INTRODUCTION
For decades, scholars have cast large American businesses as a conservative stronghold (e.g., Cohen et al. Reference Cohen, Hazan, Tallarita and Weiss2019; Grossmann, Mahmood and Isaac Reference Grossmann, Mahmood and Isaac2021). But in recent years, scholars, politicians, and journalists have devoted increasing attention to what appear to be important changes in the political activities of these firms. Major companies, once staunch allies of conservative politicians, have issued numerous progressive-leaning statements on social and environmental issues, prompting complaints about so-called woke capitalism (e.g., Douthat Reference Douthat2018). Firms have also increased transparency and adopted governance safeguards around their lobbying and election spending, potentially changing the ways they engage in more traditional forms of political activity (Center for Political Accountability 2023). Despite these visible changes, however, we know little about the evolving ideological views of the individuals who decide how these companies participate in politics: corporate directors and executives. In this article, I leverage new data on the campaign contributions of 97,469 directors and executives at 9,005 different U.S. companies to better understand the changing ideological composition of these elite actors over the first two decades of the new millennium.
The results of my analysis reveal important shifts in the political landscape of corporate America in recent years. Tracking unidimensional, contribution-based measures of corporate director and executive ideology over time, I find that between 2001 and 2022, the average observed ideology of these individuals as a whole moved meaningfully to the left, starting from a modest conservative skew and eventually landing around the middle. This leftward shift is large relative to the ideological distribution of corporate elites, amounting to about 42% of a standard deviation. It also appears to have been driven largely by turnover in the ranks of corporate directors and executives, with the proportion of liberals increasing over time. Contrary to the perception that the left has taken over big business, however, conservatives still remain common in the ranks of corporate elites. Corporate America is thus neither red nor blue; it is purple.
I also document heterogeneity across different industries and corporate roles. Certain industries, such as energy, have been and remain largely conservative, while others, such as technology, have moved considerably to the left. Meanwhile, individuals in different corporate roles follow different ideological distributions, with chief executive officers (CEOs) skewing the most conservative and senior managers skewing the most liberal. Most CEOs continue to be conservative, even as liberal CEOs have become increasingly common, whereas senior managers have changed from a roughly even split to a strong liberal skew. Overall, average observed ideology for all roles has shifted left over time.
Taking a firm-level perspective, I document similarly large changes when aggregating ideology up to the company level, as well as changes in ideological diversity within firms.
To shed light on the reasons for these changes, I perform a decomposition that fits a model predicting individual ideology based on various individual features and employs simulations to assess how average ideology would change under different counterfactual histories for these features. This exercise suggests multiple reasons for the leftward shift, with relevant factors including changes in both the mix of individuals who form the pipeline for corporate elites and the identities of the largest firms. Increased gender and racial diversity seem to have played particularly important roles, as women and people of color tend to be more liberal and these individuals have become increasingly common in corporate America. Surprisingly, however, using a difference-in-differences design to study the political effects of legally mandated gender diversity for corporate boards, I find no evidence that exogenously increasing the proportion of female directors translates into a more liberal board, complicating the diversity explanation. I also identify the importance of geographic changes, with corporate elites having increasingly made their first donations while living in liberal areas of the country. Ultimately, though, about half the leftward shift appears to be explained by other factors that have changed over time, suggesting broader forces at work.
Finally, I develop a new dataset on public stances taken by companies on state legislation involving lesbian, gay, bisexual, transgender, and queer (LGBTQ) issues to study the connections between the ideology of individual corporate elites and the political activities of firms themselves. Consistent with companies responding to the views of influential stakeholders, I find that firms with more conservative directors and executives are much less likely to take public stances supporting LGBTQ rights. To more plausibly establish causality, I use an instrumental variables (IV) design, instrumenting for average individual ideology with the average for other firms in the same industry. The results hold under this quasi-experimental design.
The transformation documented in this article has profound implications for American politics. The increased prevalence of liberals among corporate elites could fundamentally reshape the coalitions supporting each political party, potentially yielding important policy changes as well. But corporate America has still not transformed from a conservative stronghold into a liberal one, as many companies remain staunchly conservative and others are ideologically divided. It is also unclear whether these trends will continue going forward.
POLITICS IN CORPORATE AMERICA THROUGH THE START OF THE MILLENNIUM
The conventional wisdom about business and politics has cast large U.S. companies as a conservative stronghold. Countless authors, both academic and popular, have documented a longstanding alliance between the Republican Party and “big business,” as reflected by (among other things) the wave of deregulation during the Reagan administration, the passage of corporate tax cuts under the Trump administration, and the success of conservative interest groups in passing pro-business legislation at the state level (e.g., Grossmann, Mahmood, and Isaac Reference Grossmann, Mahmood and Isaac2021; Hacker and Pierson Reference Hacker and Pierson2010; Hertel-Fernandez Reference Hertel-Fernandez2019). This alliance has been evident in campaign finance data as well. Bonica (Reference Bonica, Chilton and Sen2016) reports that as recently as 2012, the political action committees (PACs), directors, and CEOs of Fortune 500 companies tended to donate more to Republicans than Democrats. Cohen et al. (Reference Cohen, Hazan, Tallarita and Weiss2019) report similar findings for S&P 1500 CEOs as recently as 2017. Fos, Kempf, and Tsoutsoura (Reference Fos, Kempf and Tsoutsoura2023) complement these findings using voter registration data for top executives at 941 different companies, showing that these individuals skewed Republican as recently as 2020.
