incumbent Governor Pete Wilson was prohibited from running due to term limits. In the Republican primary, former State Attorney General Dan Lungren ran unopposed,[8] while in the Democratic primary, three candidates—former Lieutenant Governor Gray Davis, former U.S. Representative Jane Harman, and businessman Al Checchi—competed for the nomination.[9] With the Republican nomination virtually unchallenged, Republican voters were free to cross over into the highly competitive and uncertain Democratic primary.
When they crossed party lines, however, Republican voters faced a field of three ideologically similar, centrist candidates. Harman's record in the U.S. Congress distinguished her as one of the most pro-business, fiscally conservative Democrats in the House.[10] Checchi's lack of political experience made his ideological position more difficult to assess, but even after
Nor did crossover voting into the contested Democratic primary appear to result in the nomination of a weak opponent to Lungren. Davis raised nearly $26 million in campaign contributions (California Secretary of State 1998a) and ultimately beat Lungren by nearly 20 percentage points in November (California Secretary of State 1998b). If raiders targeted Davis as the weakest Democratic candidate, it looks like they seriously miscalculated.
This is not to say that the blanket primary had no effect on campaign dynamics or the relative strength of the several candidates. Some of their centrist positioning may, in fact, have been the result of electoral pressures from the center felt during the blanket primary. Further, the fact that three moderates entered the Democratic primary and emphasized their centrist positions may reflect their beliefs that centrist candidates had a greater chance of winning the blanket primary. In fact, Checchi's campaign manager admitted that they saw the blanket primary as a unique opportunity for a candidate without traditional party ties to appeal to independents and moderate Republican crossovers (Institute of Governmental Studies 1999). However, it is clear that the blanket primary was not, at least in this case, decisive. Other factors, such as Checchi's lack of political experience, Harman's inability to mobilize financial support, and Davis's more convincing appeal as an "incumbent-like" challenger seemed to dominate any moderating effect of the blanket primary (see chapter 4 by Baldassare and chapter 7 by Salvanto and Wattenberg in this volume on the importance of incumbency in the 1998 elections).
U.S. Senate Race
The U.S. Senate race presented similarly favorable circumstances for crossover voting. In the Democratic primary, incumbent Barbara Boxer ran virtually unopposed, while former State Treasurer Matt Fong and businessman Darrell Issa competed for the Republican nomination.[11] So, in the same way that Republicans were able to cross into the competitive Democratic primary for governor, Democrats could employ their votes in a race where they might actually count—the Republican primary—while their incumbent Senator was assured of the Democratic nomination.
As in the Governor's race, however, it was less than clear which Republican Senate candidate was the more moderate and hence the more attractive to sincere crossover voters and hedgers. Issa, like Al Checchi in the gubernatorial primary, lacked political experience and spent much of the early part of the campaign staking out his ideological turf. Toward late spring, an image began to emerge of a candidate who was quite conservative on social issues and liberal economically. Fong, by contrast, tried to portray himself as more liberal on social issues and conservative on economic issues, although his endorsement by conservative Christian organizations undermined this image to some extent. Hence, it is safe to say that while the Republican primary candidates took different positions on the issues, neither was obviously more attractive to centrist independents and crossover voters.
Where the candidates did clearly differ is on their political records and ties to the political parties. Issa, like Checchi, had no prior political experience. In previous elections, when voters were in the mood to take out their frustrations about a sagging economy on incumbent politicians, this outsider status might have worked to Issa's advantage. In 1998, however, voters were in quite a different mood: incumbents and experienced politicians won up and down the ballot in both the primary and the general (see Baldassare, chapter 4; Salvanto and Wattenberg, chapter 7). To make matters worse, Issa's lack of ideological discipline and inconsistency of policy positions, no doubt a result of his political inexperience, were interpreted by the media as the mark of an erratic and unpredictable candidate. Thus, the lack of partisan ties and ideological rigidity—which should have helped Issa attract support from sincere crossovers and hedgers—failed to carry the day in 1998.
