While later chapters deal with a close up look at primary reform in one state, California, this chapter leverages a long time frame and the diversity of laws across states to examine how different primary laws affect political outcomes. This chapter builds on the database of laws described in the previous chapter. In particular, I focus on two main outcomes. First, do different primary laws produce more competitive elections? Second, do different primary laws produce more moderate candidates than others?
Data
I have supplemented the database of primary election laws with three kinds of data: demographic and economic information, the outcomes of primary and general election campaigns, and estimates of the ideal points of United States Senators. Pulled together, with some data-processing, I can then merge those data-sources with my database of laws. Each “observation” in my laws database reflects a state-even-year, e.g.
“Alabama 1968.” To the fullest extent reasonable, I have adjusted the other data to fit that format.
For non-political information about each state, I relied on the United State Decennial Census of 1940, 1950, 1960, 1970, 1980, 1990, 2000, and 2010. The database includes: state population, per capita income, percent of the population above age 65, percent of the population that graduated from high school, percent of the population with four years of college, percent of the population that is black, percent of the population identified as white, and the percent of the state that qualifies as urban. These variables seemed to have relatively stable definitions over time. Some of the other categories that might be nice to include, like percent Latino, have emerged comparatively recently as
official census categories. To fill in the intervening years, I just used a linear interpolation by state: Alabama’s 1984 percent white in the database reflects two fifths of the difference between its 1980 and 1990 census reports. A linear interpolation seems fairly safe here; these variables do not change dramatically year-by-year. For 2012, I just projected based on the data from 2000 and 2010.
Election data came from Congressional Quarterly’s “Voting and Elections Collection.” This database includes the number of votes and vote share of each candidate in U.S. Senate and gubernatorial elections back to 1968. The CQ also identified candidates as incumbents. I have collected the data for both general and primary elections. Not every state-election-year included an election for Senate or governor;
those entries are listed as “missing.” I did interpolate U.S. Presidential election vote totals for non-presidential years as a measure of state partisanship – but I intend to use that as an independent variable in the analysis. Some states elect governors in odd- numbered years; to keep the structure of the data consistent I assigned odd-numbered year gubernatorial elections to the year before them and then included an indicator variable for odd-numbered year elections to remove them later if this seemed like a problem.
For a measure of each state’s representation, I made use of the DW-Nominate scores.170 For each Congress, I computed a distance (using both dimensions) from each Senator to the point that represented that Congress’ mean (in both dimensions). Then I averaged across all Senators in each state-Congress to get a measure of the extremity of that state’s representation relative to the current Congress. The DW-Nominate database
170 I downloaded the DW-Nominate scores here: http://voteview.com/dwnominate.asp (last accessed 03/30/13).
organizes data by Senator-State-Congress; by this transformation, I have a variable that is just by State-Congress, which I can then easily map to election years. Even in years in which Senators are not up for election, they should contemplate the election laws of their state – in case those laws remain in place for the future. Conceivably, a law enacted after a Senator faced an election and repealed before the Senator faced reelection could still impact the Senator’s choices during their time in office. For example, if a state used a nonpartisan primary for a year, a Senator might consider positioning herself to appeal to a broader segment of the electorate.
Competitiveness
The type of primary election law a state uses may affect how competitive elections are in that state. Most states tend to lean towards one party or another. If we assume that elections tend to have a centralizing spatial effect, even without making very specific claims about achieving precisely the median voter, it is not a large intellectual leap to assume that a state with a closed primary rule and a dominant party should have large margins of victory. If the weaker party picks someone close to their median registered voter, and the dominant party picks someone close to their median registered voter, the candidate of the dominant party will end up much closer to the general election median voter.
Of course, if the two parties are evenly balanced and they each pick an extremist, measuring general election competitiveness does not measure proximity to the median at all. Two extremists might very well split the vote evenly between them, as most voters try to pick the lesser of two evils. Even so, competitiveness can be a desirable outcome:
it may engage citizens in the democratic process, compel politicians to take clearer issue
positions, encourage scrutiny of public officials for corruption, and so on. While political scientists can argue about the precise goods delivered by competitive elections, it at least stands to reason that the nomination process has failed to produce viable candidates for at least one party if general election margins of victory are large.