Given this body of evidence, one might be surprised to learn that many observers perceive that America’s largest companies have in recent years moved significantly to the left. Yet many journalists, politicians, and other observers have claimed exactly that. Using a term first coined by the New York Times columnist Ross Douthat (Reference Douthat2018), many conservatives have complained about the rise of so-called woke capitalism, with companies publicly taking progressive stances on social and political issues ranging from racial justice to climate change (e.g., Ramaswamy Reference Ramaswamy2021).Footnote 1 The perception of a leftward shift is also reflected in survey evidence, with Republican respondents perceiving that businesses have become more aligned with Democrats, a view shared by business leaders in both parties (Hersh and Shah Reference Hersh and Shah2025a). Nevertheless, many observers dispute this view. Survey evidence shows that unlike their Republican counterparts, Democratic mass public respondents perceive that businesses have become more aligned with Republicans (Hersh and Shah Reference Hersh and Shah2025a), and some scholars have reported that the proportion of senior executives registered to vote as Republicans has increased in recent years (Fos, Kempf, and Tsoutsoura Reference Fos, Kempf and Tsoutsoura2023).
To date, we lack sufficient evidence to adjudicate between these conflicting views. A key part of the problem is a lack of data on the ideology of a large sample of individual corporate elites over many years.Footnote 2 Although certain scholars have provided evidence on individuals using either campaign finance or voter registration data (Bonica Reference Bonica2016; Cohen et al. Reference Cohen, Hazan, Tallarita and Weiss2019; Fos, Jiang, and Nie Reference Fos, Jiang and Nie2024; Fos, Kempf, and Tsoutsoura Reference Fos, Kempf and Tsoutsoura2023), these studies—though valuable contributions to the literature—are limited in their timeframe, their sample, their measures, or a combination thereof. For example, Bonica (Reference Bonica, Chilton and Sen2016) uses the most fine-grained measure of ideology—the same campaign finance scores that I use in this article—but his study ends with the 2012 electoral cycle, which predates the purported leftward shift in corporate America. Cohen et al. (Reference Cohen, Hazan, Tallarita and Weiss2019), Fos, Kempf, and Tsoutsoura (Reference Fos, Kempf and Tsoutsoura2023), and Fos, Jiang, and Nie (Reference Fos, Jiang and Nie2024) analyze more recent data—their analyses, respectively, end in 2017, 2020, and 2021—but they use coarser measures of ideology—namely, the proportion donated to each party or voter registration. Moreover, all these studies tend to focus on the very highest echelons of corporate America, such as corporate directors and CEOs of Fortune 500 companies (Bonica Reference Bonica2016), CEOs of S&P 1500 companies (Cohen et al. Reference Cohen, Hazan, Tallarita and Weiss2019), the top five executives of 941 firms in nine states (Fos, Kempf, and Tsoutsoura Reference Fos, Kempf and Tsoutsoura2023), or solely directors albeit for a broader sample of firms (Fos, Jiang, and Nie Reference Fos, Jiang and Nie2024). This leaves us in the dark about the politics of many of the individuals who oversee key areas of the country’s largest businesses, ranging from division heads to chief marketing officers to general counsels.
DATA
I add to the literature on business and politics in the United States by undertaking the most comprehensive study of the ideology of U.S. corporate directors and executives to date, using new data on the campaign contributions of 97,469 unique individuals at 9,005 unique firms serving between 2001 and 2022. This section describes my main data (for the replication data, see Steel Reference Steel2025).
To create my dataset on corporate elite ideology, I merge two widely used databases, BoardEx (WRDS N.d.) and the Database on Ideology, Money in Politics, and Elections (DIME) (Bonica Reference Bonica2023). BoardEx provides data on over a million corporate directors and executives at tens of thousands of companies, which allows me to create a longitudinal dataset on corporate elites starting in 2001.Footnote 3 DIME in turn provides information about the political ideology of tens of millions of individuals donating to candidates for state and federal public office in the US, using a campaign finance measure developed by Bonica (Reference Bonica2014). I obtain additional firm financial data from Compustat (WRDS N.d.).
Merging BoardEx and DIME presents multiple challenges. First, the two databases lack common unique identifiers, and common fields such as employer are often reported inconsistently, especially for individuals who serve on one company’s board but report another company as their employer when making campaign contributions. Second, the inconsistencies do not fit a single pattern, suggesting that linear methods determining whether to accept a match as valid may fail to achieve optimal performance. Third, the databases are enormous, with 1.8 million unique individuals in BoardEx, 41.6 million unique contributors in DIME, and many more unique individual–employer pairs in each database. Manual review of all close calls is therefore infeasible.
To overcome these challenges, I adopt a multi-step supervised machine learning approach. I start by using the employment history for each individual that BoardEx provides to match on multiple potential employers, as in Teso (Reference Teso2025), and then identify all unique individual–employer pairs in DIME. Next, I adopt a two-stage procedure to match unique individuals in DIME with unique individuals in BoardEx. First, I perform exact matching on name and employer. Second, for unmatched DIME contributors, I identify candidate potential matches with BoardEx by requiring exact matches on name and fuzzy matches on employer with relatively low string similarity thresholds, hand-label a random sample of potential matches as valid or invalid, train a machine learning algorithm on the hand-labeled potential matches using several different features (e.g., multiple string similarity measures, name frequency, the presence of missing data), and use the trained model to classify unlabeled potential matches. My chosen model, a gradient-boosted tree algorithm, can flexibly classify candidate matches as valid or invalid using nonlinear methods, potentially accounting for multiple different types of errors that would otherwise result in a nonmatch. Section B of the Supplementary Material describes my methods in greater detail. This exercise yields matches for 366,208 out of 1,043,251 (35%) unique individuals in the North America BoardEx database. Randomly sampling matched pairs, I estimate precision (i.e., the proportion of matches that are valid) at 100%. I estimate recall (i.e., the proportion of valid matches that my algorithm has successfully identified) at 91%, though the true recall is likely lower.Footnote 4
After merging BoardEx and DIME, I limit the sample to firms headquartered in the US and select the top 4,000 firms by revenue for each year.Footnote 5 These firms generate a large amount of economic activity, earning about $21.2 trillion in revenue in 2022. The final sample contains 9,078 unique firms, 9,005 unique firms with at least one donor, 296,471 unique individuals, and 97,469 unique donors. Converting the sample into a panel, 40% of individual–year observations in BoardEx are matched to donors in DIME. Directors have a 60% match rate, much higher than executives lacking a board seat. For comparison, Bonica, Chilton, and Sen (Reference Bonica, Chilton and Sen2016) match about 43% of the lawyers in a snapshot of the Martindale–Hubbell directory to donors in DIME. Section C of the Supplementary Material presents descriptive statistics on the number of individuals and matched donors by period and address concerns with potential selection bias due to changes in BoardEx coverage over time and donor missingness. One takeaway is that BoardEx’s coverage of executives below the very top ranks appears to have increased over the first half of my sample period, so it is important to keep this change in coverage in mind when interpreting the results.
