Chapter 5
Nominating Systems, Strategic Entry
and Candidate Qualifications∗
Shigeo Hirano
Columbia University
James M. Snyder Jr.
Harvard University and NBER
May 7, 2015
∗
This is a very preliminary draft of a chapter for our book on primary elections. We are still gathering
additional data for some of the analyses so the results may change in future versions. Please do not cite or
circulate without permission of the authors.
1
1
Introduction
Do primary elections result in high quality candidates at the general election stage? For many
progressive reformers at the turn of the 20th century, improving the quality of party nominees
was one key motivation for supporting the introduction of direct primaries. These reformers
believed that nominating conventions and caucuses were easily manipulated by party bosses,
and party bosses were less interested in high quality nominees than in nominees who were
willing and able to work in the party machine’s interests. Allowing voters to directly select
party nominees would not only increase the importance of candidate quality relative to
party loyalty as a determinant of the nominations, but it would also expand the pool of high
quality candidates competing for the nomination. In particular, many of the candidates
who may have been unwilling to engage in the corrupt practices associated with the caucus
and convention systems were expected to be more willing to compete in primaries. Thus,
reformers asserted that primaries would naturally “improve the character of nominees” and
“bring into public service the best talent of the times.”1,2
Early critics of primaries argued that opening up the nomination process would have
the opposite effect. They questioned whether primary voters would actually choose high
quality nominees even if such candidates were available, since voters were largely uninformed
and easily persuaded to support candidates based on non-quality-related characteristics. In
addition, the high costs of running in the primaries, and the negative and personal nature
of primary campaigns, would deter many high quality candidates from entering.3
After the initial adoption of primaries, a number of political observers claimed that the
primaries were having only a limited effect on the quality of elected officials. For example,
1
These quotes are from a statement by Robert M. La Follette Sr. as reported in Boots (1917: 62).
Some studies of primaries outside the U.S. find evidence suggesting that voluntary primaries lead to
higher quality nominees. Carey and Polga-Hecimovich (2006) and Aragon (2014) find that presidential
candidates nominated by primaries receive higher vote shares in the general election; although Kemahlioglu,
Weitz-Shapiro and Hirano (2009) critique the data and some of the analyses in Carey and Polga-Hecimovich
(2006).
3
William H. Wilson writes, “The primary election system prevents the people from selecting the ablest
men and the men of highest character as candidates for office, and restricts the choice of the people to those
who are willing to rush forward and inject themselves into what is frequently a vicious election in order to
obtain nomination.” Case and Comment. 12:296-9. October 1916.
2
2
in 1919 William B. Munro gave the following assessment of the nomination reform in his
popular textbook on American government:
Has [the primary system] proved superior to the convention as a means of securing
capable legislators in the several states? On the whole, perhaps it has, although
there is no certainty in that direction. At its best the convention was capable
of making excellent selections, the fruit of careful deliberation. The primary has
not often shown itself able to reach as high a standard. On the other hand the
convention at its worst could strike occasionally a plane of arrogance, trickery,
and corruption to which a primary rarely if ever descends. In a word, the primary
seems to afford protection against the worst fault of the convention, which was
the frequent selection of incapable and corrupt candidates at the behest of a few
political leaders.4
However, these lukewarm assessments were of the overall impact of primaries on candidate
quality. As discussed above, we should be most concerned with how primaries are performing
in safe districts with open seat elections.
We begin this chapter by discussing a simple theoretical framework for thinking about
how primaries could affect the quality of nominees. Using a decision theoretic model with
politicians making strategic entry decisions, we provide a rationale for why primaries might
be an effective mechanism for nominating high quality candidates for advantaged parties in
safe districts. We also present some theoretical considerations for why the relative quality
of candidates nominated in safe versus competitive general election constituencies would be
higher under primaries than convention systems.
The bulk of this chapter examines the empirical evidence that primaries affected the
quality of party nominees in years immediately after the nomination system was adopted.
One of the key challenges to empirically assessing this relationship is the difficulty measuring
quality of candidates for public office. We focus on a particular aspect of quality which is the
ability of elected officials to perform the functions of office they are seeking. We use previous
4
Munro (1919: 419)
3
political experience in positions involving tasks similar to those done by officials in the
office being sought as an indicator for having qualifications for the office. We provide some
discussion of the appropriateness of prior relevant experience as a measure of qualification
for office.
Our main empirical analyses focus on the period 1892 to 1928. This period covers the
10 years prior to the year that the first state law mandating congressional primaries became
effective (Minnesota in 1902) and the 10 years after the last year such a law went into effect
among the states that passed mandatory primary laws during the progressive era (Indiana in
1918), i.e., it provides a reasonable sample of “pre-primary” and “post-primary” elections for
all of the states that passed mandatory primary laws during the progressive era. Consistent
with the predictions from our model, candidates with relevant prior experience were more
likely to compete and win in advantaged party primaries compared to competitive or disadvantaged party primaries. This relationship between nominee qualification and constituency
type is even more pronounced under the primary system as compared to the convention
system.
2
Theoretical Considerations
Since contested primary campaigns are quite costly, in most cases more costly than “throwing
ones hat in the ring” in a convention, the strategic entry decisions of candidates are an
especially important issue for primaries.5 The likelihood that a primary will attract high
quality candidates depends upon the incentives the pool of potential candidates face in a
given election. We begin this section with a simple model of strategic entry to illustrate this
point and to show how the incentives are affected by the partisan loyalties in a constituency.
After discussing strategic entry under primaries, we turn to a discussion of how the quality
of party nominees may be expected to differ between primaries and conventions.
5
Even the bribes sometimes paid in conventions were probably small compared to the costs of a serious
primary campaign.
4
2.1
Strategic Entry and Primary Elections
Two key factors are particularly relevant in determining the quality of the candidates competing in a given primary: the “pool” of potential high quality candidates in the district and
the likelihood that the winner of the primary will also win the general election.
Primary elections can be roughly divided into three types: (1) Primaries for the “advantaged” party in safe constituencies, where safe means that one party’s candidate is clearly
favored to win the general election due to the distribution of voter partisanship or ideology;
(2) Primaries for the “disadvantaged” party in safe constituencies; and (3) Primaries in competitive districts where partisan loyalties are relatively balanced between the two parties so
that neither party’s candidates are clearly expected to win the general election.6
We model one party’s primary election taking events in the opposing party as given.
There are two types of candidates, high quality (H) and low quality (L). The fraction of
high quality candidates is θ. The probability that a high (low) quality candidate wins in the
general election is GH (GL ). These parameters will vary across districts. In particular, they
will all be (weakly) larger in districts that are safer for the party. If the primary is contested
and both candidates have the same quality, then each wins with probability 1/2. If one
candidate is high quality and the other is low quality then the probability the high quality
candidate wins is P > 1/2.7 The value of office is V > 0 for all candidates. Each candidate
pays a cost of running, C, that is drawn from a uniform distribution on the interval [0, 1].
Assume that the parameters that represent probabilities – θ, GH , GL , and P – all lie strictly
between 0 and 1.
One candidate always enters. Nature then draws a second candidate, and that candidate
6
Scholars have long recognized the interplay between the general election environment and primary elections. For example, Key (1956: 79) writes, “such factors as urbanism and incumbency may affect the
frequency of primary competition through time but... in the long run the incidence of primary competition is chiefly a function of the prospects for victory in the general election.” Similarly, Sorauf (1963: 111)
states, “The basic determinant of primary competition in Pennsylvania remains the prospect of victory in
the [general] election.” These earlier scholars did not make the link to candidate quality, however.
7
Note that P is similar to, but not exactly the same as, P in the previous section. In the previous
section P was simply that information was revealed, while here it is the probability that quality is revealed
or voters choose the higher quality by chance. Thus, if the probability information is revealed by P 0 then
P = P 0 + (1 − P 0 )/2 = (1 + P 0 )/2. Note that our assumption P > 1/2 is innocuous.
