Top

Public Choice

Published in:

25-10-2018

Candidate competition and voter learning in the 2000–2012 US presidential primaries

Authors: George Deltas, Mattias K. Polborn

Published in: Public Choice | Issue 1-2/2019

Activate our intelligent search to find suitable subject content or patents.

search-config

AI-assisted search

Off

Abstract

When candidates in primary elections are ideologically differentiated (e.g., conservatives and moderates in the Republican Party), then candidates with similar positions affect each others’ vote shares more strongly than candidates with different ideological positions. We measure this effect in US presidential primaries and show that it is of first-order importance. We also show that voters’ beliefs about the candidates harden over the course of the primary, as manifested in the variability of candidate vote shares. We discuss models of sequential voting that cannot yield that pattern of results, and propose an explanation based on a model with horizontally and vertically differentiated candidates and incompletely informed voters. Consistent with the predictions of this model, we also show that, in more conservative states, low-quality conservative candidates do better relative to high-quality conservatives, and vice versa.

previous article Titles for me but not for thee: transitional gains trap of property rights extension in Colombia

next article Regulation and government debt

Dont have a licence yet? Then find out more about our products and how to get one now:

Springer Professional "Wirtschaft+Technik"

Online-Abonnement

Mit Springer Professional "Wirtschaft+Technik" erhalten Sie Zugriff auf:

über 102.000 Bücher
über 537 Zeitschriften

aus folgenden Fachgebieten:

Automobil + Motoren
Bauwesen + Immobilien
Business IT + Informatik
Elektrotechnik + Elektronik
Energie + Nachhaltigkeit
Finance + Banking
Management + Führung
Marketing + Vertrieb
Maschinenbau + Werkstoffe
Versicherung + Risiko

Jetzt Wissensvorsprung sichern!

inform now

Springer Professional "Wirtschaft"

Online-Abonnement

Mit Springer Professional "Wirtschaft" erhalten Sie Zugriff auf:

über 67.000 Bücher
über 340 Zeitschriften

aus folgenden Fachgebieten:

Bauwesen + Immobilien
Business IT + Informatik
Finance + Banking
Management + Führung
Marketing + Vertrieb
Versicherung + Risiko

Jetzt Wissensvorsprung sichern!

inform now

Available only for authorised users

We exclude George W. Bush’s and Barack Obama’s essentially unopposed renominations in 2004 and 2012.

See also Bartels (1985, 1988) and Kenny and Rice (1994), which all focus on two-candidate settings.

Among theoretical static models of primaries (i.e., those where only one vote is taken at the primary stage), our paper is most related to Adams and Merrill (2008) and Serra (2011), which point out that holding primaries allows a party to select, on average, higher quality candidates than with a direct nomination of a candidate by party insiders.

See, e.g., http://www.cnn.com/ELECTION/2000/primaries/NH/poll.rep.html, http://www.cnn.com/ELECTION/2000/primaries/SC/poll.rep.html, http://www.cnn.com/ELECTION/2000/primaries/IA/poll.html. In the 2000 Republican primary, we also identify Steve Forbes and Alan Keyes as conservatives, as they also do better with self-identified conservative voters.

See, e.g., http://www.cnn.com/2008/POLITICS/02/05/super.exit/.

The average ideological self-placement of 2012 Romney primary voters in the American National Election Survey on a seven-point scale is 5.13 (5 is “slightly conservative”, 6 is “conservative”). In contrast, the average self-placement of Santorum voters is 5.57, and 5.56 for Gingrich voters. Unfortunately, previous waves of the NES did not ask respondents about their votes in the presidential primary.

The 2008 Democratic primary is not an aberration in this respect—“liberal” and “moderate” candidates (where this classification is based on roll-call votes or expert judgments of their positions) often receive rather similar percentages of their votes from liberal and moderate voters.

For example, see http://www.cnn.com/ELECTION/2004/primaries/pages/epolls/IA/index.html.

The 1992 general election results were obtained from Dave Leip’s Atlas of U.S. Presidential Elections, available at http://www.uselectionatlas.org/.

Of course, \(\kappa \cdot \mathrm{VoteShare}^{\kappa ,\kappa '}+\kappa '\cdot \mathrm{VoteShare}^{\kappa ',\kappa }=100\) holds as an identity. Deviations from that outcome in Table 1, such as here where \(3\times 17.3\%+48.2\%=100.1\%\), are explained by rounding.