In addition to the ideology measures obtained from DIME, I gather several other important pieces of individual-level data. First, using BoardEx data, I identify whether each individual is a CEO, a member of the board of directors, an executive within the “c-suite” (e.g., chief operating officer, chief financial officer, etc.), and/or an executive outside the c-suite, whom I call “senior managers.” Second, I obtain each individual’s gender from BoardEx. Third, I use an original ensemble supervised machine learning method to predict the race and ethnicity of each individual, combining predictions generated by programatically querying OpenAI’s GPT-4o large language model and a recurrent neural network model pretrained on voter registration data (Xie Reference Xie2022). Section D of the Supplementary Material describes my method for predicting race and ethnicity in greater detail. Finally, to capture potential geographic influences on ideology, I obtain each individual’s earliest U.S. address recorded in DIME that can be geolocated with at least 0.8 confidence, identify the corresponding county using a simple features model, and calculate the average Republican proportion of the county two-party presidential vote share from 2000–2020, using election returns from the MIT Election Data and Science Lab (2024). This measure provides insight into the political context in which a donor first contributed, which may reflect the area where they grew up or some other area if they moved before making their first donation (e.g., the area where they lived early in their career).
Table A6 in the Supplementary Material contains variable definitions, and Table A7 in the Supplementary Material presents summary statistics.
THE EVOLVING IDEOLOGICAL COMPOSITION OF AMERICAN CORPORATE ELITES
Data in hand, I now turn to the main descriptive analysis. After explaining my measure for individual ideology, I document how the ideological composition of my sample of corporate elites changed between 2001 and 2022, alternately using the individual and the firm as the unit of analysis.
Measuring Ideology
I treat individual ideology as a latent variable and estimate it with the “CFScore” measure developed by Bonica (Reference Bonica2014),Footnote 6 which is based on campaign contributions. Bonica assumes a spatial utility model for donations, whereby donors generally give to ideologically proximate candidates and committees, and jointly estimates scores for both donors and recipients from a contingency matrix of donation amounts. These CFScores conceptualize ideology along a single left–right dimension and are essentially based on who donates to whom, with more liberal donors and recipients generally receiving lower scores and more conservative donors and recipients generally receiving higher scores. Importantly, unlike measures based on voter registration or the proportion donated to each party, the CFScore measure allows for distinctions within parties—such that a donation made to a progressive Democrat like Alexandria Ocasio-Cortez would be treated differently from a donation to a conservative Democrat like Joe Manchin, and likewise for Republican candidates—and they account for donations made to ideological groups that are not officially affiliated with a political party, such as the conservative Club for Growth or the liberal EMILY’s List.
Bonica and other researchers have validated these CFScores in various ways,Footnote 7 including by showing that they are powerful predictors of individual-level policy preferences as expressed in survey responses (Bonica Reference Bonica2019), but the scores do have one notable shortcoming: they are only available for donors. To the extent that donors are more ideologically extreme than nondonors, this means that the broader population of corporate elites may be more moderate than the sample of donors that I consider. But, given that donors tend to participate in politics much more actively than nondonors (e.g., Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995), there is reason to believe that corporate elite donors will be especially influential in shaping the ways that companies engage with the political system. Hence, we ought to pay careful attention to donor views even if—and perhaps especially if—they do not perfectly match those of the broader population of corporate elites.
Another potential shortcoming relates to the possibility of strategic behavior. The model underlying these scores assumes that individuals generally donate to ideologically proximate candidates. Although prior research validates this assumption even for corporate elites (Bonica Reference Bonica2016), individuals may still sometimes violate the assumption by donating to ideologically distant candidates for strategic reasons. If this means donating to “the other side,” scores for such individuals could be biased toward the center, making them appear more moderate than they truly are. Relatedly, corporate leaders may potentially temper their observable political behavior (i.e., donations) for strategic reasons, while engaging in different political behavior that we cannot observe.
A third limitation is that the scores are unidimensional, a simplification insofar as politics is multidimensional. Although a single ideological dimension explains much in contemporary American politics (e.g., McCarty, Poole, and Rosenthal Reference McCarty, Poole and Rosenthal2016), a richer, multidimensional analysis of corporate leaders’ ideology could yield additional insights. I leave such a multidimensional analysis to future research.
Because CFScores are static—that is, they do not change over time—I also estimate a dynamic version of the scores that varies by electoral cycle, similar to the dynamic scores that Bonica (Reference Bonica2014) estimates for candidates. Section E of the Supplementary Material describes my method for estimating these dynamic scores. My analyses primarily use the static scores due to their greater reliability,Footnote 8 but in certain cases I use the dynamic scores to analyze within-individual change.