5
decides whether or not to run. The candidate will run if and only if the expected benefit
exceeds her cost of running. Let t1 denote the first candidate’s type, let t2 denote the second
candidate’s type. Then there are four possible cases for the vector (t1 , t2 ): (H, H), (H, L),
(L, H) and (L, L). For the second candidate, the expected benefits of running in the four
cases are, respectively: V GH /2, V GH P , V GL (1−P ), and V GL /2. Assume that even the
largest of these, V GH P , is less than 1.8 Then, since C ∼ U [0, 1], in each case the expected
benefit is also equal to the probability the second candidate runs.
We want to study what happens in districts where the party is advantaged, districts
where it is disadvantaged, and districts that are balanced (so vigorous two-party competition
exists). Let s denote the general election advantage parties have in a constituency, and
assume that GH , GL and θ are functions of s, with dGH /ds > 0, dGL /ds > 0, and dθ/ds ≥ 0.
Let N be the expected number of candidates who run, let NH be the expected number
of high quality candidates who run, let FH be the expected fraction of candidates who run
who are high quality, and let WH be the probability that the primary winner is high quality.
Proposition 1. (i) dN/ds > 0 and dNH /ds > 0; (ii) the sign of dFH /ds and dWH /ds are
indeterminate; (iii) if dGH /ds > (dGL /ds)(1−P )/P , then dFH /ds > 0 and dWH /ds > 0.
To interpret the condition in part (iii) of the proposition, recall that P > 1/2 is the probability that a high quality candidate defeats a low quality candidate in a primary election.
Thus, the condition is definitely satisfied if dGH /ds ≥ dGL /ds. More generally, the condition
requires that as the party becomes stronger in a constituency, the probability of winning the
general election increases for both high and low quality nominees, and the probability does
not increase markedly more for low quality candidates than for high quality candidates. For
example, if P = 3/4, then the condition in part (iii) is dGH /ds > (dGL /ds)/3. Thus, it is
satisfied as long as the rate of increase in the probability of winning the general election for
high quality nominees is at least 1/3 as large as that for low quality nominees. This seems
plausible.
If part (iii) of the proposition holds, then the predictions are straightforward and can be
8
This clearly holds if V ≤ 1.
6
described most simply by the figures below. That is, as the party’s advantage in the general
election increases, the expected number of candidates, the expected number of high quality
candidates, the fraction of candidates who are high quality, and the probability that the
Parties Balanced
Advantaged
Disadvantaged
Parties Balanced
Advantaged
Disadvantaged
Disadvantaged
Parties Balanced
Advantaged
Parties Balanced
Advantaged
Fraction High Quality Cands
Number of High Quality Cands
Disadvantaged
Number of Candidates
Prob. Winner is High Quality
nominee is of high quality, all increase.9
Figure 1: Predictions of Strategic Candidate Entry Model
9
The models in Snyder and Ting (2011) and Evrenk et al. (2013) make similar predictions. Those models
go even further on one important dimension, since they model the strategic interaction between parties as well
as the behavior of candidates and voters within parties. The model in Castanheira, Crutzen and Sahuguet
(2010b) shows that primary elections induce candidates to improve the quality of the policy platforms they
offer when general elections reveal little information platform quality, or when the value of office is low.
They note that the second prediction can be interpreted as showing that, in a constituency where one party
is strong and the other is weak, primary elections induce the candidates in the stronger party to improve
the quality of their policy platforms. In another model, Castanheira, Crutzen and Sahuguet (2010a) show
that primary elections can improve the quality of the policy platforms that party-affiliated candidates offer
in order to defeat independent candidates.
7
2.2
Nominee Quality under Convention Systems
Do primaries nominate more qualified candidates than conventions? Developing a theoretical
model to capture the relationship between convention nominations and candidate quality is
not straightforward. Since convention nominations are often assumed to be under the control
of party elites, the predictions from these models will depend heavily on the assumptions
about the preferences of the party elites and the pool of potential candidates. If we simply
assume that party elites have lexicographic preferences for a candidate attribute that is
not positively correlated with quality and that the pool of high quality candidates does not
differ from the primary system, then a prediction that conventions would select lower quality
nominees than primaries could easily follow from a simple decision theoretic model like the
one discussed above.
However, there are a number of reason to be skeptical of the plausibility of such theoretical predictions. First, the composition of the pool of potential party nominees is likely to
be affected by the nomination system. This point was highlighted by those critical of primaries, who often argued that the cost of campaigning in a contested primaries would deter
high quality types from seeking any party nomination. The absence of these campaigning
costs under the convention system would presumably expand the pool candidates, including
high quality types, willing to accept a party nomination. With more options available to
party elites, they may be better able to select a high quality nominee. Alternatively, progressive reformers argued that primaries would motivate high quality types, who previously
were turned off by the alleged pathologies of the convention system, to seek their party’s
nomination.
Moreover, whether party elites actually have lexicographic preferences for non-quality
related candidate attributes and whether such preferences would have much of an impact
on the quality of the nominees is not entirely clear. Party elites could presumably have
strong incentives to select high quality nominees (if available) because of externalities across
districts, which they internalize, and because of the “bonus” they receive if their party wins
a majority (or more offices). If the pool of high quality types grows as discussed above,
8
then elites would be less constrained in recruiting high quality nominees who also have the
desired non-quality attributes – e.g., there may be a number of candidates who are both
high quality and loyal to the party. On the other hand, even if high quality types are
available, party elites may still have preferences for low quality nominees, if they believe
that these nominees have worse outside options which would make them even more loyal
to the party. So again, whether the overall quality of nominees is expected to increase or
decrease following a change in the nomination system is highly dependent upon which of the
different, but plausible, assumptions we are willing to make about elite preferences.
While theoretically modeling the relationship between nomination systems and the overall
quality of party nominees is particularly challenging, there are reasons to suspect that changing the nomination system would affect the relative likelihood of nominating high quality
candidates in safe versus competitive constituencies. In attempting to maximize the number of offices their party will win, party elites have an incentive to nominate high quality
candidates in constituencies where their party does not have an electoral advantage. This
could be achieved by subsidizing the general election campaigns in constituencies that are
competitive or where their party has a disadvantage in order to attract high quality types
who would not otherwise choose to enter these races.
Party leaders may also choose to subsidize the costs for high quality candidates to run
when the quality of a party’s nominee in one district could be perceived to affect voters’
decisions in other districts. These externalities might arise because voter utility depends
directly on the quality of all candidates who win office, not only the quality of the candidate
who wins in their district. (In a legislative context it seems especially likely that the average
quality of the majority party’s winners matters more than the average quality of the minority
party’s winners.) Externalities may also arise due to imperfect information. For example,
voters might not know the fraction of of each party’s candidates that are high quality (so,
observed quality in one district could affect voters’ beliefs everywhere). Or, voters might not
know how much value party leaders place on quality relative to loyalty (so, again, observed
quality in one district could affects voters’ beliefs everywhere).
9
Even in the absence of strategic behavior on the part of the party elites, the relationship between nominee quality and constituency safety could differ between conventions and
primaries due to differences in the costs of seeking the nomination. Since serious primary
campaigns are likely to be more costly than “throwing one’s hat in the ring” in a convention,
the strategic entry decisions, as modeled above, are probably more important considerations
under primary elections than under the convention systems.
3
Relevant Experience as a Qualification for Office
In the early debates over primaries, candidate quality was never clearly defined and often
encompassed several different attributes, such as capability or integrity. One of the challenges
in empirically investigating the effect of primaries is measuring candidate quality. We focus
on a particular dimension of quality, which is the ability of the candidate for a particular
office to perform the tasks required for that office. High quality candidates would have
some qualifications that makes them more likely to be better at performing these tasks for
the office they are seeking compared to candidates without these qualifications. Thus, our
empirical investigation will focus on the nomination of qualified candidates under primaries
and conventions.
Our measure of qualification for an office is whether a candidate has had prior experience
in positions that required her or him to perform similar tasks. A large body of literature in
economics and psychology examines the importance of prior experience and job performance.
In the labor economics literature, prior experience is often viewed as building human capital
and is used to explain wage differences (e.g. Becker (1964)). In the psychology literature, an
individual’s ability to perform new tasks are found to be related to their prior experiences.