Where Issa lacked political experience, Fong was viewed as a traditional, mainstream party candidate. He was a long-term member of the Republican party. His mother was the very popular (Democratic) Secretary of State for many years. Fong served as a member of the state Board of Equalization and then as State Treasurer. Over many years of public service, Fong established himself as an expert on taxes and fiscal policy, and promised to make reform of the tax code one of his top legislative priorities as Senator (California Secretary of State 1998d). Hence, while it is impossible to know how voters weighed the relative importance of policy positions and political experience, it is clear that Issa was not able to capitalize on his independent status as a party outsider to defeat Fong in the primary.
To the extent that Fong benefited from having greater political experience than his primary competition, and perhaps from being perceived as more moderate ideologically, he was no match for Boxer in the general election. Exit polls show that most of the Democrats that crossed over for Fong in the primary returned to their party in the general and supported
U.S. House Races
It is more difficult to generalize about the effects of the blanket primary on candidate positions in legislative races because (1) electoral circumstances vary greatly from race to race, and (2) much less is known about most candidates, especially those who lose in the primary. Furthermore, since House incumbents are reelected at an extremely high rate, there is very little competition in most House general election races and hence little opportunity for the primary to make much difference in terms of electoral outcomes.[13] Nevertheless, there are at least three types of races in which we might expect crossover voting to affect winning congressional candidates' policy positions. The first type is the open-seat race. In open seats, there are no incumbents, and so challengers have a real chance of winning. However, open seats often generate competition in both parties, and so voters may be reluctant to cross over if their own party's nomination is contested. The second type of race in which the blanket primary may matter is the contested-challenger primary. In such races, only one party is seriously contested, leaving members of the incumbent's party free to cross over into the challenger's primary. However, since challengers rarely beat incumbents in the general election, any effects on candidate positions in the primary are unlikely to translate into election outcomes in the general. The third type of race in which the blanket primary may matter is when a vulnerable incumbent faces a challenge from within his or her own party. These challenges are rare. When they do occur, they may signal to potential crossover voters that the seat is up for grabs and that their votes could matter.
Table 10.1 reports the primary election candidates, primary nominees (single asterisk), and general election winners (double asterisk) for each open seat, contested-challenger, and contested-incumbent race in 1998. I describe the results of these races in detail below.
Open Seats. There were four open House seats in 1998 due to incumbent retirements: Congressional District (CD) 1, CD 3, CD 34, and CD 36. In addition, three seats were filled in special elections in the spring of 1998 due to incumbent retirements or deaths, and although their occupants were listed on the ballot as incumbents, they were rematched with their primary competitors again in June (CD 9, CD 22, and CD 44). I refer to these races as "quasi-open" seats. I discuss these races in this section because
| District | Democratic Candidates | Ideology | Republican Candidates | Ideology |
|---|---|---|---|---|
| C=conservative, I=incumbent, L=liberal, M=moderate. | ||||
| Nominees. | ||||
| Winners. | ||||
| Open and Quasi-Open Seats | ||||
| CD 1 | Mike Thompson[**] | M | Mark Luce[*] | — |
| Jim Hennefer | — | |||
| CD 3 | Sandy Dunn[*] | M | Barbara Alby | C |
| Bob Dean Kent | — | Doug Ose[**] | M | |
| CD 34 | Grace Napolitano[**] | L | Ed Perez[*] | — |
| James Casso | L | |||
| CD 36 | Janice Hahn[*] | — | Steve Kuykendall[**] | M |
| Rudy Svorinich | — | |||
| Susan Brooks | — | |||
| Robert Pegram | C | |||
| CD 9 | Barbara Lee[**] | L | ||
| CD 22 | Lois Capps[**] | M | Tom Bordonaro[*] | C |
| James Harrison | M | |||
| CD 44 | Ralph Waite[*] | — | Mary Bono[**] | C |
| CD 46 | Loretta Sanchez[**] | M | Robert Dornan[*] | C |
| Lisa Hughes | — | |||
| James Gray | M | |||
| Contested-Challenger Primaries | ||||
| CD 10 | Ellen Tauscher (I)[**] | M | Charles Ball[*] | M |
| Donald Amador | C | |||
| Gordon Blake | C | |||
| CD 15 | Dick Lane[*] | M | Tom Campbell (I)[**] | M |
| Connor Vlakancie | — | |||
| CD 17 | Sam Farr (I)[**] | M | Bill McCampbell[*] | M |
| Mark Cares | — | |||
| CD 20 | Cal Dooley (I)[**] | M | Devin Nunes | M |
| Cliff Unruh[*] | C | |||
| CD 24 | Brad Sherman (I)[**] | M | Randy Hoffman[*] | M |
| Joe Gelman | C | |||
| William Westmiller | C | |||
| Contested-Incumbent Primaries | ||||
| CD 41 | Eileen Ansari[*] | — | Jay Kim (I) | C |
| Gary Miller[**] | C | |||
| CD 43 | Mike Rayburn[*] | — | Ken Calvert (I)[**] | C |
| Joe Khoury | C | |||
| CD 45 | Patricia Neal[*] | L | Dana Rohrbacker (I)[**] | C |
| Charmayne Bohman | M | |||
As mentioned above, we expect little crossover voting in open seats when both parties compete for the nomination, since members of both parties will be concerned, first and foremost, with their own party's nomination. However, in 1998, only one of the eight open or quasi-open seats was contested by both parties, so crossover voting is a real possibility.
