Competitiveness and Vote Share

Did the top-two produce competitive elections? We begin by looking solely at the outcomes from the 2012 election cycle. There are really three possible measures of competitiveness for an election where two candidates advance to a final stage. The first measure is just to look at what percentage of the vote the first-place candidate won in the primary. That represents, in a sense, the strength of the candidate’s core constituency. The second measure, of course, is to look at the difference between second and third place in the primary; that is, how close was the second-place candidate to elimination? The third possibility is to skip the share of the votes won in the primary – because those are just a means to an end – and ask if the primary system produces a competitive general election between the final two candidates.

Figure 4-1 displays how many elections for the House of Representatives, State Senate, and State Assembly had a first-place candidate within five-point ranges of a certain percentage of the primary vote. The modal (most common) category had the first-place finisher with between 55 to 60 percent of the vote (the mean was 54.8 percent, and the median 55.5 percent). The notion of “competitive” is not strictly defined here, but if the top finisher of the primary gets more than 55 or 60 percent of the vote, that candidate is quite likely to win the general election as well. Nevertheless, this means that roughly half of the elections were not obviously foregone conclusions after the primary stage.

Figure 4-1 Percentage of Vote for First-Place Candidate, June Primary

Restricted to all 153 House, State Senate, and State Assembly races.

The competition for second place could be quite fierce. About one-third of the 106 elections with at least three candidates had the third- and second-place finishers within 5 percentage points, and half were within 10 points. Figure 4-2 shows there were still some races for second place that were completely uncompetitive – an event that requires two circumstances: the second-place candidate has to be relatively strong (i.e., if the first-place candidate wins 99 percent of the vote, the maximum possible distance between second and third place is 1 percentage point), and the third-place candidate must be truly hopeless.

Figure 4-2 Percentage Point Difference between Second and Third Place, June Primary

Restricted to 106 districts with at least three candidates (50 AD, 10 SD, 56 CD).

One of the issues with using the actual primary vote totals to contemplate competitiveness is that some candidates may have made strategic choices not to try very hard in the primary. A candidate likely to advance to the next stage could decide, for example, to hoard financial resources for the general election. Because two candidates advance, the candidate likely to come in first place may have a lower incentive to expend effort; if things go badly, the candidate can still place second and try to make up the ground in November. In that November election, though, both candidates should exert their maximal effort as only one can win the prize and obtain public office. Figure 4-3 shows the general election results by plotting the vote share of the winning candidate. Because the top-two limits the general election to only two candidates, the losing percentage is just 100 minus the winning percentage.

Figure 4-3 Percentage of Vote for Winning Candidates, November General Election

Roughly 20 percent of the 153 general elections for Assembly, State Senate, and House had a winning percentage under 55. About 44 percent of the elections were under a winning percentage of 60. About 10 percent – sixteen total – of the elections were under 52.3. A few elections were very close indeed: AD36, 50.1 percent; CD24, 50.1 percent; AD20, 50.3 percent; AD40, 50.4 percent; AD18, 50.5 percent; AD50, 50.5 percent; SD5, 50.5 percent. Of those, AD18, AD20, and AD50 were same-party runoffs; that is, three of the seven closest races were close on Election Day because they had candidates running from the same party.

Figure 4-4 shows that coming in first in the primary typically translated into winning the general election, as well – but there are exceptions. In seven of the twenty-nine same-party runoffs, the candidate who came in first during the primary lost in the general election (24.2 percent of the time). In the 124 other elections at the House, State Senate, and State Assembly level, the top candidate in the primary lost only sixteen elections (or 12.9 percent of the time). Overall, 15 percent of candidates who came in first on the day of the primary went on to lose in November. What Figure 4-4 suggests (via the nearly empty bottom-right quadrant) is that few of those defeats were total surprises.

Figure 4-4 How Did the First-Place Primary Finishers Fare in the General Election?

1 = Same-Party Runoff

0 = Other Type of Election

A second way to think about competitiveness is to ask whether the 2012 general and primary Assembly elections were more (or less) competitive than elections in prior years, before the advent of the top-two in 2012. While this is somewhat of a tricky exercise (as we can’t easily determine cause and effect with just election returns data from a limited and recent span of elections),² we do believe it is important now to see if there are any clear signs that the 2012 elections were more competitive than the set of elections that immediately preceded them.