Individuals
Using this contribution-based measure of individual ideology, we can now describe how the ideological composition of the corporate elites in my sample has changed over time. I start by taking the individual corporate director or executive as the unit of analysis.
As a first cut, I transform the data into a panel in which each row is an individual–role–year and fit separate local regression curves for each role, with CFScore as the outcome variable and year as the predictor.Footnote 9 Each curve reflects a running average of ideology over time for the corresponding corporate role. Figure 1 plots the results.
Figure 1. Average Ideology by Corporate Position over Time
As the figure shows, average observed ideology for CEOs, directors, c-suite executives, and senior managers alike moved meaningfully to the left over the sample period. All groups started with a modest conservative skew, but most skewed left or sat around the center by the end of the sample period. Aggregating across all groups, the average CFScore was 0.29 in 2001 and −0.05 in 2022, implying that the average CFScore shifted about 0.34 points to the left between 2001 and 2022. For comparison, the CFScore standard deviation in 2001 was 0.81, so the 0.34 shift amounts to about 42% of the initial standard deviation. The results are similar if we limit the analysis to directors and c-suite executivesFootnote 10 or to individuals who have made a nontrivial number of donations,Footnote 11 or if we weight the averages by firm revenue.Footnote 12
The figure also reveals substantial heterogeneity by corporate role. During all periods, CEOs have been the most conservative group, senior managers have been the most liberal group, and boards and c-suites have sat somewhere between the two. CEOs are the only group that continue to exhibit a meaningful rightward skew as of the most recent period, and even they have moved considerably to the left. Senior managers shifted from a modest rightward skew at the beginning of the sample period to a modest liberal skew by the end of the period.
Figures A19 and A20 in the Supplementary Material reveal heterogeneity by industry as well.
Figure A7 in the Supplementary Material reports similar results using dynamic ideology scores, with the main difference being a temporary rebound in the conservative direction during the first term of the Obama administration for certain roles. Overall, average ideology still moved from modestly conservative in 2001 to roughly the center by 2022. The dynamic scores also retain the same ordering across roles.
The dynamic scores also allow us to investigate the source of the heterogeneity by corporate role. In theory, these differences could arise either because rising up the corporate ladder makes people more conservative (a causal effect of corporate position on ideology) or because more conservative individuals tend to rise further up the corporate ladder (a selection effect or possibly even a causal effect of ideology on corporate position). To evaluate these competing hypotheses, I regress the dynamic CFScore measure on dummy variables for each corporate role, with year fixed effects and alternating whether to include individual fixed effects. Positive coefficients on the CEO, c-suite, and director variables in the individual fixed effects models would support the first hypothesis, while null results would support the second. Table A20 in the Supplementary Material reports the results, which largely support the second hypothesis: without individual fixed effects, the CEO, c-suite, and director coefficients are positive and sizable, but introducing individual fixed effects drives the coefficients close to zero, with small standard errors.Footnote 13
To explore in greater detail the evolving ideological composition of the corporate elites in my sample, I break down the sample into six periods corresponding to four-year presidential terms,Footnote 14 and for each period and corporate position, I estimate the density for static individual CFScores.Footnote 15 Figure 2 plots the results, truncating the densities on the right at 2.5 for ease of presentation.Footnote 16
Figure 2. Distribution of Ideology Scores by Corporate Position and Period
Plotting the distributions reveals several interesting patterns. First, it appears that the leftward shift documented in the previous figure has been driven largely by liberals replacing conservatives, as opposed to some or all ideological groups moving to the left. Second, there is meaningful polarization across all corporate roles, and this polarization appears to have grown somewhat over time as the middle has hollowed out. Third, the heterogeneity across roles appears both in the relative proportions of conservatives versus liberals within each role and in differences within broad ideological categories. Lastly, senior managers appear to be the most polarized group, having a smaller center and a larger gap between liberals and conservatives. Sections F and G of the Supplementary Material report additional quantitative detail about these compositional patterns and changes.
To provide substantive context for the scores, Figure 3 plots the densities for the most recent period, 2021–2022, with scores for several well-known politicians included as reference points. The dots at the bottom of the plot denote recipient CFScores for the individual politicians Alexandria Ocasio-Cortez, Bernie Sanders, Joe Biden (as presidential nominee), Sherrod Brown, Joe Manchin, Lisa Murkowski, Lindsey Graham, Rand Paul, Donald Trump (as presidential nominee), and Marjorie Taylor Greene. The dotted vertical lines show party averages for winners of the 2018 through 2022 congressional elections, with Democrats on the left and Republicans on the right. To provide additional context, Figure A8 in the Supplementary Material plots densities for the recipient CFScores for all winners of the 2018 through 2022 congressional elections, broken down by party.
Figure 3. Distribution of Ideology Scores by Corporate Position, 2021–2022, with Politician Reference Points
As Figure 3 shows, the modal liberal senior manager in the most recent period leans toward the progressive wing of the Democratic Party, while the modal liberal CEO looks more like a mainstream Democrat. Liberals in the other two groups tend to sit somewhere in between. Specifically, defining the mode as the CFScore with the highest density value, the modal liberal CEO, director, c-suite executive, and senior manager, respectively, have CFScores of −0.88, −0.95,−1.06, and −1.09. For conservatives, there is a less marked contrast between the different groups, with the modal conservative corporate elite looking like a mainstream Republican regardless of corporate role: the modal conservative CEO, director, c-suite executive, and senior manager, respectively, have CFScores of 1.08, 1.07, 1.11, and 1.14. For all corporate roles, there are also quite a few moderates. This contrasts with the distribution for elected politicians, who count very few moderates in their ranks, as reflected in Figure A8 in the Supplementary Material.
Firms
In addition to analyzing individual composition, it is instructive to aggregate up to the firm level, exploring ideological trends by company. A firm-level perspective can provide helpful context for understanding changes in corporate political activities.