Of these eight open or quasi-open seats, moderate candidates were elected in five races.[14] In CD 1, moderate former State Senator Mike Thompson beat Jim Hennefer in the primary and Republican Mark Luce in the general. In CD 3, both parties nominated moderates, and Republican Doug Ose beat Democrat Sandy Dunn in the general. In CD 36, moderate Steve Kuykendall beat a field of three other Republicans for the nomination and defeated Democrat Janice Hahn in the general. In the Twenty-Second District, Democrat Lois Capps faced no primary opposition and defeated conservative Tom Bordonaro in the general. And moderate Democrat Loretta Sanchez beat conservative Republican Bob Dornan in CD 46. Extremist candidates were elected in three districts. In CD 34, liberal Democrats Grace Flores Napolitano and James Casso met in the primary, with Napolitano advancing to the general and easily defeating Republican Ed Perez. In CD 9 and CD 34, liberal Democrat Barbara Lee and conservative Republican Mary Bono, respectively, faced the same opponents they had defeated in special elections just months before and won easily. Hence, these open-seat races provide some preliminary support for the moderation hypothesis.
A closer look at the open-seat races, however, reveals that circumstances were ripe for crossover voting—sincere, hedging, or raiding—in only five races (CD 1, CD 22, CD 34, CD 36, and CD 46). In these five races, only one party held a contested primary. All five produced moderate general election winners, but only two produced moderate winners in the contested primary (Democrat Mike Thompson in CD 1 and Republican Steve Kuykendall in CD 36). The nomination and election of these moderates suggests that in these races there may have been substantial sincere crossover and hedging. Note, however, that other factors besides the blanket primary, including superior political experience, campaign resources, and key endorsements, also played important roles in all of these races.
Contested-Challenger Races. House incumbents are rarely challenged in the primary (in 1998, three House incumbents from California faced serious primary challenges from their own party—see below). The most common scenario is a contested primary to nominate the challenger. In these races, the outcome of the incumbent's primary is virtually certain and members of the incumbent's party are free to participate in the contested challenger's primary. We therefore expect any effects of crossover voting to manifest predominantly in the contested-challenger primaries. Of course, the outcome of these contested-challenger primaries rarely affects general election outcomes, since congressional incumbents are reelected at extremely high rates. Nevertheless, other less direct effects of crossover voting into the challenger's primary may still be felt in the general election, particularly if such crossover voting results in a challenger who can force the incumbent to address new issues or take new positions on old issues.
In the 1998 California primary, fourteen congressional races witnessed some competition in the challenger's party's primary. In three of these races, there was also significant intraparty competition in the incumbent's own primary. However, since challengers are virtually assured of losing in the general, these contested-challenger races rarely attract strong candidates with well-developed policy positions. Moreover, even if they have welldeveloped positions, challengers can rarely mobilize sufficient resources to run professional campaigns that effectively disseminate information about the candidate's positions to voters. This means that during the election, potential crossover voters have little information on which to base their voting decisions. After the election, analysts have little information with which to classify candidates' positions. Recognizing these information problems, I therefore focus on the contested-challenger primaries with "serious" candidates. I define these as races in which at least one candidate received at least 20 percent of the total primary vote. There were five such races in 1998.
