Making voting experiments more realistic: The ‘hybrid’ experimental design

By Delia Dumitrescu, Post-doctoral fellow, University of Montreal

What is the story?

In general, laboratory experiments that study voting behaviour use a research design based on monetary incentives. Participants decide to vote, strategically or sincerely, or to abstain in a way that maximizes their final monetary payoff. Also, most laboratory experiments usually rely on groups of a few tens of people (for an example of such design, see this previous blog entry).

Yet, these designs are disconnected from real elections. In real life, people rarely vote for immediate financial gain reasons; rather, they vote for a party because they care about what it stands for. And the vote occurs in communities of thousands of voters, making the likelihood that an individual will cast the pivotal vote nil.

To get closer to the reality of elections, we propose an original ‘hybrid’ experimental design. Two specific features make this closer to real elections compared to classical laboratory experiments. First, voters are asked to decide on the allocation of resources for a policy cause about which they care. Second, each voter is located in a wider community of thousands of voters, whose collectively tallied ballots create the election’s outcome.

Experimental protocol

To test the validity of the hybrid experimental design, we conducted an online voting experiment in which 200 individuals took part. Participants were recruited online and received a base compensation of 5$. They had to vote (anonymously) for one of the three following options that correspond to a given allocation of an amount of 600$ (see Figure 1).








If Party A won, 500$ would be donated to the pro-environment NGO ‘Equiterre’. If Party B won, this NGO would receive 300$, and if it was Party C, 100$. The support for the environmental cause thus decreased significantly from A to C. Also, participants were told that the rest of the money would be divided equally among them as a bonus (0.5$ each in case Party A is elected, 1.50$ if this is Party B, and 2.5$ for Party C).

The treatment

The winner of the election was decided on the base of 9,200 votes. Participants were made aware that 9,000 early votes had already been cast, and that their votes would be added to the count. These early votes allowed us to manipulate parties’ winning chances. Since we expected most of our sample to support Party A, we designed the distribution of early votes in such a way as to place Party A last in the race.

In order to see how individuals vote when their preferred party has various winning chances, participants were asked to vote in five independent elections with different early vote distributions (see Table 1).







With the exception of the first election, the so-called ‘open-race’ election, Party A was significantly behind the other two parties in all elections. We refer to these elections as the ‘strategic’ elections. In all of them, Party C was in front. In two of them, Party A was separated from Party C by 110 votes, while in the other two, Party A was behind Party C by at least 200 votes. In the fourth election, it was impossible for Party A to win even if all participants voted for it.


Table 2 presents participants’ voting behaviour in the strategic elections as a function of their vote in the open race election. As expected, most of them (N=119) supported Party A in the open race, while only 50 participants voted for Party B, and 31 participants for Party C. Due to the very nature of vote distribution for this election, these votes were considered as sincere.

The most striking result is the low level of defection from Party A in the strategic elections. Despite its low winning chances, 60% to 86% of Party A supporters (i.e. those who voted for it in the open race election) also voted for this party in other contexts. In particular, 60% did so even when the party had no chance whatsoever of winning (see Table 2).








These low levels of strategic defection are very close to those observed in real life elections and at the same time much smaller than those observed in other laboratory studies. This is a strong case in favour of this type of hybrid experimental design for the study of voting behaviour. It is an interesting alternative to laboratory experiments that are sometimes considered as disconnected from the reality. Also, the hybrid experimental design shares some convenient platform and procedural characteristics with other studies. In particular, it takes place online, thereby maximizing the easiness with which individuals can take part in the study. It also asks people to vote in repeated elections, like laboratory studies. Finally, hybrid experiments do not require as much resources as field experiments. They allow us to increase the external validity of the manipulation at relatively low cost.

For more information, see Dumitrescu, Delia, and André Blais. 2011. “Increased Realism at Lower Cost: The Case of the Hybrid Experiment.” PS: Political Science & Politics 44: 521-523.

Follow us on Facebook and Twitter.

This entry was posted in Blog. Bookmark the permalink. Both comments and trackbacks are currently closed.