The precise implications of the theory are for expected vote share comparisons between \(\kappa \) candidates at one position and \(\kappa '\) candidates at the other, versus \(\kappa -1\) at one position and \(\kappa '\) at the other. But comparisons between \(\kappa \) candidates at one position and \(\kappa '\) at the other versus \(\kappa -1\) at one position and \(\kappa '+1\) at the other can be obtained by applying our theoretical result iteratively.

McCain in the 2000 Delaware primary. The three remaining candidates (Bush, Forbes, Keyes) share the remaining 75% and thus have, on average, a 25% vote share as well.

However, we utilize for the purpose of inference the minimal information that candidate shares in a party’s state primary in a given year are negatively correlated (even conditional on characteristics) and that candidate-specific information available at a given time is correlated across states. That is accomplished by using White’s (1980) heteroscedasticity consistent standard errors with a two-way clustering at the state primary and candidate/round levels. Doing so tends to be conservative for the purpose of testing (i.e., ignoring clustering reduces standard errors).

Note that \(\mathrm{Conservative}_{j}\) is zero for all Democrats and \(\mathrm{Outsider}_{j}\) is zero for all Republicans, i.e., those variables include an implicit interaction with the party dummy.

Adding location dummies to this regression does not materially affect the estimates of the Democratic coefficients, but reduces the Republican location effect to essentially zero. However, with \(\mathrm{Moderate}_{i}\) being essentially a dummy for McCain, the Republican effect would be identified solely from the gain of voters by McCain as other candidates (of opposing location) depart, relative to the gain of voters by his opponents as other candidates (at same location) depart. Not only is the effective information for this specification even more limited (only four such withdrawals), but with McCain being a higher quality candidate, the location and valence effects are confounded (McCain gets a larger than expected share of the departing candidates’ voters because he is a better candidate in the vertical dimension).

The vote share variables \(Bush92\%_{s}\), \(Clinton92\%_{s}\) and \(Perot92\%_{s}\), like \(\mathrm{VoteShare}_{j,s,y}\) range from 0 to 100.

The use of exhaustive \(\text{candidate}\times \text{year}\times \text{round}\) effects eliminates the need to add party dummies in these regressions.

The Republican estimates are identical in the two regressions, as they should be since the exhaustive set of fixed effects essentially results in two distinct equations for each of the two parties. The results of Eqs. (5) and (6) are omitted from Table 2 to conserve space, but are available upon request.

The theoretical model shows that the effect of changes in electorate preferences on candidate vote shares depends not only on the candidate’s valence and political position, but also on the number of competing candidates, their valence, and their political positions. The variable \(MeanShr_{j,t,y}\) also adjusts for the number of competing candidates, their valence and political positions and, thus, in a qualitative way, reflects the factors that enter in the comparative statics of the theoretical model.

Note that, by necessity, this specification uses a different proxy for every state, since the variable MeanShr no longer takes the same value for all states in a given round.

To see this, suppose that only two states, 1 and 2, hold primaries in a given round t, and that the mean share enters directly as a regressor (rather than as an interaction). Then, the share of candidate j in state 1 is \(VS_{j,1}=BX_{j,1}+\gamma VS_{j,2}+\epsilon _{j,1}\), where BX contains all other regressors and the year subscript is suppressed. The vote share for state 2 is given analogously. Solving the two-by-two system for \(VS_{j,1}\) and \(VS_{j,2}\) yields \(VS_{j,1}=\frac{BX_{j,1}}{1-\gamma }+\frac{\epsilon _{j,1}+\gamma \epsilon _{j,2}}{1-\gamma ^{2}}\) and \(VS_{j,2}=\frac{BX_{j,2}}{1-\gamma }+\frac{\epsilon _{j,2}+\gamma \epsilon _{j,1}}{1-\gamma ^{2}}\). It can be seen from the reduced-form expressions for the vote shares that the share in state 2 is positively correlated with the structural error of the equation for the vote share in state 1 (\(\gamma <1\)).

The conclusions about relative effects are based on the parameters reported in Table 2. To obtain the level effect, one needs also to incorporate the effect of the Bush 1992 percentage on Republican vote shares, which is 1.38 (not reported in Table 2 to save space). Evaluated at a value of \(\mathrm{MeanShr}=0.4\), a 1% increase in Bush’s 1992 vote share lifts the vote share of a conservative Republican by \(0.77\%\) and reduces the vote share of the typical moderate Republican by \(0.47\%\) (the two figures are not the same because the number and typical vote shares of conservatives and moderates are not equal).