Moderates were nominated in four of the five contested-challenger primaries with serious candidates. In two races, there was one candidate who was clearly moderate and two who were clearly conservative, and the moderate was nominated in both races. In the Republican primary in CD 10, relatively moderate Charles Ball defeated conservative primary challengers Donald Amador and Gordon Blake; in the Republican primary in CD 24, moderate Republican Randy Hoffman defeated conservatives Joe Gelman and William Westmiller. In the other two races, none of the challenger primary candidates were clearly liberal or conservative, and moderate candidates Dick Lane (CD 15 Democratic primary) and Bill McCampbell (CD 17 Republican primary) were nominated. Thus, in all four of these races, conditions were ripe for crossover voting, and moderate candidates were
Contested-Incumbent Races. Congressional incumbents rarely face challenges from within their own party. When they do, the challenge usually results from extraordinary circumstances such as the incumbent's involvement in a scandal. The three intraparty incumbent challenges in 1998 (CD 41, CD 43, and CD 45) are cases in point. In all three races, the conservative Republican incumbent was involved in a major personal or political scandal. In CD 41, incumbent Jay Kim was defeated by a conservative opponent Gary Miller. In CD 43, incumbent Ken Calvert prevailed over a conservative opponent, Joe Khoury. In CD 45, incumbent Dana Rohrbacker prevailed over his moderate opponent Charmayne Bohman. Due to the extraordinary circumstances in each of these races, however, I hesitate to generalize from these outcomes about the effects of the blanket primary.
State Assembly Races
The State Assembly races provide even more opportunities to analyze the effects of the blanket primary on candidate positions. Due to term limits, there were many open seats and contested races for State Assembly in 1998. This should translate into ample opportunities for studying all types of crossover voting.
Due to the lack of reliable information on candidate positions in some low-profile races, I again restrict my analysis to the contested Assembly primaries with "serious" candidates who received at least 20 percent of the total vote cast for all primary candidates. I was able to obtain information about the winning candidate in thirty-four of thirty-seven such races. For each race, I obtained information on the winning candidate's ideology (conservative, moderate, liberal); party (Democratic or Republican); type of race (open seat, challenger's party, incumbent's party); and district ideology (safe D, strong D, split, strong R, safe R).[15] For sixteen races, I was also able to identify the losing candidate's ideology.[16] I report these raw data for 1998 in table 10.2.
One advantage of analyzing the Assembly primaries is that there were a sufficient number of contested races to conduct statistical analyses to test the moderation and polarization hypotheses directly. To test whether the blanket primary resulted in more moderate or more extreme candidates, I need to compare the results in these races to the results in races with a
| District | Party | Race Type | Winner's Ideology | Second-Place Candidate's Ideology | Third-Place Candidate's Ideology | District Ideology |
|---|---|---|---|---|---|---|
| 2 | Rep. | Open | M | C | Safe R | |
| 3 | Rep. | Open | C | C | Safe R | |
| 4 | Rep. | Inc.'s party | C | Safe R | ||
| 5 | Rep. | Open | M | C | Safe R | |
| 7 | Dem. | Open | L | Strong D | ||
| 9 | Dem. | Open | L | L | Strong R | |
| 10 | Rep. | Open | M | C | Strong R | |
| 15 | Dem. | Chal.'s party | Strong R | |||
| 20 | Dem. | Open | M | Strong D | ||
| 20 | Rep. | Open | M | Strong D | ||
| 23 | Dem. | Inc.'s party | L | Safe D | ||
| 24 | Rep. | Inc.'s party | M | C | C | Strong R |
| 29 | Rep. | Open | C | M | Safe R | |
| 30 | Dem. | Chal.'s party | L | Split | ||