Taking the firm as the unit of analysis, I start by calculating year-by-year firm-level averages for each corporate role and estimating their densities for each four-year period, similar to the individual composition analysis in Figure 2. Here, the key variable is the average ideology for the corporate role in question within some particular firm in a year, capturing overall firm-level ideology for a particular corporate role. To ensure that the analysis tracks groups rather than individuals, I also require that non-CEO firm–role–years have at least three matched individuals to be included in the analysis. Figure 4 plots the results.
Figure 4. Distribution of Firm-Level Average Ideology Scores by Corporate Position and Period
A few noteworthy takeaways emerge from this analysis. First, the dramatic changes observed at the individual level have also occurred at the firm level: firm-level averages for boards, CEOs, c-suites, and senior managers alike shifted from an initially conservative starting point in the early 2000s to become much more liberal by the end of the sample period. As of the most recent period, conservative CEOs still outnumber liberal ones, but liberal groups of senior managers and c-suite executives now outnumber conservative and even moderate groups. And, similar to individuals, the modal group of senior managers now leans toward the progressive wing of the Democratic Party, with a CFScore of −1.06. Meanwhile, corporate boards generally started around the center-right but now generally sit around the center. Second, aggregating director ideology up to the firm level makes the ideological distribution of corporate boards look very different from the distribution for individual directors. At the firm level, this distribution looks approximately normal, with a mean, median, and mode that by the end of the sample period ends up around the ideological center in American politics. The spread is also fairly tight, with few extreme boards in either direction. In other words, at the firm level, corporate boards mostly look like centrists. Indeed, among all corporate roles, corporate boards are generally the closest to the center: the firm-level average distance from the cutpoint that divides conservatives from liberals is 0.44 for boards, 0.73 for CEOs, 0.63 for c-suites, and 0.66 for senior managers. At the individual level, by contrast, directors are ideologically polarized. These results echo findings from prior research that individual directors appear partisan while boards as a whole appear bipartisan, perhaps reflecting deliberate efforts by firms to ensure that they have connections to politicians in both parties (Bonica Reference Bonica2016). Finally, there appears to be more heterogeneity in firm-level ideology for executives than boards: aggregated up to the firm level, executives continue to appear relatively polarized, though there are also many firms with center-leaning groups of executives. Thus, while the groups of individuals who exercise high-level supervisory control over companies (i.e., boards) may trend toward the center at the firm level, the same is not true for the groups of individuals who exercise day-to-day control (i.e., executives).
Section G of the Supplementary Material reports additional quantitative detail about these firm-level patterns and changes, including results on ideological diversity within firms.
EXPLAINING THE CHANGES
One of the headline findings from the previous section is that the average ideology of the individuals who run America’s largest companies moved meaningfully to the left during the first two decades of the new millennium. The obvious next question is why. In this section, I attempt to gain traction on the reasons for corporate America’s leftward turn.
Individual Change and Individual Replacement
As a threshold matter, we can distinguish between theories of individual change and theories of individual replacement—and largely rule out the former. According to a theory of individual change, ideological change comes from specific individuals becoming more liberal or more conservative over time. According to a theory of individual replacement, ideological change comes from individuals being replaced with others who have different ideological views (McCarty, Poole, and Rosenthal Reference McCarty, Poole and Rosenthal2016).
To adjudicate between these two theories, I rely on the dynamic ideology scores to perform a decomposition of the leftward shift into within-individual change and individual replacement components. Specifically, I assume that the data-generating process for individual ideology is given by a simple linear model with individual and year fixed effects, simulate different proportions of individuals in the final year of the sample, and use the fitted model to generate different predictions for individual ideology in each scenario. My procedure is as follows.
First, I estimate the following linear regression model on the full individual panel:
$$ \begin{array}{rl}{\theta}_{it}={\alpha}_i+{\gamma}_t+{\epsilon}_{it},& \end{array} $$
where i indexes individuals, t indexes years,
$ {\theta}_{it} $
is individual ideology (estimated using dynamic CFScores),
$ {\alpha}_i $
are individual fixed effects,
$ {\gamma}_t $
are year fixed effects, and
$ {\epsilon}_{it} $
is a random error term with mean zero.
Second, I create a counterfactual panel in which the proportion of individuals is held constant at 2001 levels in both periods. I construct this panel by calculating the proportions of observations with each unique individual identifier in 2001 and then using these proportions as sample weights to randomly assign individual identifiers to observations in 2022.
Finally, using the fitted model that was previously estimated on the full panel, I generate predictions for individual ideology on both the actual and the counterfactual panels and calculate the average predicted ideology for each. This provides estimates of the average ideology in the real world and in the counterfactual scenario in which the distribution of individuals did not change from 2001 to 2022. Taking the difference between these averages backs out an estimate of the “individual replacement” portion of the overall leftward shift.
Before proceeding to the results, one caveat should be noted. As explained in greater detail in Section C of the Supplementary Material, BoardEx coverage for senior managers appears to have increased over time, creating the possibility that changes in the sample may confound the decomposition. For that reason, I also report results limiting the analysis to directors only, as BoardEx coverage for directors appears to have remained relatively stable over time.
Table 1 reports the results. The first panel reports results for the complete sample of all corporate elites, and the second panel reports results for directors only. Within each panel, the first row shows the average predicted CFScore using the actual proportions (i.e., the real-world predicted average), and the second row shows the average predicted CFScore using the counterfactual 2001 firm proportions (i.e., the predicted average under the counterfactual scenario in which the distribution of firms remained the same).
