Economic Spillovers of Political Polarization

Download data and study materials from OSF

Principal investigators:

Neil Malhotra

Stanford University



Matthew Levendusky

University of Pennsylvania



Yotam Margalit

Columbia University



Sample size: 3366

Field period: 12/29/2014-06/23/2015

Partisanship is increasingly seen as an impediment to cooperation in political settings in the United States given high levels of political polarization. But does partisanship also affect behavior in non-political settings? A field experiment in an online labor market indicates that workers exhibit systematically lower reservation wages when the employer shares their political stance, reflecting a preference to work for co-partisans. A second experiment administered on a large, national classified ads platform over the course of one-and-a-half years shows that consumers, primarily in response to Republican sellers, are more likely to shun (seek) profitable market transactions with sellers affiliated with the opposing (same) party. Via a population-based survey experiment, we find that the influence of political considerations on economic choices is not confined to ardent partisans; rather, it extends throughout the electorate. These negative externalities in the economic domain suggest that partisanship shapes cooperation in everyday behavior.

If partisanship spills over into the economic domain, then we would expect respondents to forgo material gains when explicitly presented with an option to express their partisan identities. Comparing across these conditions, we expect that subjects will be least likely to accept the non-partisan offer (lower payment, but no benefit to the other geographic region/religion/party) in the Geography Condition compared to the other three conditions. That is, subjects will not pay to express their minimal-group geographic preference, but they will leave money on the table to express religious or partisan preferences, though the higher payment may cause some people to be willing to donate to the other party.

Experimental Manipulations
At the beginning of the survey, we measured respondents' partisan identification using the standard question wording employed by the American National Election Study (see Online Appendix 8 for the full questionnaire). Respondents also answered several additional questions. First, they reported their religious affiliation. Second, they were asked whether they live East or West of the Mississippi River. We use these two questions to develop baseline (apolitical) benchmarks, described below. They also answered a series of distractor items to disguise the true purpose of the study.

The final question in the survey constitutes the experimental manipulation. We told respondents: "As an additional thank you for filling out this questionnaire, we would like to give you a bonus cash payment. You can choose one of the two options below." One option is a simple payment of $3, which we refer to as a non-partisan offer. The second option, which we call the partisan offer, is a higher payment plus a donation to a group that is presumably disliked by the respondent. Respondents were randomly assigned to one of four experimental conditions, each of which presents a different partisan offer.

First, respondents could be randomized into the baseline political treatment, which we call the Partisanship Condition. Here, they are given the choice between the non-partisan offer of receiving $3, or the partisan offer of receiving a $6 payment themselves with an additional $4 donation to the opposing party's national committee (which we explained works to elect candidates from the other party). The donation here is designed to be small enough such that respondents know it will not affect the outcome of elections, but large enough to evoke animus toward a disliked group. In all conditions, respondents saw this trade-off both as text, and as figures in a comparison table, to ensure that they actually understood the proposed exchange. Thus, to express their political preferences and avoid contributing to the other party, respondents have to leave half of the money they have been offered on the table. This design produces a fairly straightforward test of the economic consequences of partisanship: Will respondents forego gains to express their partisan preferences?

Second, respondents could be offered the same setup as the baseline political treatment, except now they are offered $9 instead of $6 (all other variables, including the payment to the other party, remain the same); we refer to this condition as the Higher Payment Condition. In this case, instead of simply doubling the payment they receive for helping the other party, we triple it, making it even more difficult to accept the non-partisan offer. This condition allows us to test the elasticity of these partisan effects: As the cost of expressing one's partisanship increases, how does behavior change?
In the third condition (the Religion Condition), respondents are presented with an offer to fund a disliked religious group. Christian respondents were told they would be funding the American Atheists, and atheist/agnostic respondents were told they would be funding the Christian Legal Society. All dollar amounts remain the same: a $6 payment to the individual, and $4 to the disliked religious group. This gives us an apolitical benchmark against which we can compare the effects of partisanship. Religion is a large, and socially meaningful cleavage (Pew Research 2016). So by assessing how the effects of partisanship vary relative to religion, we can better understand the effect sizes in the partisan conditions and put them into context.