| 33 | Rep. | Open | M | Safe R | ||
| 35 | Rep. | Open | M | Split | ||
| 35 | Rep. | Open | M | Split | ||
| 37 | Rep. | Open | C | M | Safe R | |
| 47 | Dem. | Open | M | Safe D | ||
| 49 | Dem. | Open | M | L | L | Safe D |
| 50 | Dem. | Open | L | L | Safe D | |
| 53 | Dem. | Open | Split | |||
| 54 | Rep. | Open | M | Split | ||
| 54 | Dem. | Open | M | Split | ||
| 55 | Dem. | Inc.'s party | M | Safe D | ||
| 57 | Dem. | Inc.'s party | L | Safe D | ||
| 60 | Rep. | Open | M | Split | ||
| 61 | Rep. | Open | C | Split | ||
| 62 | Dem. | Open | Strong D | |||
| 65 | Rep. | Inc.'s party | C | Safe R | ||
| 67 | Rep. | Inc.'s party | C | M | Strong R | |
| 68 | Rep. | Open | C | M | Safe R | |
| 73 | Rep. | Open | C | C | Safe R | |
| 75 | Rep. | Open | M | C | Safe R | |
| 76 | Rep. | Chal.'s party | C | M | Strong D | |
| 78 | Rep. | Chal.'s party | M | Split |
Table 10.4 compares the percentage of contested Assembly primaries won by moderate candidates in 1996 (under the closed primary) and in 1998 (under the blanket primary). The first row shows that moderates won half of all contested primaries in 1998, compared to 37 percent in 1996. A one-tailed difference of proportions test produces a p-value of .12, indicating that we can reject the null hypothesis that the proportions of moderate winners in the two elections are the same (or that the 1996 proportion is higher) with 88 percent confidence. A two-tailed difference of proportions test produces a p-value of .24. While these tests fail to achieve the standard 95 percent level of confidence, due largely to the small number of races in both election years (forty-one in 1996, thirty-four in 1998), they do indicate some support for the moderation hypothesis.
To test the robustness of the moderation effect, subsequent rows in table 10.4 break down the races in a number of ways. We see that the percentage of open-seat and challenger party primaries won by moderates is higher in 1998. Both moderate (i.e., less liberal) Democrats and moderate (i.e., less conservative) Republicans were more likely to win in 1998. These results are starker when the winner faced an extreme challenger and when voters in the district were moderate.
These comparisons indicate that, while the moderation effect holds across the various subsamples of races, there are also important differences in the overall level of candidate moderation in the different races. In other words, factors such as race type, party, presence of an extreme challenger, and district ideology appear to affect the probability that moderates will be nominated, independent of the effect of the blanket primary. To test this more complete model of candidate policy positions, I ran a multivariate logistic regression analysis. Table 10.5 reports the results of this analysis. The dependent variable is whether or not a moderate was elected in each race. The logit coefficients report the effect of each independent variable on the log of the odds ratio, that is, on the log of the ratio of the probabilities that a moderate candidate is and is not nominated.
The logit coefficients in column 2 (Model 1) show that once we control for the other factors that affect the probability of nominating a moderate
| District | Party | Race Type | Winner's Ideology | Second-Place Candidate's Ideology | Third-Place Candidate's Ideology | District Ideology |
|---|---|---|---|---|---|---|
| 1 | Dem. | Open | L | M | L | Split |
| 1 | Rep. | Open | M | M | Split | |
| 4 | Rep. | Open | C | Safe R | ||
| 8 | Dem. | Open | M | Split | ||
| 8 | Rep. | Open | M | Split | ||
| 9 | Dem. | Open | L | M | Safe D | |
| 11 | Dem. | Open | L | Safe D | ||
| 14 | Dem. | Open | L | L | L | Safe D |
| 15 | Rep. | Open | C | M | Strong R | |
| 16 | Dem. | Open | M | L | Safe D | |
| 19 | Dem. | Open | L | M | Safe D | |
| 22 | Dem. | Open | M | Strong D | ||
| 22 | Rep. | Open | M | C | Strong D | |
| 23 | Dem. | Open | L | M | L | Safe D |