Both literatures highlight the variation in types of prior experience and the variation in how
the different experiences are related to expected performance in new positions. In general,
the degree to which prior experiences are related to the tasks expected of workers has significant impact on their performance. Various studies find that prior experience that builds
human capital for particular industries, occupations or firms have different effects on worker
10
compensation. These differences are particularly salient when workers move to a new industry or firm (e.g., Neal (1995)). More recently, economists have focused on the implications of
task-specific human capital (e.g.Gibbons and Waldman (2004)). The personnel psychology
literature provides some empirical support to the notion that task-specific prior experience
is strongly related to performance (e.g., Quinones, Ford and Teachout (1995)).
Although this literature has tended to focus on the private sector, prior experience in
positions requiring similar tasks would presumably also affect the performance of elected
officials. Since the tasks required of public officials, including elected officials, vary across
offices, not all prior experiences are likely to have the same impact on performance for all
offices. The prior experiences relevant for being a member of Congress are likely to differ
from those relevant for being a state treasurer. For the former, prior experiences as a state
legislator, e.g. learning how to craft legislation, are likely to be more relevant than managing
funds for a local government. For the latter the reverse is likely to be the case. Candidates
are occasionally required to have some minimal qualification for the task required of certain
offices. For example, candidates for state attorney general are often required to be lawyers.
However, this is a minimal qualification and successful candidates often demonstrate their
qualifications with other relevant prior experiences, such as being a district attorney.10
There is some evidence that prior relevant experience may improve the performance of
elected officials. Various studies of legislators in recent years have found that legislative
effectiveness improves with seniority. In studying legislative effectiveness rankings in the
North Carolina state legislature, Padro-i Miguel and Snyder (2006) find that the performance
of legislators did increase during their tenure in office. In studying the U.S. House, Volden
and Wiseman (2014) find that the legislative effectiveness of House members, which is based
on their ability to enact their sponsored legislation, increases with seniority.
10
One noteable example where the attorney general did not have relevant prior experience was in North
Carolina. North Carolina Attorney General, Zeb V. Walser, who was nominated through a convention in
1896, apparently had no additional prior relevant experience other than being a lawyer and a member of
the state legislature. He was criticized during his term in office as being “incompetent” and being forced to
hire additional attorneys to represent the state. An article in The North Carolinian states, “If contradictory
decisions are what the people desire, then in Attorney-General Walser they have a man after their own heart;
but when it comes to performing the duties which were formerly attached to this position, the present official
is woefully deficient.” (The North Carolinian, September 9, 1897 p. 8)
11
For more direct evidence regarding the influence of prior relevant experience on future
performance, we compared the effectiveness of first year state senators in North Carolina who
had prior experience in the state House to those who did not have this experience.11 The
first year state senators with the prior relevant experience do have higher initial legislative
effectiveness scores as compared to those without the experience.12
Further evidence that prior relevant experience, at least in the modern period, may be
a reasonable measure of candidate qualification is that experience is a common part of
newspapers’ justifications for endorsing certain candidates. We have collected thousands
of primary election newspaper endorsements for the period 1990 to 2010.13 Presumably,
journalists and newspaper editors have more information about the attributes that are most
relevant for candidates seeking a particular office. In reading through these endorsements,
it was apparent that prior experience was a qualification used to assess future performance.
In races where at least two newspapers made endorsements during the primary, most of
the endorsements would go to candidates with prior experience. For example in open seat
primary U.S. House races where there are exactly two candidates and one has experience,
the experienced candidate receives the endorsement(s) 85% of the time. In cases with an
incumbent, the experienced candidate (i.e. the incumbent) receives the endorsement(s) 95%
of the time (233 U.S. House races).
Whether or not prior relevant experience actually affects future performance in office,
politicians tend to campaign as if this relationship exists. This is apparent even in candidates’
campaign advertisements soon after primaries were adopted. For example, in the 1910
Kansas primary for state treasurer, which was just two years after Kansas had adopted
mandatory primaries, the two term incumbent, Mark Tulley, announced his candidacy for
11
This is measure of effectiveness is described in detail in Padro-i Miguel and Snyder (2006) and also in
Chapter 9.
12
Volden and Wiseman (2014) find evidence that state legislators who come from states with more professionalized legislatures also tend to have higher legislative effectiveness in the U.S. House. However, House
members with prior state legislative experience, even in professionalized state legislatures, do not appear to
be more effective in their first year or two in Congress relative to other members of their incoming class. This
may have something to do with the particular LES measure, as the extreme right skew in the distribution
of scores may make it better for identifying unusually effective legislators rather than capturing the more
modest differences in ineffectiveness that would accrue from prior relevant office holding experience.
13
This data is discussed in greater detail in Chapter 9.
12
the position by stating: “I assure you that the experience is an advantage and that I feel
better qualified to discharge the duties of the office on account of the experience which I
have had in conducting the affairs.” Tulley’s challengers emphasized their relevant prior
experiences in related office in their announcements. F. J. Altswager highlighted his prior
experience by stating, “I am a candidate because I believe I am competent to fill the position.
My duties as county treasurer of Reno county for the past four years, and my eight years’
service to the State Board of Normal School regents have given me insight to public questions
which I feel would greatly aid me in serving the state in the capacity of State Treasurer.”14
In looking through numerous primary campaign advertisements, this is a common pattern
in primary election campaigns, particularly in races for statewide offices where office-specific
human capital may be relevant, such as auditor or superintendent of public instruction.
Finally, another indication that prior relevant experience may be related to expected future performance is that voters are more likely to support candidates with relevant prior
experience. In both the early period after primaries were first introduced and in recent
decades, primary voters are more likely to support candidates with relevant prior experience
than they are to support candidates with less relevant prior political experience. For example, in primary races for treasurer and auditor, primary voters are more likely to support
a candidate with prior experience managing local public funds than a candidate without
this experience. However, primary voters in treasurer and auditor races are not more likely
to support a candidate with prior legislative experience over a candidate without prior political experience. We discuss this evidence in more detail in the later chapters on voting
behavior. McDermott (2005) provides some additional evidence that relevant prior experience affected survey respondents’ willingness to support particular candidates in the 1994
California general election. Thus, irrespective of whether relevant prior experiences actually impact future performance in office, the primary voting behavior is consistent with an
electorate that believes prior experience matters.
We should note that our measure of relevant prior experience differs from the more com14
The Tulley and Altswager annoucements were reported in The Topeka Daily Capital January 30, 1910,
p4.
13
mon use of previous office holder experience as a measure of “candidate quality” in studies
of post-WWII U.S. elections – see e.g., Jacobson and Kernell (1983); Bond, Covington and
Fleisher (1985); Jacobson (1989, 2009); Lublin (1994); Bond, Fleisher and Talbert (1997);
Cox and Katz (1996); Van Dunk (1997); Goodliffe (2001, 2007). A few studies – e.g., Carson
and Roberts (2005, 2013) and Carson, Engstrom and Roberts (2007) – use previous office
holder experience as a measure of quality in the pre-war period. These papers are mainly
interested in estimating the impact of candidate attributes on election outcomes, thus, they
treat prior officeholder experience as a broad measure of “electability,” and include any
prior elected experience in constructing their measures. Aspects of prior experience with
less relevant offices may be a general indicator of political ability, but it may also capture
the ability to win office rather than a specific measure of how the candidate would perform
in a particular office. We restrict attention to a narrower set of “relevant” offices that are
more clearly associated with the accumulation of office-specific human capital.
3.0.1
Prior Experience as a Measure of Party “Insider” Status
Early skeptics of primary elections argued that the system favored “insiders” or professional
politicians. Party organizations could use their resource advantage to support their favored
candidate, which could potentially allow the party elite even more control over the nomination process than they had under the convention or caucus system. Thus, one concern
with using prior experience as a measure of office-specific human capital is that political
experience may also simply be an indication that the candidate is an “insider.”