Our strategy was to collect the general election vote shares of the top-two finishers, as well as the vote percentages received by the top-two Democrats and top-two Republicans in primary Assembly elections, going back to the 2004 elections. We obtained these data from the various Statements of the Vote, available from the California Secretary of State.

The easiest approach is to look at general election outcomes and to look simply at the vote returns of winning candidates in Assembly general elections, in 2004–2010 as compared with 2012. In a sense, this is similar to the seminal studies of Mayhew (Reference Mayhew1974) and Fiorina (Reference Fiorina1977), both of whom studied the “vanishing marginal.” The hypothesis here is whether we see that so-called marginal elections, those that are highly competitive and thus have a winning percentage near 50, are more or less likely to have occurred in 2012 relative to the prior Assembly general elections.

The top panel of Figure 4-5 shows the distribution of general election votes received by winning candidates in Assembly races, 2004–2010. The bottom panel gives the same distribution for winning candidates in the 2012 Assembly general elections. The results shown in Figure 4-5 are relatively clear – there is a shift toward more competitive races in 2012, relative to past Assembly general election races. That is, in 2012 there were more races clustered right at 50 to 60 percent than in the prior elections, so this graph seems to show a trend toward more competitive Assembly general elections in 2012 than in the previous decade. We also computed all of the summary statistics for the 2012 general elections at the Assembly level, relative to those in the previous decade:

1. – The mean in 2012 was 63.07, compared with 66.82 in 2010, 68.50 in 2008, 67.64 in 2006, and 66.37 in 2004.

2. – The median in 2012 was 61.15, compared with 63.75, 66.30, 64.65, and 64.30 in 2010, 2008, 2006, and 2004, respectively.

3. – The first and third quartiles of the distributions also shifted downward in 2012, compared with these previous elections.

Again, we cannot easily separate out cause and effect, nor can we isolate the specific effects of the top-two as compared with other factors at play in 2012. But these data show a trend toward more competitive Assembly general elections in 2012, and we will need to study additional elections in the future to determine if the “marginals” are reappearing.

Figure 4-5 “Vanishing Marginals,” General Elections

We can ask a similar question about the relative competitiveness of the primaries in 2012 as compared with earlier Assembly primary elections. This is in a sense something of an “apples and oranges” comparison, as primaries in California prior to 2012 were intra-party affairs, while the top-two primary in 2012 was a cross-party affair. But we feel it is nonetheless instructive to examine how winning vote margins, by both Democratic and Republican primary winners, appeared in primary elections before 2012, as compared with the 2012 election. We provide these graphs in Figures 4-6 and 4-7.

Figure 4-6 Democratic Party Primary Highest Vote, 2010–2012

Figure 4-7 Republican Party Primary Highest Vote, 2010–2012

What we see in both figures is dramatic. The degree to which primary elections prior to 2012 were uncompetitive is easy to see – the large number of races where winning candidates achieved a 100 percent vote share in Assembly primary elections before 2012 is amazing. However, in both of the graphics for 2012, we see a shift away from that tendency (which of course is exactly how the top-two primary will work, by definition – unless a candidate runs without any opposition at all in the primary, then by definition candidates will all tend to have vote shares of less than 100 percent). But what we see in 2012 is stark – the top candidate in either party received a vote share usually in the 20 to 40 percent range. Thus, we are seeing more competitive primaries, by this metric, which will be an outcome of the top-two primary. Whether shifting from a primary system in which many candidates receive 100 percent of the vote (and face no competition at all in a primary election) is a better system than the top-two is of course one of the questions that needs further research, after we can accumulate more data on the behavior of those elected to office under the top-two primary system.