| 26 | Dem. | Open | M | Split | ||
| 27 | Dem. | Open | L | Strong D | ||
| 27 | Rep. | Open | M | C | Strong D | |
| 28 | Rep. | Chal Open | C | M | Safe R | |
| 47 | Dem. | Open | M | Safe D | ||
| 49 | Dem. | Open | M | L | Safe D | |
| 50 | Dem. | Open | L | Safe D | ||
| 53 | Dem. | Open | Split | |||
| 54 | Rep. | Open | M | Split | ||
| 54 | Dem. | Open | M | Split | ||
| 55 | Dem. | Inc.'s party | M | Safe D | ||
| 57 | Dem. | Inc.'s party | L | Safe D | ||
| 60 | Rep. | Open | M | Split | ||
| 61 | Rep. | Open | C | Split | ||
| 62 | Dem. | Open | Strong D | |||
| 65 | Rep. | Inc.'s party | C | Safe R | ||
| 67 | Rep. | Inc.'s party | C | M | Strong R | |
| 68 | Rep. | Open | C | Safe R | ||
| 73 | Rep. | Open | C | Safe R | ||
| 75 | Rep. | Open | M | C | Safe R | |
| 76 | Rep. | Chal.'s party | C | M | Strong D | |
| 78 | Rep. | Chal.'s party | M | Split |
| 1996 | 1998 | p<z (1-tailed) | p<|z| (2-tailed) | |
|---|---|---|---|---|
| Average value lower in 1998; hence 1-tailed test is p > z. | ||||
| All | 37% | 50% | .12 | .24 |
| Open | 41 | 58 | .10 | .20 |
| Challenger's party | 25 | 33 | .40 | .81 |
| Democrats | 32 | 46 | .20 | .40 |
| Republicans | 41 | 52 | .23 | .45 |
| Extreme challenger | 42 | 60 | .18 | .36 |
| Moderate district | 48 | 63 | .18 | .36 |
| Vote margin (VM) | 18.66 | 21.03 | .26 | .52 |
| VM—Open seats[a] | 18.65 | 16.31 | .28 | .56 |
| VM—Challenger's party [a] | 21.25 | 16.25 | .29 | .59 |
| No. of candidates[a] | 3.61 | 3.51 | .40 | .80 |
| No. of candidates—Open seats[a] | 3.88 | 3.88 | .50 | 1.00 |
| No. of candidates—Challenger's party[a] | 2.25 | 2.00 | .31 | .62 |
| Independent Variable | Model 1 | Model 2 | Model 3 |
|---|---|---|---|
| NOTE: Asymptotic standard errors are in parentheses. | |||
| p < .10, 1-tailed test | |||
| p < .05, 1-tailed test | |||
| Blanket | 1.02[**] | 0.99[**] | 0.45 |
| (0.54) | (0.54) | (0.83) | |
| Open seat | 1.67[**] | 1.74[**] | 1.20 |
| (0.69) | (0.70) | (1.24) | |
| Moderate district | 1.49[**] | 1.46[**] | 1.53 |
| (0.55) | (0.56) | (0.89) | |
| Democrats | −0.39 | −1.73[**] | |
| (0.54) | (0.89) | ||
| Extreme challenger | 2.23[**] | ||
| (0.98) | |||
| Constant | −2.92[**] | −2.78[**] | −3.59[**] |
| (0.86) | (0.88) | (1.47) | |
| Pseudo-R2 | .13 | .14 | .26 |
| N | 75 | 75 | 46 |
Model 2 adds a dummy variable scored 1 if the race is a Democratic primary and scored 0 otherwise. The negative coefficient is small relative to the other estimated effects and not significant. Thus, once we control for primary type, race type, and district ideology, Democrats are no more likely to elect moderates than Republicans. Finally, Model 3 adds a dummy variable indicating whether the winner faced an extreme challenger. The coefficient on the extreme challenger variable is positive and significant, indicating that moderates are more successful when they face extreme challengers.[18] Inclusion of the extreme challenger variable increases the size and significance of the Democratic primary dummy. However, it also decreases the size and significance of the blanket primary variable. Further research, perhaps analysis of elections in other blanket primary states, is necessary to determine whether these effects are due to limited observations or some other considerations.[19]
Since logit coefficients express the estimated effect of each explanatory variable on the log of the odds ratio, they are difficult to interpret on their own. Therefore, it is common to report how the independent variables of interest affect the predicted probabilities of the event (in this case, of nominating a moderate candidate), holding constant the values of the other independent variables. I therefore convert the coefficients from column 2 into predicted probabilities. Table 10.6 reports the predicted probabilities of nominating moderates associated with closed and blanket primaries, under various combinations of race type and district ideology.