One piece of evidence that prior relevant experience is not simply capturing “insiderness”
is the relationship between this measure and party loyalty. Since party insiders presumably follow the direction of the party leadership, party loyalty in roll call voting behavior
is arguably an indicator of being an insider. Thus, if relevant prior experience is mainly
capturing whether a candidate is an insider or not, then we would expect it to be highly
correlated with the party loyalty scores introduced in Chapter 4 – i.e. overall loyalty to party
leaders, overall loyalty to the party majority, loyalty to party leaders on progressive issues,
etc. We find no evidence that House representatives with prior state legislative experience
14
have significantly higher party loyalty scores during their first term in office. This is true
when we consider all members whose first term was during the period 1892 to 1928. It is
also true when we restrict attention to members nominated under the convention system or
members nominated by a direct primary. We also find no significant correlations when we
divide the sample by party and/or region.
The behavior of primary voters provides additional evidence that relevant prior experience
may be capturing beliefs about how qualified a candidate is for a particular office. As
briefly discussed above, primary voters appear to be distinguishing between relevant and
less relevant prior political experience. If party elites were mobilizing primary voters to
support “insiders” or if relevant prior experience is simply an indicator for electability, then
we would have expected the electoral advantage to be of similar magnitude for candidates
with either relevant or irrelevant prior political experience. In later chapters, we provide
evidence that primary voters do appear to offer more support to candidates with relevant
prior experience than those with less relevant experience.
4
Candidate Qualification in Early Primary Elections
To examine whether primaries during the early period were effective at nominating qualified
candidates, we examine the prior experience of primary candidates for the following offices:
U.S. House, U.S. Senate, governor, state treasurer and state auditor, during the period prior
to 1928. Each of the offices require different tasks, so the relevant prior experience that
would indicate some qualification should differ across the offices.
Since legislators are expected to develop skills in writing bills and shepherding these bills
through the committee system, compromising to build broader support for passing bills on
the floor, trading favors with fellow legislators, working with party leaders, etc., we use
prior experience as a state legislator or a member of Congress (although not currently the
incumbent) as indicators of having a qualification for U.S. House and Senate candidates. We
restrict attention to states for which we are able to construct complete lists of all individuals
15
who served in the state legislature over the period 1863 to 1928.15,16 We avoid relying on
biographical sources that do not have comprehensive coverage, since this can easily lead
to sample selection bias. Winning candidates are much more likely to leave a biographical
paper trail than losing candidates. Also, among the losers, those who were successful in
some other aspect of life – and are therefore likely to be “high quality” on dimensions other
than office-specific human capital – are more likely to leave biographical trail than others.
For gubernatorial candidates we assume that prior experience as an elected statewide
executive, e.g. secretary of state, treasurer, auditor or comptroller, attorney general, would
develop skills in running large bureaucracies, and expertise about specific areas of state
policy-making.17 These skills are especially relevant for serving in statewide executive offices,
including governor (the “chief bureaucrat” of state government), and less relevant for serving
in the U.S. senate. Thus, a gubernatorial candidate is considered qualified if she or he has
prior experience in an elected statewide executive position. We use a complete list of elected
statewide officials for the period 1880 to 1928.
Although it is somewhat simplistic, we treat state treasurers and auditors together as
“accountants and guardians of public funds.” Both offices require general familiarity with
bookkeeping, accounting, and public finances, as well as a high degree of fiduciary trust.
Our indicator of office-specific human capital is prior experience as a deputy or assistant
auditor or treasurer or any comparable local office, e.g. county or city auditor or treasurer.18
Successfully serving in these lower offices not only builds relevant human capital, but also
helps establish a reputation for being trustworthy. One concern with the analysis of these
offices is that we lack a comprehensive a list of the same prior relevant experience for all
15
The states we include are: Arkansas, Arizona, Colorado, Delaware, Florida, Idaho, Illinois, Indiana,
Iowa, Kansas, Massachusetts, Michigan, Minnesota, Missouri, Montana, Nebraska, Nevada, New Jersey,
New York, North Dakota, Ohio, Pennsylvania, South Dakota, Washington, West Virginia, Wisconsin.
16
We disregard service completed more than 30 years prior to an election. The data for California are
also available, but we do not include California because it permitted cross-filing, allowing candidates to run
in several parties’ primaries simultaneously. Many candidates won both the Democratic and Republican
primaries. This was the case in 6 out of 11 congressional races in 1920, and in 9 out of 11 races in 1922 and
1924. It is difficult to code party affiliations and seat safety in such cases.
17
We hope to eventually include prior experience as the mayor of a major U.S. city as another indicator
of relevant experience, but we have not yet gathered all of these data.
18
Ideally, we would include private sector experience in accounting and banking, but we do not have a
comprehensive source for this information so we do not include it.
16
primary candidates. We gathered information from newspapers, state manuals, and various
biographical sources. In the analyses below regarding the composition of the pool of candidates, we focus on those races where we have information about the prior experience of at
least three quarters of the candidates.
With these measures we can examine the extent to which the qualifications of candidates
and nominees in U.S. House and statewide office primaries match the theoretical expectations
in Figure 1. The figure has four panels each with the same horizontal axis: whether the primary is for a disadvantaged party, parties balanced or advantaged party. The four outcomes
of interest are: i) the percentage of the party nominees who have relevant prior experience;
ii) the average number of candidates; iii) the average number of candidates with relevant
prior experience; iv) the average proportion of candidates with relevant prior experience.
For U.S. House primaries, we find patterns of candidate qualification and constituency
partisanship that match the predictions from our model. In the top left panel of Figure 2 we
see that the probability that a qualified candidate wins the primary election increases from
about 0.10 in the disadvantaged party primary, to almost 0.30 in parties balanced primaries to
over 0.45 in the advantaged party primary. The remaining three panels show that increasing
the general election safety of a party not only increases the number of candidates entering
the party’s primary, but also that the proportion of candidates with prior state legislative
experience is higher in these primaries. The proportion of qualified candidates competing in
the primary almost triples, as we move from disadvantaged party primaries to primaries in
parties balanced districts to advantaged party primaries.
In Figure 3 we observe the same relationships between the qualification of primary candidates and primary type when we focus on races for statewide office – i.e. governor, U.S.
senator, treasurer and auditor. The patterns in this figure again show that advantaged
party primaries are more likely to nominate candidates with the relevant prior experience
than other primaries. The proportion of high quality candidates competing in a party’s
primary also increases as the party’s advantage in the general election increases.
These figures provide evidence consistent with the predictions from the theoretical model.
17
2.5
1.5
3
2.5
N
4
3.5
Number
Disadvantaged
Parties
Advantaged
umberBalanced
of Candidates
.1
1.5
Number of Candidates
2
2.5
3
3.5
Prob. Winner has Legislative Experience
.15 .2 .25 .3 .35 .4 .45
4
.45 Winner
.4
.35
.3
.25
.2
.15
.1
Prob.
Disadvantaged
Parties
Advantaged
Balanced
has Legislative Experience
Disadvantaged
Parties Balanced
Advantaged
Disadvantaged
Number of Candidates with Legislative Experience
.1
.3
.5
.7
Disadvantaged
Parties Balanced
Advantaged
25
20
15
10
Percentage
Disadvantaged
Parties
Advantaged
Balanced
of Candidates with Legislative Experience
Percentage of Candidates with Legislative Experience
10
15
20
25
.7
.5
.3
.1
Number
Disadvantaged
Parties
Advantaged
Balanced
of Candidates with Legislative Experience
Parties Balanced
Advantaged
Disadvantaged
Parties Balanced
Advantaged
Figure 2: Prior Offieholder Experience in U.S. House Open-Seat Primaries in the Early
Period
18
2.5
1.5
N
3
2.5
Number
Disadvantaged
Parties
Advantaged
umberBalanced
of Candidates
.1
1.5
2
Number of Candidates
2.5
3
Prob. Winner Has Relevant Experience
.2
.3
.4
.4
.3
.2
.1
Prob.