The residuals of Model 5 give similar results.

The adjustment is exact when no covariates are entered.

It is not surprising, and in fact reassuring, that when one includes the variable \(PriorSignals_{j,s,y}\) in the vote share regressions it turns comes out to be insignificant.

In fact, it is larger in absolute value than those reported in Table 3 because on average multiple state contests occur in a single round.

It is not appropriate to do so for the less parametrized models 1, 2, and 3 because the skedastic function almost surely also depends on other factors that dominate learning effects. However, we did verify that the conclusions of Table 2 are not affected when those models were estimated with GLS.

The idea is that voters in the same state receive their news about the candidates from the same local news sources so that the errors are not individual-specific. To simplify the model and gain some tractability, we ignore nationally observed errors, although they are accounted for in the estimation.

Since we focus on the implications of voters’ learning and preferences for vote shares, the specific rules for who wins the nomination do not matter; therefore, we are silent on that issue.

Adams, J., & Merrill, S. (2008). Candidate and party strategies in two-stage elections beginning with a primary. American Journal of Political Science, 52(2), 344–359.CrossRef

Bartels, L. M. (1985). Expectations and preferences in presidential nominating campaigns. The American Political Science Review, 79(3), 804–815.CrossRef

Bartels, L. M. (1987). Candidate choice and the dynamics of the presidential nominating process. American Journal of Political Science, 31(1), 1–30.CrossRef

Bartels, L. M. (1988). Presidential primaries and the dynamics of public choice. Princeton: Princeton University Press.

Callander, S. (2007). Bandwagons and momentum in sequential voting. Review of Economic Studies, 74, 653–684.CrossRef

Dekel, E., & Piccione, M. (2000). Sequential voting procedures in symmetric binary elections. Journal of Political Economy, 108(1), 34–55.CrossRef

Deltas, G., Herrera, H., & Polborn, M. K. (2016). Learning and coordination in the presidential primary system. Review of Economic Studies, 83(4), 1544–1578.CrossRef

Grossmann, M., & Hopkins, D. A. (2016). Asymmetric politics: Ideological Republicans and group Interest Democrats. Oxford: Oxford University Press.CrossRef

Kenny, P. J., & Rice, T. W. (1994). The psychology of political momentum. Political Research Quarterly, 47(4), 923–938.CrossRef

Klumpp, T., & Polborn, M. K. (2006). Primaries and the New Hampshire effect. Journal of Public Economics, 90, 1073–1114.CrossRef

Knight, B., & Hummel, P. (2015). Sequential or simultaneous elections? An empirical welfare analysis. International Economic Review, 56, 851–887.CrossRef

Knight, B., & Schiff, N. (2010). Momentum and social learning in presidential primaries. Journal of Political Economy, 118, 1110–1150.CrossRef

Krasa, S., & Polborn, M. K. (2010). The binary policy model. Journal of Economic Theory, 145(2), 661–688.CrossRef

Poole, K., & Rosenthal, H. (1985). A spatial model for legislative roll call analysis. American Journal of Political Science, 29(2), 357–384.CrossRef

Schwabe, R. (2015). Super Tuesday: Campaign finance and the dynamics of sequential elections. Social Choice and Welfare, 44(4), 927–951.CrossRef

Serra, G. (2011). Why primaries? The party’s tradeoff between policy and valence. Journal of Theoretical Politics, 23(1), 21–51.CrossRef

White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48(3), 817–838.CrossRef

Title: Candidate competition and voter learning in the 2000–2012 US presidential primaries
Authors: George Deltas
Mattias K. Polborn
Publication date: 25-10-2018
Publisher: Springer US
Published in: Public Choice / Issue 1-2/2019
Print ISSN: 0048-5829
Electronic ISSN: 1573-7101
DOI: https://doi.org/10.1007/s11127-018-0620-7

Springer Professional

Abstract

Please log in to get access to your license.

Dont have a licence yet? Then find out more about our products and how to get one now:

Springer Professional "Wirtschaft+Technik"

Springer Professional "Wirtschaft"

Other articles of this Issue 1-2/2019

Regulation and government debt

Editorial Announcement

Land lotteries, long-term wealth, and political selection

A banana republic? The effects of inconsistencies in the counting of votes on voting behavior

How robust is the welfare state when facing open borders? An evolutionary game-theoretic model

Evolving hierarchical preferences and behavioral economic policies