Table 1. Counterfactual Simulations (Individual Proportions)
The results of this decomposition indicate that changes in which individuals comprise the sample are responsible for the vast majority of the leftward movement in average ideology between 2001 and 2022, though within-individual change in ideology also played a small role. For the complete sample of all corporate elites, assuming real-world proportions implies that the average predicted dynamic CFScore drops by 0.36 points between between 2001 and 2022. By contrast, assuming 2001 individual proportions in both periods implies that the average predicted dynamic CFScore drops by only 0.05 points. This implies that individual replacement is responsible for 0.31 points of the leftward shift (0.36 − 0.05 = 0.31), or about 86% of the overall change (
$ 0.31/0.36=0.86 $
). Limiting the decomposition to directors only, a similar analysis implies that individual replacement is responsible for about 84% of the leftward shift.
As an additional check to confirm that within-individual change constituted a small portion of the overall leftward shift, I fit to the same panel a linear model with individual fixed effects and a linear time trend. The time trend coefficient is only −0.001, which implies that a 21-year passage of time would predict a leftward shift in ideology of only 0.03 points (about 7% of the overall leftward shift). I obtain similar results when limiting the analysis to individuals who have contributed during every four-year period: the time trend coefficient is insignificant and only −0.0008. These results are untabulated but available from the author on request.
Firm Replacement
Within the set of individual-replacement theories, we can further divide these theories into two categories, based on (i) the replacement of conservatives with liberals due to broad changes in the population of corporate elites or (ii) the replacement of entire firms populated more by conservatives with firms populated more by liberals. According to theories in the first category, change comes from corporate elites who retire or otherwise exit the sample being replaced by more liberal individuals because the pipeline for corporate elites has become more liberal. According to theories in the second category, change occurs because certain firms exit the sample altogether—whether due to bankruptcy, an acquisition, or a decline in revenue that causes the firm to drop out of the top 4,000—and are replaced with firms populated by more liberal individuals. Of course, changes in the population of corporate elites can interact with the replacement of firms, but for now, I treat the theories as separate.
To test these theories, I turn back to static CFScores and perform a decomposition similar to the previous individual-level analysis, this time substituting firm for individual fixed effects and simulating different proportions of firms in the final year of the sample. That is, I estimate the following linear regression model on the full individual panel:
$$ \begin{array}{rl}{\theta}_{ift}={\alpha}_f+{\gamma}_t+{\epsilon}_{ift},& \end{array} $$
where i indexes individuals, f indexes firms, t indexes years,
$ {\theta}_{ift}={\theta}_i $
is individual ideology (assumed to be fixed),
$ {\alpha}_f $
are firm fixed effects,
$ {\gamma}_t $
are year fixed effects, and
$ {\epsilon}_{ift} $
is a random error term with mean zero. I then simulate different proportions of firms for 2022 and use the fitted model to predict individual ideology in each period. Table 2 reports the results.
Table 2. Counterfactual Simulations (Firm Proportions)
The decomposition results indicate that changes in which firms comprise the sample are responsible for a meaningful proportion of the leftward movement in average ideology between 2001 and 2022, but they do not tell the entire story. For the complete sample of all corporate elites, assuming real-world proportions implies that the average predicted CFScore drops by 0.35 points between 2001 and 2022. Assuming 2001 firm proportions in both periods, the average predicted CFScore drops by 0.23 points. This implies that the change in firms is responsible for about 34% of the overall change. Limiting the decomposition to directors only, a similar analysis implies that the change in firms is responsible for about 31% of the leftward shift.
In short, while broad changes in the population of corporate elites and the replacement of firms both seem to have some explanatory power, the broader population changes seem to be more important. In the next section, I dig deeper into these broader changes.
Changes in Individual Characteristics
To identify the types of broad changes in the population of corporate elites that have contributed to the leftward shift in average individual ideology, I look for changes in the proportions of corporate elites who bear individual characteristics that are associated with left-leaning ideology. The logic underlying this individual-characteristics theory of change is straightforward. Different individual characteristics are associated with different ideological views. The distribution of individual characteristics in the population of corporate elites has also changed over time. Therefore, changes in the distribution of individual characteristics in the population of corporate elites may have contributed to the leftward shift. Based on prior research in political behavior, I focus on five characteristics: gender, race, geography, education, and industry.
In Figures A10–A20 of the Supplementary Material, I provide descriptive evidence suggesting that these factors may have plausibly contributed to the leftward shift in ideology. First, within my sample of corporate elites, each of these characteristics is correlated with observed ideology in the expected way. That is, on average, women are more liberal than men, people of color are more liberal than non-Hispanic whites, people first donating in more liberal areas are more liberal than those in more conservative areas, people with more education tend to be more liberal than those with less education, and people who work in certain industries such as technology (i.e., “business equipment”) are more liberal than people who work in certain other industries such as energy. Second, the proportions of corporate elites with these characteristics have changed over time in a direction that would predict a leftward shift in average ideology. That is, the proportions of women, people of color, people from more liberal areas, people with certain higher degrees, and people working in more liberal industries have all increased over time.
The descriptive analysis also suggests that no single factor tells the entire story. Although there are predictable differences in average ideology between individuals based on their characteristics, average ideology has also moved to the left even within these characteristics. For example, although women are generally more liberal than men, both men and women have generally become more liberal over time.
To more formally analyze the contribution of changes in individual characteristics to the overall leftward shift, I perform another decomposition with counterfactual simulations, similar to the firm replacement analysis. Instead of a simple linear model with firm and year fixed effects, however, I allow for a more complex data-generating process for individual ideology. Specifically, I assume individual i’s ideology at firm f in role r at time t is given by
$$ \begin{array}{rl}{\theta}_{ifrt}=f({\mathbf{x}}_{ifrt})+{\epsilon}_{ifrt},& \end{array} $$
where
$ {\theta}_{ifrt}={\theta}_i $
is individual ideology (assumed to be fixed),
$ f(\cdot ) $
is a potentially nonlinear function,
$ {\mathbf{x}}_{ifrt} $
is a vector of potentially time-varying individual characteristics, and
$ {\epsilon}_{ifrt} $
is a random error term with mean zero.