The main finding from table 10.6 is that for all combinations of race type and district ideology, the probability of nominating a moderate candidate is higher under the blanket primary. In the least favorable setting—when there is an incumbent running and the district is safe Democratic or safe Republican—the probability of nominating a moderate increases from .05 under a closed primary to .13 under a blanket primary. In the most favorable setting—where the race is an open seat and the district ideology is moderate—the probability of nominating a moderate increases from .56 under a closed primary to .78 under a blanket primary.
| Blanket | Open Seat | Moderate District | Z | P |
|---|---|---|---|---|
| 0 | 0 | 0 | −2.92 | .05 |
| 0 | 0 | 1 | −1.43 | .19 |
| 0 | 1 | 0 | −1.25 | .22 |
| 0 | 1 | 1 | 0.24 | .56 |
| 1 | 0 | 0 | −1.90 | .13 |
| 1 | 0 | 1 | −0.41 | .40 |
| 1 | 1 | 0 | −0.23 | .44 |
| 1 | 1 | 1 | 1.26 | .78 |
SUMMARY AND CONCLUSION
One of the most debated questions about the blanket primary is whether it would result in the nomination and ultimate election of moderate or extreme candidates. The potential advantages enjoyed by such candidates result from strategic voter behavior. Under the blanket primary, the costs of engaging in three types of voter behavior—sincere crossover, hedging, and raiding—are all substantially lowered. If voters engage primarily in sincere crossover and hedging, the blanket primary advantages moderate candidates. If voters engage primarily in raiding, the blanket primary advantages extreme candidates.
Analyzing the results of the 1998 California races for Governor, U.S. Senate, U.S. Congress, and State Assembly, I find some evidence for the moderation hypothesis and little evidence for the polarization hypothesis. At the top of the ballot, the blanket primary did not seem to have a major effect on election outcomes. Other factors, such as incumbency (or an incumbency-like appeal), seem to have been more important. If anything, the U.S. Senate race provides some evidence for the polarization hypothesis, since the least moderate Senate candidate was elected. The blanket primary seemed to make some difference in the competitive U.S. Congressional primaries, with moderates prevailing more often than extremists. And it clearly made a difference in the State Assembly races, leading to the nomination of a significantly greater number of moderates in 1998 compared to 1996. This is not to say that raiding never occurred or that extreme candidates never benefited from crossover voting made possible by the blanket primary. Rather, the evidence strongly suggests that the overall net effect of the blanket primary was to produce more moderate candidates.
NOTES
1. By "moderate" candidates, I simply mean those with centrist policy positions. Thus, moderate Democrats are relatively conservative, and moderate Republicans are relatively liberal. By "extreme," I mean far to either the liberal or conservative end of the ideological spectrum.
2. The other forms of primary elections used in the United States are the open and semiclosed. Twenty-three states use open primaries. In open primaries, participation is open to all registered voters, but each party has a separate ballot, and voters are restricted to participating in a single party's nominations in a given election. Eight states use semiclosed primaries, which are an intermediate form that allows independents and/or new voters to participate but not members of other parties. Louisiana uses a nonpartisan "primary," in which any candidate who receives a majority of the total vote automatically wins the office. If no candidate receives a majority, the top two vote receivers from any parties meet in a runoff (see Bott 1991).
3. It is also possible that existing candidates will change their policy positions. In Downsian fashion, candidates may move away from their hard-core partisan constituencies and converge towards the center to compete for the favor of the sincere crossovers and hedgers (Downs 1957). This "move" towards the center may occur either as existing candidates change their positions to adapt to the new electoral environment, or as new candidates enter with more moderate positions.
4. I also assume that all voters turn out, and that all actors know all aspects of the example.
5. This example is deliberately simplistic to illustrate the basic dynamics of voter choice under open or blanket primaries. I consider the consequences of relaxing some of the most restrictive assumptions, such as the voter distribution and candidate positions, below.
6. See Gerber and Morton 2001 for a more detailed examination of this basic model.
7. ADA scores are voting indices constructed by the liberal interest group Americans for Democratic Action. They indicate the proportion of times the legislator cast a liberal vote on selected bills. Gerber and Morton use ADA scores as indicators of legislators' ideology.
8. In fact, three other Republican candidates were listed on the ballot. However, none was ever considered a real threat to Lungren and, in the end, none received more than 4 percent of the vote.