Disadvantaged
Parties
Advantaged
Winner
Balanced
Has Relevant Experience
Disadvantaged
Parties Balanced
Advantaged
Disadvantaged
Number of Candidates with Relevant Experience
.4
.6
.8
1
.2
Disadvantaged
Parties Balanced
Advantaged
35
30
25
20
15
10
Percentage
Disadvantaged
Parties
Advantaged
Balanced
of Candidates with Relevant Experience
Percentage of Candidates with Relevant Experience
10
15
20
25
30
35
NumberBalanced
1
.8
.6
.4
.2
Number
Disadvantaged
Parties
Advantaged
of Candidates with Relevant Experience
Parties Balanced
Advantaged
Disadvantaged
Parties Balanced
Advantaged
Figure 3: Prior Officeholder Experience in Statewide Open-Seat Primaries in the Early
Period
19
Primaries, even in the early period, have a greater impact on the qualification of candidates in
the advantaged party’s primary. Primaries appear to be contributing to the electoral system
where they are most needed – i.e. where, in the absence of primaries, candidates might be
elected without facing any serious electoral competition. However, to properly gauge the
contribution of primaries in nominating qualified candidates, we need to compare primary
nominations to nominations under a different system. In the next section we compare the
nomination of qualified candidates under primaries and conventions.
5
Qualified Candidates in Primaries vs Conventions
Did the introduction of primaries lead to more qualified candidates being nominated, particularly in the constituencies where party nominees could expect to face little general election
competition? As discussed above, developing a theoretical model to predict the overall effect of primaries on candidate qualifications is not straightforward and often depends on
the assumptions we are willing to make about about elite preferences. However, there are
more plausible theoretical reasons why we would expect advantaged party nominees to be
more qualified than disadvantaged party nominees under the primary as compared to the
convention system.
We examine the effect of moving from a convention to a primary system on the qualification of party nominees during the period 1892 to 1928. This period is roughly 10 years
prior to the first statewide adoption of mandatory primaries and 10 years after the initial
wave of progressive era adoptions of primaries. The main sample is the set of races with
no incumbent in the party’s primary. There might or might not be an incumbent in the
opposition party.19
19
One complication is that under the convention system, we only know that the candidate nominated
by the convention was not an incumbent. The incumbent might have “run” in the convention but lost.
In the primary period we exclude cases without an incumbent in the race. For the analysis of the U.S.
House, we analyzed the sample including cases where an incumbent ran in the primary and lost in order to
make the convention and primary samples more comparable. We also analyzed a sample that excludes cases
where House incumbents sought renomination but were not nominated by the convention. We used ICPSR
7428 Biographical Characteristics of Members of the United States Congress, 1789-1979 for information
about House incumbents who sought renomination under the convention system. In both of these additional
analyses the substantive findings are the same as in Table 1. This is perhaps not surprising since incumbents
20
We exploit the panel structure of the data, estimating models with both state-specific
and year-specific fixed effects, and even state-specific time trends. We denote district, which
is a congressional district in our analysis of the U.S. House and state in our analysis of the
statewide offices, by i and years by t. Experiencedijt is a dummy variable equal to 1 if party
j’s nominee in district i in year t has the relevant prior experience for the office they are
seeking. Primaryit is a dummy variable equal to 1 if nominations are made in a state that
has primary elections in year t. Let Safeijt be a trichotomous variable taking on the values
+1 if district i is “safe” for party j in year t (i.e. party j is the advantaged party), -1 if
district i is “safe” for the opposition party in year t (i.e. party j is the disadvantaged party),
and 0 if district i is “competitive” (safe for neither party). As above, we use a threshold of
15 percentage points to define safe districts.
The main specifications are of the form:
Experiencedijt = αk + θt + β1 Primaryit + β2 Safeijt + β3 Primaryit ×Safeijt + ijt
(1)
The αk terms are state fixed effects.20 The θt terms are year fixed effects. The state specific
fixed effects capture factors such as differences in the professionalization of the state legislatures. We might expect the degree to which prior relevant experience reflects a qualification
for a specific office to vary with the exact responsibilities of the various offices in the different
states – e.g. highly professionalized state legislatures may mean that prior experience is a
better reflection of the accumulation of legislative skills or the state auditor or treasurer
position in certain states may encompass additional responsibilities beyond what is expected
of local auditors and treasurers. The year effects capture the changes in these factors over
time. The state specific trends allow for differential career trajectories across states.21
losing renomination was relatively rare under both systems (95.4% of House incumbents running or reelection won their primaries and 95.9% of incumbents who sought renomination through the convention
were renominated). We do not focus on these results since we are unsure about sources for convention
system renomination outcomes.
20
In the analysis of statewide offices k = i.
21
We also estimated models that included Opp Pty Incumb and Primaryit ×Opp Pty Incumb as regressors,
where Opp Pty Incumbijt is a dummy variable equal to 1 if the opposition party’s nominee is the incumbent.
This specification allows us to test the hypothesis that the introduction of primary elections led many high
21
5.1
Primaries and Conventions for the U.S. House
For the reasons discussed above, prior experience in the state legislature or U.S. Congress
are indicators for having qualifications to be a U.S. House member.22 In our sample of House
races about half of the observations, 48.9%, are under the convention system. On average,
26.9% of non-incumbent nominees were Experienced under the convention systems, and
28.7% were Experienced under the primary system.23 The distribution of Safe is also similar
under the two systems. Under the convention system, 9.0% of the cases are in districts that
were safe for the party in question, 32.9% are in districts that were safe for the opposition
party, and 58.1% are in competitive districts. Under the primary system, the corresponding
figures are 10.3%, 40.6%, and 49.1%, respectively.24
A relationship between the qualification of party nominees and district safety is evident
in the basic summary statistics. In districts that were safe for the opposition party, a party’s
nominee was Experienced in 14.9% of the cases under the convention system and in 14.7%
of the cases under primaries – essentially the same. In competitive districts, the figures
are 31.1% and 35.1% – a bit higher under the primary system. In districts that are safe
for a party, however, the party’s nominee was Experienced in 43.3% of the cases under the
convention system and in 52.8% of the cases under the primary system – a large difference.
This appears to be the main impact of primaries: a dramatic increase in the proportion of
high quality nominees in parties’ safe districts.
To examine this relationship in a slightly more rigorous manner, we can estimate the
parameters in equations 5.1. The increasing proportion of experienced nominees in safe
districts is clearly evident in the results presented in the top panel of Table 1. The estimated
quality candidates to refuse to run against incumbents, (Carson and Roberts (2013)). The results are similar
to those reported in the table below, and the estimated coefficients on the two additional variables were
never statistically significant, so we do not report them separately.
22
We also include elected lieutenant governors, since they are typically heads of their state senates. The
only other positions that seem potentially comparable are city and county councilors. We do not include
these, however, because we do not have comprehensive information about this type of experience for losing
candidates.
23
We are keeping in cases that are uncontested and treating the opposition nominee as inexperienced. The
findings are not affected if we drop them.
24
Notice the share of cases that are safe is smaller than the share of cases that are unsafe because we are
focusing on cases without an incumbent. A disproportionate number of incumbents are in safe districts.
22
coefficient on the interaction term Primary ×Safe, β3 , is positive and statistically significant.
Thus, after a state adopted primary elections, in districts that were safe for a party, the
probability that the congressional nominee from that party was experienced increased by
0.070 in the simple differences-in-differences estimate. The increase is relatively large when
we consider that only 26.9% of our sample under the convention system had relevant prior
experience. Thus, the estimated change implies an increase of 26% over the mean under the
convention system.
Table 1: Primaries vs. Conventions and Relevant Experience, 1892–1928
U.S. House
(N = 5069)
Direct Primary × District Safety
Direct Primary
District Safety
Governor and U.S. Senate
0.045
(0.016)
0.029
(0.020)
0.150
(0.019)
0.070
(0.021)
0.064
(0.025)
0.119
(0.026)
0.062
(0.021)
0.054
(0.032)
0.122
(0.026)
0.068
(0.038)
0.063
(0.038)
0.141
(0.031)
0.076
(0.037)
0.102
(0.045)
0.131
(0.031)
(N = 1251)
Direct Primary × District Safety
Direct Primary
District Safety
0.033
(0.038)
0.142
(0.031)
0.142
(0.027)
State Auditor and Treasurer
(N = 680)
Direct Primary × District Safety
0.115
(0.059)
0.180
(0.047)
0.056
(0.046)
0.199
(0.052)
0.132
(0.069)
-0.004
(0.037)
0.179
(0.060)
0.104
(0.094)
-0.000
(0.038)
State and Year FE
No
Yes
Yes
State-Year Trends
No
No
Yes
Direct Primary
District Safety
Standard errors, clustered by state, are in parentheses.