Coordination, “Strategic Desertion,” and Pathological Results

The top-two primary system, like any electoral system (and, indeed, formal theory would suggest that this is not a solvable problem – see Chapter 2), has some weaknesses. Because many candidates can enter the contest, it is possible that extreme candidates relying on a committed base could take advantage of voters’ failures to coordinate on a reasonable alternative. Take, for example, an election with one very right-wing candidate, one very left-wing candidate, and many centrist candidates. If the moderate voters cannot decide which moderate candidate to follow, they could split the vote between all of them. Instead of picking, as intended, two moderate alternatives, the two extreme candidates could get the vote from the extreme of their side of the electorate and advance to the general election. Despite the fact that most of the electorate would probably prefer any of the moderate alternatives to either extremist, they would be stuck with those two candidates because they had failed to coordinate. The ultimate election might be quite close – that is, competitive – without the candidates locating anywhere close to the median voter.

This is not just an abstract academic concern. Figure 4-8 shows the vote share of the first-place candidate in the primary and the number of candidates competing in the election. As the number of candidates increases, the vote share of the first-place candidate decreases. In Congressional District 8, the first-place (Gregg Imus, R) and second-place (Paul Cook, R) candidates earned 16 percent and 15 percent of the vote in a field of ten Republicans, two Democrats, and one “No Party Preference” candidate. The third-place candidate (Phil Liberatore, R) fell 240 votes shy of Cook. One risk with many candidates and a top-two primary is that the vote splits among so many possible contenders that two candidates from the same party desired by a very small number of voters (in this case, 31 percent between both Imus and Cook) could end up as the only alternatives. The data in Figure 4-8 show that the Imus-Cook election is a rare event; in just about every election, at least one candidate got more than 25 percent of the vote. Cook would defeat Imus with 57 percent of the vote in November. Nevertheless, the race in CD8 demonstrates that it is possible for someone to advance with very little support in the district if the vote splits enough ways.

Figure 4-8 Vote Share and Number of Candidates

Voters may find it difficult to coordinate on an alternative candidate in races with many choices, even if they wish to do so. One measure of whether a “coordination failure” may have occurred is if the percentage of the vote for the fourth-place (and on to the last-place) candidate exceeds the difference between the second- and third-place candidates. That is, if all the voters who picked the fourth- to last-place candidates picked the third-place candidate, the result of the election would be different. In the races with at least four candidates, this happens frequently (see Figure 4-9). Out of sixty-four races (Assembly, State Senate, and U.S. House) with at least four candidates, voters may have lost an opportunity to change the outcome in forty of them (if they all voted for the third-place candidate instead), as shown in Table 4-4. This is particularly interesting because these coordination failures may be associated with same-party runoffs – the situation where one party splits enough of the vote so that two candidates of the other party advance to the November election.

Table 4-4 Potential Coordination Failures and Same-Party Runoffs

Table 4-4 (Minimum Number of Candidates = 4).

Figure 4-9 Coordination Failures

The most problematic example of this kind of coordination failure occurred in Congressional District 31. CD31 featured a race between two Republicans in November: Gary Miller (an incumbent) and Bob Dutton. In the primary, Miller earned 27 percent of the vote, and Dutton earned 25 percent. Four Democrats also ran, splitting between them 48 percent of the vote. The closest Democrat, Pete Aguilar, had 23 percent of the vote; that meant that the fourth-, fifth-, and sixth-place Democrats split 25 percent of the vote. Because the difference between Dutton and Aguilar was only 2 percent, these votes for later-place Democrats cost the Democratic Party an opportunity to have a candidate in a congressional district in which the Democrats have a six-point registration advantage over the Republicans, and 49 percent of the population is Latino and 12 percent African-American. Miller, the candidate endorsed by the California Republican Party, would go on to win the election in November with 55.2 percent of the vote.

The Miller-Dutton type of election is unusual. Table 4-5 shows the types of elections that occur in different kinds of districts: Republican-leaning districts, Democratic-leaning districts, and districts in which Republican and Democratic registration numbers are within 5 percentage points (“toss-ups”). There is one race (Miller-Dutton) between two Republicans in a Democratic-leaning district and no elections between two Democrats in Republican-leaning districts. These results hold true even if the “toss-up” category is eliminated and we look at just the difference between Republican and Democratic registration. Even if the Miller-Dutton-type race is not normatively optimal, it is rare.

