The Politics of Small Business Owners
The Politics of Small Business Owners (DOI: https://doi.org/10.1017/S0007123425000274) was written by Neil Malhotra, Yotam Margalit, and Saikun Shi in 2025. It was published in the British Journal of Political Science.
The authors begin by exploring small business owners in existing national and international data sets. Within the U.S., the ANES and GSS ask about employment (which includes a self-employed option), which the authors use as a proxy for owning a small business. They regress Republican partisanship on this proxy, while also controlling for age, race, gender, education, income, and region. Based on the significance of the proxy's estimated coefficient, they determine that there is a consistent Republican partisan lean among this subpopulation over time.
The ISSP asked a similar question across 18 countries in 2022. The authors repeat the prior analysis, running once on a pooled multinational data set and again for each national subpopulation. They control for gender, age, education, and economic standing.
- Most estimated coefficients indicate a significant right lean. More specifically, all estimated coefficients that are significant indicate a right lean.
- The estimated coefficient for the US subpopulation actually has the smallest degree among those that are significant.
The authors then use other U.S.-based data collections for electoral behaviors of medical practitioners. If a medical entity is only linked to one doctor, they consider it to be a small business and consider that doctor to be its owner. The authors repeat the analysis once more using this proxy. They determine that this subpopulation within the set of medical practitioners is more likely to register as a Republican and more likely to donate to a Republican campaign.
The authors then assemble a novel frame from the set of applications for Paycheck Protection Program (PPP) loans. To avoid duplicates they use only the first round of applications: 557,859 in total.
A random sample of the PPP application list is selected, for which efforts are made to identify the owner and their contact information. From the successfully identified subsample, a stratified sample is selected for contact. This process is repeated for each wave of the survey. Over four waves, 262,347 small business owners were identified, of which 221,142 were contacted.
Invitations to a web survey link were sent by email and text message. An incentive of $20 was offered. The survey collected 701 respondents.
A similar web survey is also administered using the Prolific web panel, to establish a gen pop reference. Quotas were set on region, gender, race, age, and political party. 2,472 responses were collected, of which 1,586 self-reported as not owning and never having owned a small business.
The authors first construct two identification strategies for right-leaning individuals:
- A series of questions asking for party identification; then strong vs not strong partisan; or for independents, whether they lean Democrat or Republican. This yields a 6-point scale, which they recast to be between 0 and 1.
- A question asking for a vote choice in a hypothetical election between Biden and Trump. Biden is coded as 0; Trump is coded as 1; anything else is coded as 0.5.
They also construct several categorical variables to test on:
- current vs past owner vs non-owner
- founder vs inheriting owner vs non-owner
- employer vs nonemployer owner (i.e., does the small business have more than 1 employee) vs non-owner
They regress on these categorical variables with the non-owner level as the left-out dummy.
Lastly, they construct a wide variety of behavioral measures: an index for attitudes about regulation; an index for attitudes about taxation; and index for attitudes about filing taxes; social conservatism markers; markers for attitudes about risk; markers for attitudes towards the "Weberian concept of the Protestant work ethic"; and markers for attitudes about individualism.
The authors begin analysis by regressing on the candidate partisan identifiers, to test for a selection effect. These include only demographic controls.
- They find that nonemployer owners are not distinct. Employer owners are however more likely to lean right by either partisan identifier.
- When looking at the hypothetical election partisan identifier, currently owning a small business and having founded a small business have a significant positive (right-leaning) coefficient.
- When looking at the party-based partisan identifier, those same identifiers lose statistical significance when controls are included. Inheriting a small business however does have a significant effect.
The authors then introduce their behavioral measures. They regress on the owner identifier and on the two partisan identifiers. For these measures to be a plausible explanation for differences, there must be a significant coefficient for both. Furthermore, the authors assert that a selection effect would be indicated by one of the personality measures while a treatment effect would be indicated by the opinion measures. They split this analysis between employer and non-employer owners based on the above analysis.
- The only plausible explanation is the index for attitudes about regulation.
- The index for attitudes about filing taxes is predictive of small business ownership. The index for attitudes about taxation is predictive of both partisan identifiers. It is notable that there is no crossover.
Repeating the earlier models with the index for attitudes about regulation as a control, small business owners are no longer more right-leaning than the general population.
Reading Notes
The authors have experimented with partisan identification, and not in a good way.
Asking about how they would vote in a hypothetical election is purely inferior to asking how they voted in a past election. Recall is an error term, but so is overestimation of the likelihood to vote. At least with a past election, estimates can be post-stratified to known results.
The construction of the party identification scale fits independents into a middling value. The larger literature around partisan identity indicates that independents should instead be removed from analysis. I also take issue to the questionnaire design (what sort of respondent volunteers an identity of "not strong"?). Forcing independents to make a dichotomous choice of their lean is also questionable.
I don't get the assertion that only personality measures can indicate a selection effect. There seems to be so much plausible overlap between attitudes about regulation and about Protestant work ethic. Need to think about this a bit more.