Data and Methods
We leverage five data sources; see Supplementary Information Section A for summary statistics.Footnote 2 Our outcome of interest is a categorical variable summarizing PAC pledges (or lack thereof) made by 280 major U.S. companies to the Capitol insurrection based on a CNN survey, which sampled all Fortune 500 corporations with PACs that contributed to Republican Objectors before 2021 (Hernandez and Yellin Reference Hernandez and Yellin2021). Thirty-six companies pledged to halt contributions to Objectors, which we label as “Targeted Response.” Another 87 firms announced a pause to all federal giving in order to reevaluate their criteria for candidate selection (Hernandez and Yellin Reference Hernandez and Yellin2021), which we code as “Nontargeted Response.” Of the total sample, 147 companies did not respond and 10 reported no change to their PAC contribution strategies. We group these companies in the “No Response” category given their shared avoidance of committing to costly constraints on their PAC contributions even if such constraints may demonstrate firms’ democratic values. Our findings are robust to excluding the 10 PACs that refused to make PAC pledges (Supplementary Information Table C1).
To test these corporate PAC pledges’ responsiveness to stakeholder partisanship, we collect campaign finance records (OpenSecrets 2021) and an original dataset of followers of corporate Twitter accounts (see Supplementary Information Section B for further details on measurement strategies). First, firms may be more likely to pledge changes in their contributions after the Capitol insurrection if those who oversee their PACs—executives and public-affairs specialists—perceive a strategic need to support Democratic candidates (Center for Political Accountability 2021). We measure the long-run partisan orientation of corporate PACs as the percentage of each PAC’s contributions to Democratic (versus Republican) candidates or party committees throughout the 2010–2020 election cycles.
Second, we infer employees’ partisanship from their individual campaign donation histories. Federal campaign finance records disclose each donor’s self-reported employment. For each employee donor, we calculate the share of their contributions to Democratic (versus Republican) recipients during 2010–2020 (most of them only donated to candidates from one party; see Li Reference Li2018). We average individual-level shares of contributions to Democrats across employee donors within a given firm, weighting all employees equally.
Although it would be ideal to construct comparable partisanship measures for consumers and shareholders, individuals do not report their consumption choices or asset ownership when making contributions or registering to vote. Given data constraints, we present another proxy of stakeholder partisanship (particularly for nonelite, nonemployee stakeholders) based on firms’ Twitter followers. To overview the construction of this variable, we collected Twitter handles of Fortune 500 firms and obtained their followers via an academic license to the Twitter API. We calculate the two-party share of corporate Twitter followers who additionally follow Senator Elizabeth Warren versus Senator Ted Cruz, who are comparable with respect to Twitter following, level of political office, and ideological extremism (VoteView 2021). Because Twitter follower networks for political accounts exhibit ideological homophily (Barberá Reference Barberá2015), firms that share more followers with Senator Warren than with Senator Cruz will generally have more left-leaning stakeholders. Compared with our campaign finance-based partisanship measures, this Twitter-based measure may better represent nonelite corporate stakeholders (e.g., consumers) given lower resource barriers to political participation via social media than via campaign finance (Brady, Verba, and Schlozman Reference Brady, Verba and Schlozman1995).
For auxiliary measures, we collect data on firm revenue, employment, and assets in 2019 from the Fortune magazine (Fortune 2021), merged on firm using the fastLink R package (Enamorado, Fifield, and Imai Reference Enamorado, Fifield and Imai2019) and manual monitoring. We also obtain sector classification from the Center for Responsive Politics (OpenSecrets 2021).
To test our hypothesis that stakeholder partisanship predicts corporate PAC pledges following the Capitol insurrection, we estimate a multinomial logistic regression because the different types of pledges may be qualitatively distinct:
(1) $$ {\displaystyle \begin{array}{cc}& \hskip-10.12em \ln \left(\frac{\Pr \left({Y}_j\hskip0.35em =\hskip0.35em a\right)}{\Pr \left({Y}_j\hskip0.35em =\hskip0.35em \mathrm{No}\;\mathrm{Response}\right)}\right)\\ {}=& \hskip-2.12em {\beta}_{0a}+{\beta}_{1a}\hskip0.35em \%\hskip0.35em \mathrm{of}\;\mathrm{PAC}\;\mathrm{Donations}\ \mathrm{to}\;{\mathrm{Democrats}}_j\\ {}+& \hskip-1.12em {\beta}_{2a}\hskip0.35em \%\hskip0.35em \mathrm{of}\ \mathrm{Employee}\ \mathrm{Donors}\;\mathrm{who}\;\mathrm{are}\;{\mathrm{Democrats}}_j\\ {}+& \hskip-0.5em {\beta}_{3a}\hskip0.35em \%\hskip0.35em \mathrm{of}\ \mathrm{Twitter}\ \mathrm{Followers}\;\mathrm{who}\;\mathrm{are}\;{\mathrm{Democrats}}_j\\ {}+& \hskip-3.5em {\beta}_{4a}\log \left({\mathrm{Assets}}_j\right)+{\beta}_{5a}\log \left({\mathrm{Employees}}_j\right)\\ {}+& \hskip-10.12em {\beta}_{6a}\log \left({\mathrm{Revenue}}_j\right)+{\lambda}_{sa,}\end{array}}\hskip0.48em $$
where $ {Y}_j $ represents firm $ j $ ’s response as one of three categories: No Response (NR), Nontargeted Response (NTR), and Targeted Response (TR). This regression can be seen as two equations, one for each alternative level $ a $ to the baseline NR. In these two equations, the regressors are the same but the coefficients differ. We hypothesize $ {\beta}_{1a} $ through $ {\beta}_{3a} $ to be positive as the likelihood of PAC pledges should on average increase with the Democratic leaning of stakeholders. Coefficients $ {\beta}_{4a} $ through $ {\beta}_{6a} $ capture variation in the likelihood of corporate PAC pledges owing to different measures of firm size. Finally, $ {\lambda}_{sa} $ represents sector (s) fixed effects.