Finally, in the fourth condition (the Geography Condition), respondents have the option of funding another geographic region. We divided people into those east and west of the Mississippi River using the item above. We told those east (west) of the Mississippi River they would be funding the Association of Western (Eastern) States, which advocates for policies that benefit those living west (east) of the Mississippi River. We use this as a relatively meaningless placebo division: the Mississippi River does not represent any meaningful division in American political, economic, or social life. This scenario directly parallels the minimal groups paradigm from social psychology (Tajfel and Turner 1979). This condition therefore provides a floor effect: How much will people pay to express their group identity when that group identity is effectively trivial? Contrasting the effects of partisanship to those of religion and geography helps us contextualize our findings.

The outcome variable is whether the respondent chose the non-partisan option (forgoing material gains to express out-group animus against the other party, religion, or geographic group, depending upon condition).

Summary of Results
We find strong evidence that partisanship leads people to forgo economic gains, and therefore distorts conventional decision making. The acceptance rate for the non-partisan option in the Partisanship Condition was 75.4%. This means that three-fourths of respondents were willing to give up a doubling of their bonus payment simply to avoid the alternative option that offered a benefit to the other party. This high rate suggests that partisan animus is real and has behavioral consequences.
Does this high percentage merely reflect people's unwillingness to donate to any political cause as opposed to negative affect toward the opposing party specifically? To address this issue, we conducted a follow-up experiment on Mechanical Turk, where we offered people a choice between receiving a payment of $0.50 and receiving a higher payment of $1.00 plus a $0.50 donation to their own party. The expectations here are reversed. If our mechanism is correct, then most people should be selecting the latter option since they receive personal benefits and get to help out their own party. On the other hand, if people were against donating to anything political, then they would prefer to take the lower payment. However, we do indeed find that the acceptance rate of the higher payment is 85% (with a 95% confidence interval of 80% to 90%), which is substantially different from the acceptance rate of 24.6% in the original experiment. This makes us confident that our results are not simply due to an aversion to politics.

Returning to our main experiment, the acceptance rate in the Partisanship Condition was significantly higher than the 33.7% acceptance rate in the minimal-group Geography Condition, again suggesting that partisan animus is associated with real, behavioral ramifications. The acceptance rate in the Partisanship Condition was statistically indistinguishable from and substantively similar to the 77.0% acceptance rate in the Religion Condition, meaning that partisanship reflects as large a cleavage as religion, another long-standing and deeply rooted division in American society.

We did, however, find that partisan preferences were slightly elastic. Increasing the payment in Option B from $6 to $9 decreased the acceptance rate of the non-partisan offer by 4.8%. Although this difference is statistically significant (given that the study is very well-powered), it is not that substantively large, implying an elasticity of only .06, a doubling of the price of expressing a partisan opinion from $3 to $6 is only associated with a 4.8% reduction in demand. Further, almost the entirety of the elasticity effect is concentrated among respondents making under $35,000 per year. Among respondents making more than $35,000, the difference in the acceptance rates between the Partisanship and Higher Payment conditions is only 2.8 percentage points (p = .20), while among lower-income respondents it is 10.1 percentage points (p = .01). For most respondents (those above the bottom third of the income distribution), partisan animus is relatively inelastic.

Consistent with our expectations, and supportive of the idea that partisan attachment has meaningful behavioral implications, the acceptance rate of the non-partisan offer among strong partisans in the Partisanship Condition was an extremely high 89.1%. This figure is statistically distinguishable from and substantially larger than the 68.5% acceptance rate among non-strong partisans. Further, this 20.7 percentage point difference is significantly larger than the 6 percentage-point gap between the two groups in the Religion Condition. Finally, the placebo test finds (as expected) that strength of partisanship does not predict the acceptance rate in the Geography Condition; the difference between the subgroups is about 1 percentage point.

It is not terribly surprising to find that strong partisans overwhelmingly reject the partisan offer (indeed, if we had not, it would have called our design into question). What is more surprising and unexpected is that fully two-thirds of weak and leaning partisans similarly reject the partisan offer. Even those with only modest ties to their party are willing to forego additional gains to avoid benefiting the other party. The economic consequences of partisanship are not confined to a narrow segment of the public, but rather extend broadly throughout the electorate.


The Economic Consequences of Partisanship in a Polarized Era. Christopher McConnell, Yotam Margalit, Neil Malhotra, Matthew Levendusky. Paper presented at the 2015 Meeting of the American Political Science Association.