23
The coefficient estimate on District Safety is also positive, which suggests that parties
were also more likely to have experienced nominees in safe districts even under the convention
system. This relationship is not unexpected given the differences in the pool of potential
candidates. More of the state legislators are likely to be affiliated with the advantaged party
than the disadvantaged party in safe districts. Thus, this result may simply reflect differences
in the pool of potential candidates in safe versus competitive districts.
The coefficient estimates on Primary are also positive, but with varying statistical significance. These estimates provides some suggestive evidence that the experience of the
nominees in competitive districts increased after the introduction of primaries. However,
as predicted in the theoretical considerations, the effect is less robust in the competitive
districts as compared to the safe districts.
5.2
Primaries and Conventions for Statewide Offices
We use the same differences-in-differences specification in equation 5.1 to analyze the effect
of changing nomination systems on the qualifications of candidates for statewide offices. We
again focus on govenor, U.S. senator, state treasurer and state auditor.
In our sample of gubernatorial and U.S. senate races, 49.5% of the observations are under
the convention system. On average, 21.2% of non-incumbent nominees were Experienced
under the convention systems, and 39.6% were Experienced under the primary system. The
distribution of Saf e is somewhat different under the two systems. Under the convention
system, 14.5% of the cases are in districts that were safe for party in question, 35.1% are in
districts that were safe for the opposition party, and 50.4% are in competitive districts. Under
the primary system, the corresponding figures are 33.2%, 28.3%, and 38.4%, respectively.
A relationship between prior office-holding experience of party nominees and district safety
is again evident in the basic summary statistics. In districts that were safe for the opposition
party, a party’s nominee was Experienced in 9.7% of the cases under the convention system
and in 18.4% of the cases under primaries – slightly higher under the primary system. In
competitive districts, the figures are 24.4% and 42.8% – again a bit higher under the primary
24
system. In districts that are safe for a party, however, the party’s nominee was Experienced
in 37.8% of the cases under the convention system and in 53.8% of the cases under the
primary system – a large difference. This appears to be the main impact of primaries: a
dramatic increase in the proportion of high quality nominees in parties’ safe districts.
Similar to the results for the U.S. House in the top panel of Table 1, the results in the
second panel of Table 1 show a significant change in the qualification of governors and U.S.
senators nominated in safe districts following the introduction of primaries. Governors and
senators nominated in advantaged party primaries were more likely to have relevant prior
experience as compared to the nominees under the convention system. The magnitude of this
interaction effect is larger than for the other offices examined. The coefficient on β3 is 0.068
in the simple differences-in-differences estimation. The increased probability of nominating
an Experienced candidate is relatively large when we consider that the average experience
level in our sample under the convention system is just 21.2%. Thus, the estimated change
implies an increase of 32% over the mean under the convention system.25
In our sample of state auditor and treasurer races, 54.0% of the observations are under
the convention system. On average, 25.4% of non-incumbent nominees were Experienced
under the convention systems, and 46.2% were Experienced under the primary system. The
distribution of Saf e is somewhat different under the two systems. Under the convention
system, 19.1% of the cases are in districts that were safe for party in question, 30.9% are in
districts that were safe for the opposition party, and 50.0% are in competitive districts. Under
the primary system, the corresponding figures are 29.5%, 19.6%, and 51.0%, respectively.
A relationship between the prior relevant experience of party nominees and district safety
is again evident in the basic summary statistics. In districts that were safe for the opposition
party, a party’s nominee was Experienced in 12.4% of the cases under the convention system
and in 21.3% of the cases under primaries – slightly higher under the primary system. In
25
In a separate analysis, we focus on the prior experience of only the general election winners. This
addresses some of the sample selection concerns with our inability to find the biographical information of the
candidates who lost the general election. Since there are few cases where an disadvantaged party candidate
wins the general election, we code District Safety as 1 if the winner is from the advantaged party in a safe
district and 0 otherwise. The coefficient of interest, β3 , is even larger in this analysis compared to the results
including all party nominees.
25
competitive districts, the figures are 36.1% and 49.1% – again a bit higher under the primary
system. In districts that are safe for a party, however, the party’s nominee was Experienced
in 18.6% of the cases under the convention system and in 57.6% of the cases under the
primary system – a large difference. This appears to be the main impact of primaries: a
dramatic increase in the proportion of high quality nominees in parties’ safe districts.
We again estimate equation 5.1 from above, but now focus on state auditor and treasurer
elections. Similar to the results for the U.S. House, governor and U.S. Senate in the top
two panels of Table 1, the results in the third panel of Table 1 show a significant change
in the qualification of state treasurer and auditor nominated in safe districts following the
introduction of primaries. State auditors and treasurers nominated in advantaged party
primaries were more likely to have relevant prior experience as compared to the nominees
under the convention system. The magnitude of this interaction effect is larger than for the
other offices examined. The coefficient on β3 is 0.199 in the simple differences-in-differences
estimation. The increased probability of nominating an Experienced candidate is relatively
large when we consider that the average experience level in our sample under the convention
system is just 25.4%. Thus, the estimated change implies an increase of 78% over the mean
under the convention system.26
5.2.1
Less Relevant Political Experience
As briefly discussed above, one concern with prior experience as an indicator for having
qualifications for office is that it may also be capturing some other characteristic of the
candidate, such as “insiderness”, political professionalism, name recognition, or campaigning
ability. We examine whether the introduction of primaries in safe districts also led to an
increase in the nomination of governors, senators, auditors and treasurers with prior political
26
In a separate analysis, we focus on the prior experience of only the general election winners. This
addresses some of the sample selection concerns with our inability to find the biographical information of the
candidates who lost the general election. Since there are few cases where an disadvantaged party candidate
wins the general election, we code District Safety as 1 if the winner is from the advantaged party in a safe
district and 0 otherwise. The coefficient of interest, β3 , is even larger in this analysis compared to the results
including all party nominees.
26
experience that is less relevant for these offices.27 For governors we consider experience as a
U.S. House member as being less relevant. For U.S. senators statewide executive experience
is considered less relevant experience. For state auditors and treasurers, state legislative
experience is considered less relevant experience. If primaries are mainly selecting politicians
with skills operating electoral campaigns or an advantage given to insiders, then we would
expect to see similar patterns when less relevant experience is the dependent variable.
Table 2: Primaries vs. Conventions and Less
Relevant Experience, 1892–1928
Dep. Var. = Experience
Governor and U.S. Senate
Direct Primary ×
District Safety
Direct Primary
District Safety
(N = 1143)
-0.132 -0.045
(0.051) (0.044)
-0.152 -0.007
(0.035) (0.037)
0.235
0.165
(0.043) (0.031)
State Auditor and Treasurer
Direct Primary ×
District Safety
Direct Primary
District Safety
-0.041
(0.049)
0.015
(0.059)
0.157
(0.033)
(N = 1302)
-0.086 -0.039
(0.036) (0.037)
-0.099 -0.014
(0.039) (0.033)
0.190
0.160
(0.027) (0.028)
-0.044
(0.040)
-0.029
(0.046)
0.165
(0.031)
State and Year FE
No
Yes
Yes
State-Year Trends
No
No
Yes
Standard errors, clustered by state, are in parentheses.
Table 2 presents our estimates of the coefficients in equation 5.1 with the dependent
variable now less relevant experience. The coefficient estimates provide no evidence that
parties were more likely to nominate former state legislators to be auditor or treasurer
candidates either in safe or competitive districts after the introduction of primary elections.
27
We could use other political experience that is not obviously relevant for auditors and treasurers, but
unfortunately we do not have comprehensive data of other prior offices for a large sample of candidates.
27
Perhaps not surprisingly, the districts that were safe under both the convention and primary
system were more likely to have nominees with prior state legislative experience. This most
likely reflects differences in the pool of candidates under these two systems. These findings
suggest that the increase in the prior relevant experience of advantaged party nominees after
the introduction of primaries was not simply reflecting the electoral advantage insiders or
professional politicians have under the new system.