For the individual characteristics,
$ {\mathbf{x}}_{ifrt} $
, I consider race and ethnicity, gender, geography, education, job title, and industry. Race and ethnicity are measured in a binary manner based on whether the person is predicted as non-Hispanic white; gender is measured in a binary manner as male or female; geography is measured as the Republican proportion of the two-party presidential vote share for the individual’s initial donation location; education is measured as the highest degree received by the individual; job title is measured by CEO, c-suite, and director dummy variables; and industry is defined by Fama–French 12 industry (Volkova Reference Volkova2017). I also include the year. Variable definitions are contained in Table A6 in the Supplementary Material.
Using the full individual panel, I then estimate
$ f(\cdot ) $
via machine learning. Initially remaining agnostic about the best model for the task, I try out two: a lasso linear regression model and a random forests model. After tuning the hyperparameters, I find that the random forests model significantly outperforms the linear regression model,Footnote
17 so I use the random forests model.
Figure A21 in the Supplementary Material displays a variable importance plot for the trained model. The plot indicates that geography is by far the most important predictor for individual ideology. Industry is a distant second, followed by education, time, gender, director status, race and ethnicity, and c-suite status. CEO status seems to be unimportant—or at least it does not add much beyond the combined effect of director and c-suite status. Table A22 in the Supplementary Material reports similar results for a simpler linear regression model.Footnote 18
Next, I create different counterfactual panels in which the distribution for either gender, race, geography, or industries is held constant at 2001 levels, similar to the firm-replacement counterfactual simulations.
Finally, I use the trained model to generate predictions for individual ideology on both the actual and the counterfactual panels. This provides estimates of the average ideology in the real world and in the counterfactual scenarios in which the distribution of either gender, race, geography, or industries did not change from 2001 to 2022. Once again, taking the difference between the actual and counterfactual averages provides a rough estimate of how much the change in the proportion of individuals bearing the characteristic contributed to the overall leftward shift.
Before proceeding to the results, I flag that some of the individual characteristics are measured with error (e.g., my model for predicting race sometimes produces inaccurate predictions), which can yield attenuation bias, making the estimates of the effects of these characteristics seem smaller than the true effects. By contrast, time is measured with precision. As a result, my results may understate the influence of certain individual characteristics while overstating the influence of other factors related to the passage of time.
With this caveat in mind, Table 3 reports the results. The first panel reports results for all corporate elites, and the second panel reports results for directors only. Within each panel, the first row shows the average predicted CFScore using the actual proportions; the second, third, fourth, fifth, and sixth rows, respectively, show the average predicted CFScore using the counterfactual 2001 gender, race, geography, industry, and education proportions.
Table 3. Counterfactual Simulations (Individual Characteristics)
The results of this decomposition indicate that increased gender diversity, increased racial diversity, and an increased tendency for corporate elites to come from liberal areas have all meaningfully contributed to the leftward shift, while broad changes in the distribution of industries that corporate elites work in have contributed relatively little. In the sample of all corporate elites, the differences between average ideology under the actual and counterfactual scenarios imply that increased gender diversity accounts for about 16% of the leftward shift, increased racial diversity for about 8%, geographic changes for about 18%, educational changes for about 3%, and industry changes for about 3%. In the director-only sample, increased gender diversity accounts for about 22% of the leftward shift, increased racial diversity for about 11%, geographic changes for about 17%, educational changes for about 1%, and industry changes for about 3%.
As a robustness check, Table A23 in the Supplementary Material reports results using a simpler linear model to generate predictions instead of the random forests model. The results are qualitatively similar.
Political Effects of Mandated Gender Diversity
Before concluding this section, I turn to a source of exogenous variation in gender diversity to get causal leverage on its political impact: a legal gender diversity mandate for corporate boards passed in California in late 2018. To the extent that diversity mandates force companies to hire more women and women are generally more liberal than men, we might expect such mandates to shift average boardroom ideology to the left.
The California law, known as S.B. 826, required a minimum number of women on boards of publicly held corporations incorporated or headquartered in California, implemented through phased-in requirements that became effective starting in 2019. Although a California court ultimately struck down the statute as unconstitutional, the law was effective through the end of 2021, creating an opportunity to study the effects of mandated gender diversity using a DID design. Specifically, I focus on its impact on the ideological composition of corporate boards, comparing changes in the average ideology of directors of firms covered and not covered by the statute before and after it went into effect. This research design relies on a parallel trends assumption to establish causality, meaning that the treated and control units must have followed parallel trends in the relevant outcome variable in the absence of the mandate. Section H of the Supplementary Material contains additional details about institutional background and the statistical models used to implement the design. I implement both a classic DID, which compares California and non-California firms before and after the law went into effect, and a triple DID, which additionally leverages the fact that only firms without enough women on their boards had to add more women.
Before presenting results on the ideological impact of the mandate, I first confirm that the law achieved its intended effect: increasing the number of women on corporate boards. Figure 5a shows an event-study plot of the average effect of the law on the number of women on the boards of California firms under the classic DID. For reference, the blue horizontal line on the plot is the standard deviation of the outcome variable. Table A3 in the Supplementary Material tabulates results for both the classic DID and the triple DID, as well as results for the proportion of women and the number of men as the outcome. These results confirm that California’s gender diversity mandate successfully increased both the number of women and the proportion of women on the boards of California firms, while not meaningfully affecting the number of men on boards.