9. Three other Democrats were listed on the ballot. All received less than 1 percent of the party vote.
10. Harman's 1996 ADA score was only 60 (with 100 being the most liberal), placing her among the most conservative Democrats in the House (Block and Buck 1997, 390).
11. One other Democratic candidate won 7.85 percent of the party vote. Four other Republican candidates were also listed on the ballot. One received 10.36 percent of the party vote, and the others received less than 2 percent each (California Secretary of State 1998c).
12. Boxer's 1996 ADA score was a perfect 100 (Block and Buck 1997, 341).
13. It is possible, of course, that candidates' retirement decisions are tied in part to strategic considerations related to the blanket primary. See Kiewiet and Zeng 1993 on strategic retirements.
14. My assessment of candidate policy positions is based on information from several sources, including candidate statements and election materials, newspaper coverage, and political consultant Allan Hoffenblum's California Target Book, a careful race-by-race assessment of state legislative races used extensively by campaign professionals. The Target Book is produced and updated throughout the election cycle. I thank Mr. Hoffenblum for his generosity in providing me access to the Target Book.
15. As with the U.S. Congressional races, I relied primarily on candidate statements, election materials, newspaper reports, and Hoffenblum's California Target Book to assess winning candidates' positions and district ideology.
16. In most cases, this information was available for the second-place candidate. In a few, it was also available for the third-place candidate.
17. Candidates who predated term limits, instituted in 1992, may have an incumbency advantage not enjoyed by post-term-limits candidates. Therefore, comparisons between 1998 and years prior to 1996 are severely suspect.
18. An alternative explanation posits the opposite causal story; that is, strong moderate candidates attract ideologically extreme challengers.
19. Reliable data on challenger policy positions were available for only forty-six races (thirty in 1996, sixteen in 1998).
REFERENCES
Block, A. G., and Claudia Buck. 1997. California Political Almanac. Sacramento, CA: State Net.
Bott, Alexander. 1991. Handbook of United States Election Laws and Practices.New York: Greenwood Press.
California Secretary of State. 1996. Ballot Pamphlet. Sacramento.
California Secretary of State. 1998a. California General Election Campaign Receipts, Expenditures, Cash on Hand, and Debts for State Candidates and Officeholders (July 1, 1998 through December 31, 1998). Sacramento.
California Secretary of State. 1998b. Statement of the Vote: Primary Election, June 2, 1998. Sacramento.
California Secretary of State. 1998c. Statement of the Vote: General Election, November 3, 1998. Sacramento.
California Secretary of State. 1998d. Ballot Pamphlet. Sacramento.
Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row.
Gerber, Elisabeth R., and Rebecca B. Morton. 1998. “Primary Election Systems and Representation.” Journal of Law, Economics, and Organization14,no. 2: 304–24.
Gerber, Elisabeth R., and Rebecca B. Morton. 2001. “Electoral Institutions and Party Competition: The Effects of Nomination Procedures on Electoral Coalition Formation.” Unpublished working paper. University of California, San Diego.
Grofman, Bernard, and Thomas L. Brunell. 2001. “Explaining the Ideological Differences Between Two U.S. Senators Elected from the Same State: An Institutional Effects Model.” In Peter F. Galderisi, Marni Ezra, and Michael Lyons, eds.,
Hoffenblum, Allan, ed. 1996a. California Target Book, March 1996. Los Angeles: Allan Hoffenblum and Associates.
Hoffenblum, Allan, ed. 1996b. California Target Book, May 1996. Los Angeles: Allan Hoffenblum and Associates.
Hoffenblum, Allan, ed. 1998a. California Target Book, Final Primary Edition. Los Angeles: Allan Hoffenblum and Associates.
Hoffenblum, Allan, ed. 1998b. California Target Book, General Election Edition. Los Angeles: Allan Hoffenblum and Associates.
Institute of Governmental Studies. 1999. Transcript, 1998 Governor's Race Conference. Berkeley, CA: n.p.
Kiewiet, D. Roderick, and Lanche Zeng. 1993. “An Analysis of Congressional Career Decisions.” American Political Science Review87:928–41.
King, David C. 1998. “Party Competition and Polarization in American Politics.” Paper presented at the Annual Meeting of the Midwest Political Science Association, Chicago.