Government elections tend to fall into one of two categories: district elections and at-large elections.

In the first type, a city, state, or country is split into districts, and residents can vote only for candidates in their district. For example, each ward in a city might elect its own alderman. In the second type, every person can vote for any candidate. For instance, an entire city might choose from the same pool of candidates to elect city council members.

But do the election rules affect what types of people get voted into office?

Yes, according to a recent experiment in Afghanistan. In 2002, the Afghan government designed a program to fund public resources, such as wells and bridges, in villages. Each village needed to elect a council to manage the projects. Some villages were instructed to use district elections, while others used at-large elections.

When researchers analyzed election results, they saw a clear pattern. In at-large elections, villagers elected more educated people. And these officials had less extreme policy preferences than those chosen in district elections.

“The rules do affect who gets elected,” says study coauthor Georgy Egorov, a professor of managerial economics and decision sciences at Kellogg.

Does that mean at-large elections are better? Not necessarily, Egorov says.

A society that values having a wide diversity of views represented might prefer district elections, which result in more polarized councils. When extreme views are represented in government, Egorov says, perhaps citizens who support those policies do not have to resort to protests or violence to make their voices heard.

The study cannot say whether district or at-large elections are better, but it is clear that voting systems can influence who ends up in power.

While the Afghanistan elections differ in important ways from elections in other countries, some lessons might be relevant to the U.S. Take the 2010 Congressional campaigns, during which the extremely conservative Tea Party won seats in district elections to the House of Representatives.

“It’s consistent with their electoral success,” Egorov says.

A Unique Data Set

The Afghanistan program, called the National Solidarity Programme, was set up with the help of NGOs to distribute more than $1.1 billion to about 32,000 villages. Elected councils would implement projects such as wells, bridges, irrigation canals, electrical generators, and literacy classes.

The program dictated that any adult villager could be elected, and votes would be cast by secret ballot. But beyond that, the Afghan government did not have a strong sense of “the right way” the elections should be held, Egorov says. So the government allowed Egorov and his colleagues to randomly assign 250 villages to either a district or an at-large voting system as an experiment. Egorov collaborated with Andrew Beath at the World Bank, Fotini Christia at the Massachusetts Institute of Technology, and Ruben Enikolopov at the Barcelona Institute of Political Economy and Governance.

Data collected by NGOs about the outcomes of these elections provided the researchers with a few key pieces of information.

One was each elected council member’s education level. The researchers used this as a rough proxy for a person’s likely competence. While the correlation is imperfect, Egorov says, one would hope that education provides people with skills, such as literacy, that may help them implement projects successfully.

Another data set contained the GPS coordinates of each male council member’s house in 140 villages. (Councils were required to be 50 percent women, but for religious reasons, the NGOs collected locations only for men.)

This allowed the researchers to gauge a key policy preference for representatives: where a given resource, such as a bridge or literacy class, should be located. Presumably, each council member wanted the resource as close to his house as possible. Using the GPS coordinate of a council member’s house to infer his “policy” preference was likely more reliable than asking council members to report their views, since participants may not always tell the truth.

“It’s probably among the very few data sets where you can claim you observe political preferences,” Egorov says.

Making a Trade-off

The team started by analyzing the education level of council members.

In villages with district elections, 7 percent of male representatives had finished high school; in villages with at-large elections, that figure was 11 percent. (No difference was seen among female council members because Afghan women generally received very little education.)

To investigate further, the team developed a mathematical model of elections, where villagers considered both the education levels and the home locations of their fellow villagers when casting their votes. In some cases, “you might rather vote for somebody who’s incompetent but at least shares your political preferences,” Egorov says.

Consider a simple scenario where there are two districts—say, east and west—and the village must elect two council members. Only one person in the village is educated, as schooling is rare in these villages. Each voter wants the public resource close to their house. If one were to average all voters’ preferences, the ideal location would be the village’s center.

The model assumed that the two council members would compromise on location. “They will just settle on choosing a location halfway between them,” Egorov says.

So what would happen in the two types of elections?

The model predicted that in an at-large election, villagers would choose the single educated person as one council member. For the second member, voters would, on average, counterbalance the educated person with someone who held the opposite views on where the resource should be located—that is, someone living in the opposite location as the competent candidate. That way, the public resource would be near the center.

In district elections, voters faced a more complex choice. They knew that the public resource would end up between the two district representatives’ homes. So each district might be likely to elect an extreme candidate—that is, someone on the far east or far west side. “He will bargain tougher and pull the location of the public good closer to himself,” Egorov says.

If the educated candidate lived close to the village center, his district might be less likely to elect him.

The model also predicted that such effects would be weaker in homogeneous villages where people share similar preferences. For example, in a smaller village, everyone lives close to the center, so district voters face less of a trade-off between competence and extremism.

Paths to Power

The team then tested the model’s predictions on the Afghanistan data. They found that, as predicted, villages with district elections chose more extreme representatives. Their male council members lived an average of 25 percent farther from the village center than those elected in at-large elections.

The researchers also assessed each village’s homogeneity. They studied village size, as well as other indicators such as ethnic diversity and villagers’ responses to surveys about project preferences.

As expected, the village’s homogeneity made a difference. Among villages where people were likely to share policy preferences on the type and location of resources, there was no link between the election system and the council members’ education level.

Finally, the researchers analyzed whether council members who were educated—which the researchers were using as a proxy for competence—did indeed implement more successful projects. Controlling for factors such as the type and cost of projects, the team found that projects were started and completed more quickly in villages with at-large elections, where more educated council members were more likely to be elected.

And there were some signs that residents of these villages enjoyed better welfare after the project’s implementation. For example, children suffered from diarrhea less frequently, and harvests produced slightly more income.

These elections were, of course, vastly different than those in countries such as the United States. For instance, candidates did not campaign and were not organized into political parties.

However, the researchers do see a parallel with the 2010 U.S. election when Tea Party–supported candidates claimed seats in Congress, where both houses are elected via district elections. Interestingly, other scholars did not see a corresponding rise in extreme conservative views among Americans at the time that these more conservative candidates won seats.

One possible reason is that the 2009 stimulus package had pushed the issue of national debt to the forefront. So voters who disliked government spending supported Tea Party candidates who railed against it as a way to try to swing the debt pendulum back toward the center, in the same way Afghan voters wanted to pull their local resource toward the center of the village.

The Tea Party representatives would “pull the blanket quite a bit in your direction” during policy-making, Egorov says. Like the Afghan villagers voting in district elections, Americans might have elected extreme candidates so that the government’s final position would be closer to their own.

While the study cannot say whether district or at-large elections are better, Egorov says, it is clear that voting systems can influence who ends up in power.

“The type of election matters,” he says.