New article about choosing political candidates with Simon Calmar Andersen in British Journal of Political Science
How do we choose between alternatives? In our new article, we introduce lexicographic preferences as a possible way to understand citizens’ choices of political candidates. We show that the mixed evidence on the influence of candidate gender and race might be explained by voters having lexicographic preferences, treating the party affiliation of the candidate as a first-order preference. Only when party affiliation is satisfied, for example when the choice of candidate is within parties like in a presidential primary, can other, second-order, preferences play a larger role. We show that conjoint experiments systematically delivers a way to analyze lexicographic preferences, and how gender and race can play a role when candidate party affiliation is constant.
In political science and many other disciplines, classical choice theory (e.g., ordinal and cardinal utility functions and substitutability between alternatives) is king. Here, the decision-maker takes all (or most) attributes of the decision-object into account and he/she can trade-off between them. In practice, this means that in the choice of a political candidate, for example, candidate attributes like policy preferences, party affiliation, gender, race, experience, job, age etc. are considered and can be substituted between each other. If the choice is between two candidates, say in a presidential election, the first candidate might not have enough political experience in your view but instead have your prefered job experience outside politics. For the other candidate, this may be the other way around. If you substitute between alternatives, you might end up thinking these candidates are equally preferable.
Not so with lexicographic preferences. These allow a ranking of preferences for attributes, so one attribute is above all other, and you seek to maximize that one no matter what the levels on other attributes might be. In the example with presidential candidates, if you have lexicographic preferences for political experience, you would always choose the candidate with most experience – no matter the difference to the other candidate on this attribute, and no matter the levels of other atttributes like job, age, gender, race etc. With lexicographic preferences, you are unwilling to trade or accept compensation for that first-order prefence, and no indifference curve can be drawn. These are strict lexicographic preferences, and they probably rarely occur in real decision situations. A modified version of lexicographic preferences accepts minimum levels that when satisfied, decision-makers can move on to consider other attributes1.
In the article, we specifically look at the candidate attributes of gender and race, which are two attributes that are also publicly debated when candidates for the US presidency are nominated2. In the candidate literature, there is mixed evidence on the influence of these attributes. Our take is that the possibility of hierarchical decision-making has been overlooked in this debate, especially in the US, where a large political polarization has happened over the last decades3. Party affiliation is likely a first-order preference for many voters, and therefore other attributes like gender and race are first considered when choosing between two candidates from the same party. This is, we believe, the first contribution of our article.
The second contribution is methodological. Lexicographic preferences are hard to measure4. We propose to use conjoint experiments that rutinely randomize different levels of different attributes on (typically) two profiles of some sort – in our case political candidates. Respondents are then asked to choose between the two profiles. In terms of lexicographic preferences and party affiliation, this means that sometimes respondents are presented with two candidates from the same party. Actually, this will happen about half the time when there is just two parties to choose from: Republicans and Democrats. Our simple idea is to compare candidate choices when the party affiliation is the same between the candidates (i.e., a republican and a republican) to when it is different (i.e., a democrat and a republican). Below is one of the results of doing that from the article. The results are further divided by respondent party.

As can be seen from the third “difference” panel, there seems to be evidence of party affiliation being a first-order preference, and only when this preference is made “irrelevant” by the conjoint experiment (when the two candidate profiles present candidates from the same party), is gender and race allowed to play a larger role.
See the article for more details:

We hope that the theory of lexicographic preferences (which we of course did not invent) will enjoy larger popularity in the coming years, because there could be a number of applications of it – also outside candidate choice. For example, do parents have lexicographic preferences when choosing daycare institutions (could distance or public/private be first-order preferences)?
- Rosenberger, RS et al. (2003) Measuring dispositions for lexicographic preferences of environmental goods: Integrating economics, psychology and ethics. Ecological Economics 44(1), 63–76. ↩︎
- Goldmacher, S, Nagourney, A, and Medina, J (2020) The Kamala Harris pick: Geographic balance takes back seat to gender, race. The New York Times, 11 August 2020. Available from https://www.nytimes.com/2020/08/11/us/politics/harris-biden-geography-balance.html ↩︎
- Druckman, JN et al. (2021) Affective polarization, local contexts and public opinion in America. Nature Human Behaviour 5(1), 28–38. ↩︎
- Scott, A. 2002. “Identifying and analysing dominant preferences in discrete choice experiments: An application in health care.” Journal of Economic Psychology 23, no. 3 (June 1, 2002): 383–398. ↩︎