6
Contribution of Primaries in the Early Period
Did primary elections contribute to the U.S. system in the early period? The analyses above
suggest that open-seat primaries are more likely to nominate a candidate with relevant prior
experience for the advantaged party in safe districts as compared to either party in competitive districts. In earlier chapters, we suggested that the lower quality of party nominees in
competitive districts is less of a concern since the general election will serve as an additional
screening mechanism. In this section we now evaluate the “bottom line,” and ask whether
the qualifications of the eventual general election winners for open seats differs between
competitive and uncompetitive districts. If primaries in the absence of competitive general
elections are as effective as having competitive general elections, then we would suspect that
the proportion of public officials with relevant prior experience to be about the same in safe
and competitive constituencies despite not having competitive general elections.
Table 3 shows the results. The top panel of the Table focuses on gubernatorial and U.S.
Senate races. The middle panel focuses on state treasurer and auditor races. The bottom
panel focuses on U.S. House races. In each panel the first row shows the percentage of general
election winners who have relevant prior experience in competitive and uncompetitive states
or districts. In the uncompetitive areas we only examine the advantaged party, since it wins
most of the offices in these areas. If primary and general elections are both effective in
selecting qualified candidates, then we should observe a similarly high percentage of general
election winners having relevant prior experience in both types of states or districts.
The second row in each panel shows the percentage of all candidates competing in primary
28
elections – i.e., the “pool” of candidates who run – who have relevant prior experience.28 In
the uncompetitive areas, the pool of candidates is only for the advantaged party primary,
since the winner comes from this group. In the competitive areas, the pool consists of
candidates from both parties’ primaries. If the electoral system as a whole selects for qualified
candidates, then the percentage of general election winners who are qualified should be
significantly higher than the percentage of qualified candidates in the overall pool.
Table 3: Quality of General Election Winners vs.
Candidates for Open Seats
Uncompetitive
State or District
All
Competitive
State or District
Governors and U.S. Senators, 1900-1948
Winners with
43.3%
39.3%
Relevant Experience
[224]
[117]
All Candidates with
34.2%
29.5%
Relevant Experience
[892]
[713]
Auditors and Treasurers, 1900-1948
Winners with
64.8%
54.4%
Relevant Experience
[128]
[136]
All Candidates with
56.6%
49.7%
Relevant Experience
[242]
[352]
U.S. House Representatives, 1900-1948
Winners with
47.3%
38.6%
Legislative Experience
[241]
[254]
All Candidates with
38.9%
32.4%
Legislative Experience
[602]
[870]
Cell entries give the percentage of candidates of each type. Number of
observations in brackets.
There are two notable patterns in Table 3. First, in all three cases the percentage of general
elections winners who have relevant prior experience is higher in uncompetitive as compared
to competitive states and districts. This pattern is what we would expect if primaries have
an important role in selecting qualified candidates in uncompetitive states and districts. If
general elections were the only truly effective mechanism for selecting qualified candidates,
28
We do not include token candidates who win less than 1% of the vote.
29
then the percentage of qualified winners would be much higher in the competitive areas than
in the uncompetitive areas.
Second, there is a higher percentage of general election winners with relevant prior experience as compared to the pool of candidates. This suggests that primary elections are
doing something other than simply selecting a random draw from the pool. However, the
difference between the proportion of qualified candidates among the winners versus the pool
of candidates is not particularly overwhelming. This suggests the contributions of primaries
may largely occur through the strategic entry decisions which affect the composition of the
pool of potential candidates. The next chapter will focus on voting behavior in primaries
to examine whether the primary electorate was voting for candidates with relevant prior
experience during this earlier period.
7
Discussion
Two ways in which primaries are expected to contribute to the U.S. political system are: 1)
Primaries provide an opportunity for voters to weigh non-partisan or ideological attributes of
the candidates, such as their qualifications for the office; 2) Primaries introduce competition
in constituencies where the partisan loyalties make one party safe in the general election. By
contributing in these ways, primaries should facilitate the election of qualified public officials
even in the absense of general election competition. This chapter provides evidence that
primaries have been contributing to the system in the ways we had expected, even in the
years immediately after the system was introduced. Moreover, advantaged parties in safe
constituencies were even more likely to nominate a qualified candidate under the primary
system than the convention system.
One potential concern with the above analyses is the use of relevant prior experience to
indicate candidate qualification. In later chapters, which focuses on primary competition
in more recent years, we introduce two alternative indicators of candidate qualification to
provide further validation that primaries are contributing to the political system where they
are most needed. Thus, we will return to this discussion of measuring candidate qualification
30
in those later chapters.
In focusing on the behavior of elites, this chapter also does not provide much direct
evidence that primary voters have much of a role in the selection of qualified public officials.
How informed were primary voters during the first years after the reform was introduced? Did
primary voters in this early period identify and support the candidates with prior relevant
experience? Did they vote along dimensions other than candidate qualifications? We turn
to these questions regarding primary voting behavior during the first decades of the 20th
century in the next chapter.
31
8
Appendix
Proof of Proposition 1. Denoting cases where the second candidate does not run as
∅, there are six possible primary election scenarios. The scenarios, together with their
probabilities of occurring, are:
Scenario
Probability scenario occurs
(H, H)
(H, L)
(H, ∅)
(L, H)
(L, L)
(L, ∅)
θ2 V GH /2
θ(1−θ)V GL (1−P )
θ[1 − θV GH /2 − (1−θ)V GL (1−P )]
θ(1−θ)V GH P
(1−θ)2 V GL /2
(1−θ)[1 − θV GH P − (1−θ)V GL /2]
Proof of (i). The expected number of candidates is N = 1 + Prob(N = 2). Denote
Prob(N = 2) by N2 . Then N2 = θ2 V GH /2+θ(1−θ)V GL (1−P )+θ(1−θ)V GH P +(1−θ)2 V GL /2.
Thus, dN/ds = dN2 /ds = (∂N2 /∂GH )(dGH /ds) + (∂N2 /∂GL )(dGL /ds) + (∂N2 /∂θ)(dθ/ds).
Differentiating, ∂N2 /∂GH = V [θ2 /2 + θ(1−θ)P ] > 0 and ∂N2 /∂GL = V [θ(1−θ)(1−P ) +
(1−θ)2 /2] > 0. Also, after canceling some terms and rearranging, ∂N2 /∂θ can be written
∂N2 /∂θ = V (GH −GL )[θ(1−P ) + (1−θ)P ] > 0. Thus, dN/ds > 0.
The expected number of high quality candidates is NH = 2Prob(H, H) + Prob(H, L) +
Prob(H, ∅) + Prob(L, H). This can be written NH = θ + θV GH [θ/2 + (1 − θ)P ]. Thus,
dNH /ds = (∂NH /∂GH )(dGH /ds)+(∂NH /∂θ)(dθ/ds). Differentiating, ∂NH /∂GH = θV [θ/2+
(1 − θ)P ] > 0. Also, after canceling some terms and rearranging, ∂N2 /∂θ can be written
∂NH /∂θ = 1 + V GH [θ(1−P ) + (1−θ)P ]. Thus, dNH /ds > 0.
Proof of (ii). The fraction of high quality candidates is FH = Prob(H, H) + Prob(H, ∅) +
(Prob(H, L)+Prob(L, H))/2. This can be written FH = θ+θ(1−θ)V [GH P −GL (1−P )]. Thus,
dFH /ds = (∂FH /∂GH )(dGH /ds)+(∂FH /∂GL )(dGL /ds)+(∂FH /∂θ)(dθ/ds). Differentiating,
∂FH /∂GH = θ(1−θ)V P > 0 and ∂FH /∂GL = −θ(1−θ)V (1−P ) < 0. Thus, if dθ/ds ≈ 0,
dGL /ds ≈ 0 and dGH /ds > 0 is “large,” then dFH /ds > 0, while if dθ/ds ≈ 0, dGH /ds ≈ 0
and dGL /ds > 0 is “large”, then dFH /ds < 0. So, the sign of dFH /ds is indeterminate.
The probability that the primary winner is high quality is WH = Prob(H, H)+Prob(H, ∅)+
(Prob(H, L) + Prob(L, H))P . This can be written WH = θ + θ(1−θ)V [GH P 2 − GL (1−P )2 ].