As a general rule, with many primary reform attempts, “moderate” and “viable”
are synonymous. Typically, parties attempt to open their primary process to produce more moderate (and therefore viable) candidates. Arguably this is what happened in Connecticut in the 1980s; at least there the Republicans considered a desire to win more elections.171 Moderate California Republicans also pushed for the blanket and top-two primaries in that state. There’s a back-of-the-envelope application of the median voter theorem that underlies all of this: letting unaffiliated moderate voters into the minority party primary “should” allow more moderate candidates to win the primary and then run more competitively in the general election.
The problem with all back-of-the-envelope political theory rears its ugly head in this case: the formal predictions are nowhere unambiguous and depend to a great deal on the assumptions of the strategic choices of both candidates and voters. As noted in the chapter reviewing the literature, several political scientists have commented that moderation may not vary along a linear “openness” dimension. Gerber and Morton (1998) specifically suggested that semi-closed primaries might produce the most moderate candidates because they would prevent partisan “raiding” while allowing unaffiliated moderate participation. Gerber and Morton hypothesized, and found, that semi-closed primaries produced more moderate candidates than open primaries.
171 Although, as mentioned in the previous chapter, Lowell Weicker emphasized the habitual voting aspect over the spatial modeling notions in Connecticut; nevertheless: ‘so he said.’
Overall, elections for “top of the ticket” statewide offices tend not to be very competitive in any case. Figure 3-1, below, plots the mean and median difference between the first and second place candidates for U.S Senate – or, if a Senate election did not occur, state governor – by election year across all the states. Notably, both the means and the medians increase over time; on average, elections for U.S. Senate and governor become less competitive. In 2012, both the mean and the median differences were more than fifteen percentage points. Most Senators and governors win big; one interpretation overall is that the nomination process fails to provide the winners with credible competition more often than not.
Figure 3-1: Margin of Victory (Percent Difference Between 1st & 2nd Place); Means and Medians for “top of the ticket” U.S. Elections, 1968–2012.
0 5 10 15 20 25 30
1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 2012
Margin of Victory
Mean Median Linear (Mean) Linear (Median)
Figure 3-2: Percent Margin of Victory, “Top of Ticket” Elections, 1968–2012 (N=968)
Of the 968 elections included in this study, the victor won by less than five percent in less than 200 of them. Figure 3-2 provides additional context for Figure 3-1;
while the high means are driven in part by a few very high outliers (uncontested elections), most of the explanation lies with a broader lack of competition affecting most elections. There are two ways to contemplate “competitiveness.” As discussed in Born (1981) and Atkeson (1998) in the context of primary competitiveness, early studies in that area tended to come up with an arbitrary cutoff of what constituted “competitive.”
Using the whole continuous variable does have some advantages; the continuous version of the difference does capture much more information about the election. On the other hand, the authors of the older studies also had a point: when one or both sides stop trying, the margin of victory may not measure much about that election at all. It is a bit like
050100150200
Number of Elections
0 20 40 60 80 100
Margin of Victory
yards generated at the end of a lopsided football game; they are not very predictive of future values because the backups are in the game. I created a variable which indicates if the winner triumphed by less than ten percentage points (342 elections, or 35 percent of the elections in the study); I will use both measures in the analysis. These cutoffs are arbitrary; to capture the sense of the competitive election I had hoped to use the five percent cutoff but needed to adopt the ten percent to ensure that all states had at least one
“competitive” election.
Figure 3-3: Types of Primary Laws Used 1946–2012
Some policymakers may view the lack of competition as a problem to solve; this section addresses whether or not different primary laws might help solve it. Figure 3-3 shows the number of each type of election law used in each year from 1968 through
0 5 10 15 20 25 30
1944 1948 1952 1956 1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 2012
Number of States Using Each Type
Open Blanket Semiclosed Closed
Asymmetric Nonpartisan Convention
2012. Over this period of time, the number of true closed primaries diminishes and states (and parties) use more semi-closed primaries. The Tashjian case accelerated this trend as parties periodically opted for semi-closed rules. Figure 3-3 shows the results of what parties decided; in some cases, both major parties made different rules (the “asymmetric”
category). Most states over this period used either an open, closed, or semi-closed primary. Typically states that used an open primary stuck with it, although there are periodic changes. I have included the open runoff primaries in with the other open primaries in Figure 3-3 as well. States rarely used the alternative nomination procedure types: conventions, blanket primaries, nonpartisan primaries.