Table 4-5 Election Type by Partisan District Type (Based on Registration)

Race, Ethnicity, and the Same-Party Runoff

The top-two raises some voting rights questions in part because it shares some characteristics with the Southern runoff primary system. While we return in more detail to this subject in Chapter 9, there are some methodological approaches that make sense to discuss here. For many years, most states in the American South employed a three-level election system. First, each party would hold a primary election. If one candidate failed to reach a certain threshold of the vote (typically 50 percent), then that party would hold a runoff between the top-two finishers within that party. The winner of the runoff would be the nominee of the party, and the nominees of all parties would compete in the general election. Because most of these states were effectively controlled entirely by the Democratic Party (see Key 1949), winning the nomination of the Democratic Party effectively won most elections. So while the formal structure of the election was quite different (allowed as many parties as qualified at the general election, and each party got one nominee), the practical effect was quite similar.

The Southern runoff primary raised voting rights concerns for some scholars. The general charge is that the system worked to prevent African-American candidates from winning the primary. If the vote split between too many white candidates, the runoff would allow the white voters to coordinate on a white candidate at the expense of a black candidate. While scholars debate the extent that this is true (see Kousser Reference Kousser1984, Bullock and Johnson Reference Bullock and Johnson1985, Bullock and Smith Reference Bullock and Smith1990, and Bullock and Johnson Reference Bullock and Johnson1992), the potential for such an effect in California deserves scrutiny. In California, racial politics are a bit more complicated than in many of the Southern states in the heyday of the Southern runoff: the state has a large and diverse population of many non-white racial and ethnic groups.

Figure 4-10 displays the number of districts within certain ranges of Latino population (as reported from the 2010 Census in Hoffenblum et al. Reference Hoffenblum, Pross, Griego, Howard, Quinn and Sragow2012). This includes districts of varying size, since State Senate districts (of which there are forty, with twenty up for election in 2012) are larger than House districts (fifty-three of those), and House districts are larger than State Assembly districts (eighty of those). The existence of majority-minority Latino districts is quite evident in Figure 4-10; there is a whole group of districts with between 65 and 70 percent Latino population. Latinos still make up electorally very significant fractions of the population in most districts, though; the modal category of 25 to 30 percent still gives a candidate capable of uniting the Latino electorate a considerable amount of voting power.

Figure 4-10 Number of Districts and Hispanic/Latino Population Percentage

The relationship between racial composition and whether a same-party runoff occurs can provide one piece of evidence to evaluate whether the top-two may disadvantage non-white groups. Figure 4-11 plots the percentage of a district registered Republican with the percentage of a district that is white (and not Hispanic). Generally speaking, those variables are positively correlated because Republicans are more likely to be white. Nevertheless, there are still some very Democratic districts that are “more white” than others. If the same-party Democratic runoffs strongly tended to occur mostly in the districts that were less white (while still very Democratic), this might be cause for concern. Nevertheless, Figure 4-11 shows no obviously strong tendency for this to occur; same-party Democratic runoffs are marked with a “1.” There are some same-party Democratic runoffs in districts that have very few white voters but also some in districts that are majority white.

Figure 4-11 Race, Partisanship, and Democratic Runoffs

0: Other Election

1: Democratic Same-Party Runoff

Summary

The top-two appears to have produced more competitive elections than we would have seen under the old primary system, although it is important to be cautious about causality. Figure 4-12 shows the percentage of the vote earned by the winning candidate in the November election and the difference between Democratic and Republican registration. The pattern looks like a V, with the point falling near 0 on the horizontal axis and 50 on the vertical axis – the point representing a very close election in a very competitive district. The same-party runoffs, represented in that figure with 1s instead of 0s, appear to be raining out from under the V – the same-party runoffs look like they were closer elections than the party split would suggest.

Figure 4-12 Partisanship and Winning Percentage

The top-two primary did not completely upend the California political process and magically transform a state with many uncompetitive elections into a state with mostly competitive elections. Nevertheless, some of the elections seem to have been more competitive. This intended, if small, result seems to have come without too many unintended negative consequences. In only one of 153 elections did the November ballot not feature a candidate of the strongest party because that party split its vote among too many candidates. Furthermore, most first-place candidates in the primary won at least 25 percent of the vote, so at least one of the candidates on the November ballot had a reasonable quantity of support in the district. These results suggest that the policy was at least a limited success in its first election year.