Thus, dWH /ds = (∂WH /∂GH )(dGH /ds) + (∂WH /∂GL )(dGL /ds) + (∂WH /∂θ)(dθ/ds). Differentiating, ∂WH /∂GH = θ(1−θ)V P 2 > 0 and ∂WH /∂GL = −θ(1−θ)V (1−P )2 < 0. Thus,
if dθ/ds ≈ 0, dGL /ds ≈ 0 and dGH /ds > 0 is “large,” then dWH /ds > 0, while if dθ/ds ≈ 0,
dGH /ds ≈ 0 and dGL /ds > 0 is “large”, then dWH /ds < 0. So, the sign of dWH /ds is
indeterminate.
Proof of (iii). Recall that ∂FH /∂GH = θ(1−θ)V P > 0 and ∂FH /∂GL = −θ(1−θ)V (1−P ) <
0. Thus, the first two terms of dFH /ds can can be written as θ(1−θ)V [P (dGH /ds) − (1−
P )(dGL /ds)]. This sum is strictly positive if dGH /ds > (dGL /ds)(1−P )/P . Regarding the
32
third term of dFH /ds, differentiate to obtain ∂FH /∂θ = 1+(1−2θ)V [GH P −GL (1−P )]. This
is decreasing in θ, so set θ = 1 (the largest possible value). Then ∂FH /∂θ = 1 − V GH P +
V GL (1−P ). This is positive, since V GH P < 1 (see assumption above regarding the most
favorable expected benefit of running). Thus, ∂FH /∂θ > 0. Thus, dFH /ds > 0.
Finally, recall that ∂WH /∂GH = θ(1−θ)V P 2 > 0 and ∂WH /∂GL = −θ(1−θ)V (1−P )2 < 0.
Thus, the first two terms of dWH /ds can can be written as θ(1 − θ)V [P 2 (dGH /ds) − (1 −
P )2 (dGL /ds)]. Since (1 − P )2 /P 2 < (1 − P )/P , this sum is strictly positive if dGH /ds >
(dGL /ds)(1−P )/P . Regarding the third term of dWH /ds, differentiate to obtain ∂WH /∂θ =
1 + (1−2θ)V [GH P 2 − GL (1−P )2 ]. This is decreasing in θ, so set θ = 1 (the largest possible
value). Then ∂WH /∂θ = 1 − V GH P 2 + V GL (1 − P )2 . This is positive, since V GH P < 1
(see assumption above regarding the most favorable expected benefit of running). Thus,
∂WH /∂θ > 0. Thus, dWH /ds > 0.
33
Table 4: Primaries vs. Conventions and Relevant Experience, 1892–1928, Using Different Measures of Safety
Dep. Var. = Experience
Governor and U.S. Senate
(N = 1192)
Direct Primary × District Safety
Direct Primary
District Safety
0.028
(0.039)
0.128
(0.031)
0.147
(0.028)
0.068
(0.040)
0.046
(0.040)
0.143
(0.033)
0.079
(0.039)
0.074
(0.050)
0.133
(0.032)
State Auditor and Treasurer
(N = 669)
Direct Primary × District Safety
0.117
(0.063)
0.189
(0.048)
0.068
(0.047)
0.197
(0.065)
0.145
(0.072)
0.010
(0.039)
0.174
(0.072)
0.118
(0.100)
0.014
(0.039)
State and Year FE
No
Yes
Yes
State-Year Trends
No
No
Yes
Direct Primary
District Safety
Standard errors, clustered by state, are in parentheses.
34
References
Aragon, Fernando M. 2014. “Why do parties use primaries? Political selection versus candidate incentives.” Public Choice 160:205–225.
Becker, Gary S. 1964. Human capital: a theoretical analysis with special reference to education. New York: Columbia University Press.
Bond, Jon R., Cary Covington and Richard Fleisher. 1985. “Explaining Challenger Quality
in Congressional Elections.” The Journal of Politics 41:510–29.
Bond, Jon R., Richard Fleisher and Jeffrey C. Talbert. 1997. “Partisan Differences in Candidate Quality in Open Seat House Race, 1976-1994.” Political Research Quarterly 50:281–
300.
Boots, Ralph Simpson. 1917. The Direct Primary in New Jersey. Columbia university.
Carey, John and John Polga-Hecimovich. 2006. “Primary Elections and Candidate Strength
in Latin America.” The Journal of Politics 68(3):530–543.
Carson, Jamie L., Erik J. Engstrom and Jason M. Roberts. 2007. “Candidate Quality,
the Personal Vote, and the Incumbency Advantage in Congress.” The American Political
Science Review 101(2):289–301.
Carson, Jamie L. and Jason M. Roberts. 2005. “Strategic politicians and U.S. House elections, 1874-1914.” The Journal of Politics 67(2):474–496.
Carson, Jamie L and Jason M. Roberts. 2013. Ambition, Competition, and Electoral Reform:
The Politics of Congressional Elections Across Time. University of Michigan Press.
Castanheira, Micael, Benoit Crutzen and Nicolas Sahuguet. 2010a. “The Impact of Party
Organization on Electoral Outcomes.” Revue Economique 61(4):677–696.
Castanheira, Micael, Benoit Crutzen and Nicolas Sahuguet. 2010b. “Party Organization and
Electoral Competition.” Journal of Law, Economics and Organization 26(1):212–242.
Cox, Gary W. and Jonathan N. Katz. 1996. “Why Did the Incumbency Advantage in the
U.S. House Elections Grow?” American Journal of Political Science 40(2):478–497.
Gibbons, Robert and Michael Waldman. 2004. “Task-Specific Human Capital.” American
Economic Review: Papers and Proceedings 94(2):203–207.
Goodliffe, Jay. 2001. “The Effect of War Chests on Challenger Entry in U.S. House Elections.” American Journal of Political Science 45:830–44.
Goodliffe, Jay. 2007. “Campaign War Chests and Challenger Quality in Senate Elections.”
Legislative Studies Quarterly 32(1):135–56.
Jacobson, Gary C. 1989. “Strategic Politicians and the Dynamics of U.S. House elections,
1946-86.” The American Political Science Review 83(3):773–793.
35
Jacobson, Gary C. 2009. The Politics of Congressional Elections. 7th ed. New York: Longman.
Jacobson, Gary C. and Samuel Kernell. 1983. Strategy and Choice in Congressional Elections.
New Haven: Yale University Press.
Kemahlioglu, Ozke, Rebecca Weitz-Shapiro and Shigeo Hirano. 2009. “Why Primaries in
Latin American Presidential Elections?” The Journal of Politics 71(1):339–352.
Key, V.O. 1956. American State Politics: An Introduction. New York: Alfred A. Knopf.
Lublin, David. 1994. “Quality, Not Quantity: Strategic Politicians in U.S. Senate Elections,
1952-1990.” The Journal of Politics 56:228–41.
McDermott, Monika L. 2005. “Candidate Occupations and Voter Information Shortcuts.”
The Journal of Politics 67(1):201–219.
Munro, William Bennett. 1919. The Government of the United States: National, State, and
Local. New York: Macmillan Company.
Neal, Derek. 1995. “Industry-Specific Human Capital: Evidence from Displaced Workers.”
Journal of Labor Economics 13(4):653–77.
Padro-i Miguel, Gerard and James M. Snyder, Jr. 2006. “Legislative Effectiveness and
Legislative Life.” Legislative Studies Quartely 31:347–391.
Quinones, Miguel A, J Kevin Ford and Mark S Teachout. 1995. “The relationship between
work experience and job performance: a conceptual and meta-analytic review.” Personnel
Psychology 48(4):887–910.
Snyder, Jr., James M. and Michael M. Ting. 2011. “Electoral Selection with Parties and
Primaries.” American Journal of Political Science 55(4):782–796.
Sorauf, Frank J. 1963. Party and Representation: Legislative Politics in Pennsylvania. New
York: Atherton Press.
Van Dunk, Emily. 1997. “Challenger Quality in State Legislative Elections.” Political Research Quarterly 50(4):793–807.
Volden, Craig and Alan E. Wiseman. 2014. Legislative Effectiveness in the United States
Congress: The Lawmakers. New York: Cambridge University Press.
36