I have tried conducting the analysis in several different ways, including combining some of these primary categories. Open primaries, broadly defined, could include the (now unconstitutional) blanket primary. Especially since I have drawn competitiveness data from the top of the statewide ticket, the multiple crossover effect in the blanket primary is a lot more likely to take place farther down the ballot. For example, if a Republican voter wants to vote for a Democratic U.S. Senate primary candidate, that voter is likely to vote in the Democratic race in both the open and the blanket primary. If a voter wants to vote in their own Senate primary, the desire to vote in the other party’s primary for other races – lower house of the state legislature, perhaps – seems unlikely to draw the voter to cross over. I suspect that the blanket primary would have a larger effect against an open primary farther down the ballot; both because a voter drawn across for the top of the ticket might change back and because a voter might jump across the aisle farther down the ticket. This is a hypothesis rather than a fact, of course;
nevertheless, it is sufficiently credible to deserve testing in the models. In the broad-open
category I have excluded challenge primaries (like Utah); instead, I have grouped the
“challenge primary” states with the conventions for these broader and simplified categories.
I have also generated a category for closed primaries, broadly defined. The semi- closed primary has much more in common with the closed primary than with the open primary. For most of the electorate (in most cases, anyway) semi-closed primaries and closed primaries operate identically; registered partisans cannot cast crossover ballots without changing their registration status. In most semi-closed primaries, commentators across many states lamented low participation levels by unaffiliated or independent voters. Campaigns may not have had the microtargeting infrastructure (at least, until very recently) to figure out which unaffiliated voters to try to mobilize. In this category I have excluded challenge primary states (like Connecticut, until recently) since they may be more properly grouped with the conventions.
I have grouped the challenge primaries and the conventions together under the heading of “party control.” As with the broadly defined open and closed categories, I will use this as a variable for alternative specifications to my original classification scheme. I have not included here the states with weak pre-primary institutions (like Colorado); I have only included the challenge primaries that have no petition “work- around” for alternative candidates to use to get on the ballot. Starting in 1968, there are very few elections that use conventions in the database – only six. The “party control”
category has 43 observations over that span (from Connecticut, Delaware, Indiana, Utah).
Since meaningful primaries rarely occurred with challenge primaries, considering these together also makes sense.
In some of the regressions, I have also grouped together the “nonpartisan” and
“blanket” primaries under an “anti-party” heading. In Washington, California, and Alaska one can make plausible inferences about the intentions of the authors of the laws:
to prevent strong party control over the nomination process. Louisiana seems to have acquired the nonpartisan primary for other reasons (to cut out the Republicans entirely;
see previous chapter). Nevertheless, these types of laws have constituted popular reforms. It may make sense to consider them together. It is, though, not possible to use both the “anti-party” grouping and the “broad open” heading simultaneously as blanket primaries appear in both.
In the analysis of these laws, the party-choice laws could be coded two different ways. I have tested both. The first alternative, leaving them all coded as “party choice,”
reflects the notion that whatever choice the parties make is an equilibrium outcome in a strategic contest; that type of coding tests whether letting the parties choose increases or reduces competition. The second alternative, coding their choice, sets up a test on the more specific rule. A “party” is not a rational actor; it is instead made up of many individuals with their own agendas. Parties may not always make sensible choices (or their leadership may consider different dimensions of the problem). Periodically parties do not make symmetric choices. There are 21 cases in which an election occurred for one of these top offices with asymmetric rule choices. Those are interesting opportunities for analysis that would be missed by just coding each primary as “party choice.”
Do primary laws seem to correlate with margins of victory? Table 3-1, below, displays the mean and median margin of victory for the “top of the ticket” race by primary type for 1968 to 2012. Some of these categories obviously have more entries
than others; there are just 15 nonpartisan elections that meet the requirements while there are 439 open primaries. The figures in Table 3-1 should be surprising; contrary to expectations, closed primaries (either broadly or narrowly defined) have smaller margins of victory than the alternatives. The six general elections generated by conventions had the smallest margins of all (about three times smaller), although with only six observations that should be taken with a grain of salt.
Table 3-1: Margin of Victory, Primary Type
Category Type Mean Median N
Broad Broad Open 21.3 16.6 450
Nonpartisan 23.2 21.7 15
Broad Closed 18.4 14.3 460
Party Control 21.2 16.9 43
Narrow Open 21.6 16.9 439
Blanket 22.7 13.7 30
Semi-Closed (Choice) 20.6 14.6 97
Closed (Choice) 17.5 14.1 360
Asymmetric (Choice) 23.4 23.0 21
Nonpartisan 23.2 21.7 15
Convention 7.0 3.9 6
Of course, many considerations besides primary type determine general election margin of victory. This section presents several approaches to estimating the effect of primary type on general election victory margins. Closed primaries do seem to produce closer elections than some of the other alternatives, even when controlling for other possible sources of variation. The different models vary by how primary types are categorized or what types of elections are included. The main result remains robust to all of the different models presented below.
In Table 3-2, I display the results for two models using narrowly defined primary election type categories. Both of these linear models include time series and cross-
section (state) fixed effects.172 The only difference between them is that the left model codes states by their laws and the right model codes states by their laws and, if allowed, the choices of the parties. Both models have the same number of observations (968), cross sections (50), and observations per cross section (ranging from 14 to 23). Both models use the winning percentage of the ‘top of the ticket’ candidate – either a Senator or, lacking a Senator, a governor. A positive coefficient implies an increase in the winning percentage, a less competitive election. In each model, I have specified all the types of primaries but one; the coefficients on the other primaries should be interpreted relative to the variable I excluded.
Open primaries produce much less competitive elections than closed primaries.
In the “laws” model reported in Table 3-2, both open and blanket primaries had notably larger margins of victory, statistically significant at the .05 level. The semi-closed and nonpartisan primaries had positive coefficients that did not reach the .05 level. The
“party choice” and convention nomination procedures had negative and insignificant coefficients. Many parties decided to continue with the closed primary and there are very few convention observations, making neither of those results wholly unsurprising. In the version of the model that replaces the “party choice” variable with the actual choice made (now excluding both primaries closed by law and closed by both parties by choice), the general results are the same. Both open and blanket primaries produced greater margins of victory than closed primaries; the model estimated not only significant coefficients but also large ones. Once again semi-closed primaries are not significantly different than
172 Estimated using STATA’s xtset and xtreg commands.
closed primaries and the results for nonpartisan primaries and conventions are much the same. Interestingly, the “asymmetric” primaries do not produce different margins of Table 3-2: “Top of Ticket” Margin of Victory, Narrow Categories; TSCS Fixed Effects
.
Model:
Laws
Model:
Laws & Choice
Variable Coef. T-Stat. Coef. T-Stat.
Open 13.08* 2.35 13.54* 2.44
Blanket 15.63* 2.77 17.15* 3.04
Semi-Closed (Law) 3.76 1.04
Semi-Closed (Law & Choice) 1.82 0.60
Party Choice -0.58 -0.20
Asymmetric Choice -0.16 -0.04
Nonpartisan 6.10 0.81 7.05 0.93
Convention -0.55 -0.05 -0.32 -0.03
Incumbent in Race 7.90* 6.88 7.88* 6.85
Pre-Primary Hurdle -1.04 -0.28 -0.96 -0.26
Runoff -8.26 -1.65 -8.41 -1.70
Presidential MofV 0.03 0.41 0.02 0.27
Population (100,000) -0.09 -1.90 -0.10* -2.05 Per Capita Income (1000s) 0.53* 2.56 0.55* 2.71 Percent Over Age 65 -0.81 -1.34 -0.84 -1.39 Percent High School Grad -0.06 -0.55 -0.06 -0.58
Percent Black -0.39 -0.55 -0.37 -0.51
Percent Urban -0.04 -0.18 -0.06 -0.30
Gubernatorial Odd Year Elections 0.42 0.05 0.40 0.05 Presidential Election Year -0.26 -0.25 -0.24 -0.24
Constant 26.31 1.50 28.29 1.63
Excluded Variable
Closed (Law)
Closed (Law & Choice)
Overall R-Sq. 0.0392 0.0387
N 968
50 14 23 Groups
Obs. Min Obs. Max