Figure 5. Effects of California’s Gender Diversity Mandate (Classic DID)
I now turn to the ideological impact of the law, where the outcome variable is the average CFScore of the firm’s directors. This allows me to test the hypothesis that the diversity mandate moved average ideology to the left. Figure 5b shows an event-study plot of the average effect of the law on the average board ideology of California firms under the classic DID. For reference, the blue horizontal line on the plot is the standard deviation of the outcome variable, multiplied by −1. Table A4 tabulates results for both the classic DID and the triple DID. Surprisingly, these results indicate that despite increasing the representation of women on corporate boards, California’s gender diversity mandate had little impact on the average observed ideology of covered firms’ boards.
Given that women tend to be more liberal than men—a generalization that holds for corporate directors as well as the general public—why did the California law fail to move average board ideology to the left? Two possibilities arise, each potentially consistent with strategic efforts by firms to maintain ideological balance in their boardrooms. First, the newly added women directors may have been more conservative than the average woman director. Second, when new women join the board in response to the law, liberal men may leave the board.Footnote 19 In Table A24 in the Supplementary Material, I conduct the same analysis separately by gender and find evidence consistent with the latter explanation. Contrary to the hypothesis that average ideology for women became more conservative in response to the law, the ideology coefficient for women is negative and insignificant. But, supporting the hypothesis that average ideology for men became more conservative in response to the law, the ideology coefficient for men is positive and statistically significant in the triple DID specification, albeit insignificant in the classic DID specification.
CONSEQUENCES FOR CORPORATE POLITICAL ACTIVITY
The previous sections documented substantial ideological changes among corporate directors and executives in recent years and presented evidence about the reasons for these changes. In this section, I turn to the consequences, asking whether individual corporate elite ideology influences the ways that companies engage with the political system. That is, do the ideological views of a company’s directors and executives affect the political activities of the company itself?
More concretely, I study corporate positions on state legislation involving LGBTQ issues. In recent years, state legislators have proposed and in some cases passed legislation derided by its critics as anti-LGBTQ, such as Florida’s “Parental Rights in Education Act,” commonly known as the “Don’t Say Gay” bill (ACLU 2023). In 2020, the Human Rights Campaign (HRC), an LGBTQ advocacy group, started calling on businesses to sign a statement publicly opposing this wave of legislation (Human Rights Campaign Reference Campaign2024). According to the HRC’s website, 337 companies had signed the statement as of July 2024. To investigate whether the ideological views of directors and executives have influenced corporate signing decisions, I download lists of signatories to the statement and hand-code whether each of the companies in a sample of the largest U.S. companies signed the statement. Section I of the Supplementary Material provides the full text of the statement, additional details about the sample construction, and summary statistics. On the theory that corporate political activities are driven in part by the ideological views of corporate leaders and other stakeholders (Donaldson and Preston Reference Donaldson and Preston1995; Hambrick and Mason Reference Hambrick and Mason1984), I hypothesize that firms with more conservative directors and executives will be less likely to sign the statement.
As a first cut, Figure 6 plots the bivariate relationship between corporate decisions to sign the HRC statement and within-firm average individual ideology. On the x-axis is the within-firm mean CFScore, capturing the overall ideological leanings of the firm’s directors and executives, and on the y-axis is a binary variable indicating whether the firm signed the HRC statement. The curve is a local regression curve and thus can be understood as the estimated probability a firm signs the statement given a particular ideology level. The hash marks at the top and bottom of the plot are individual data points, where firms at the top and bottom, respectively, signed and did not sign the statement.
Figure 6. Corporate Stance on Anti-LGBTQ Legislation by Average Individual Ideology
The results from this simple bivariate analysis support the hypothesis that firms with more conservative directors and executives will be less likely to sign the HRC statement. For the most liberal firms, the estimated probability of signing the statement is around 60%, and this probability decreases more or less monotonically until reaching 0% for the most conservative firms.
To test this hypothesis more formally, I regress a dummy variable for whether a firm signed the HRC statement on the previous within-firm average individual ideology variable, a firm size control, and state and industry fixed effects. Section I of the Supplementary Material contains details about the statistical models as well as certain robustness checks, and Table A5 in the Supplementary Material tabulates the results. Consistent with the previous bivariate results, the results of this analysis support the hypothesis that firms with more conservative directors and executives will be less likely to sign the HRC statement. In each specification, the average ideology coefficient is negative and statistically significant. The coefficients are also substantively large. Depending on the specification, a one-point increase in average individual ideology in the conservative direction is associated with a 22 to 38 percentage point reduction in the probability that a firm signs the HRC statement. Considering the relatively low baseline probability of signing the statement—the intercept in the simplest model is
$ 0.23 $
, indicating that a roughly centrist firm signs the statement with 23% probability—a 22 to 38 percentage point reduction is enormous.
However, an important obstacle stands in the way of giving the foregoing results a causal interpretation: individuals may select into firms based in part on the firms’ political activities. Indeed, companies frequently highlight their records on LGBTQ issues when recruiting employees, suggesting that individuals do select into firms based in part on how much the companies support LGBTQ rights. A company’s average individual ideology may therefore be endogenous to the likelihood that the company signs the HRC statement. If so, the previous estimates may partly (or even entirely) reflect a selection effect.
To address this endogeneity concern, I use an IV design (Angrist and Pischke Reference Angrist and Pischke2009). To establish that the relationship between average individual ideology and the probability of signing the HRC statement is causal, this research design relies on an exogenous instrument that influences a firm’s average individual ideology but not the probability that the firm signs the HRC statement, except indirectly through the firm’s average individual ideology. Specifically, I adopt a “leave-one-out” strategy in which I instrument for a firm’s average individual ideology with the average for all other firms in the same industry. The theory underlying this instrument rests on the fact that individual ideology tends to vary predictably across industries. For example, corporate elites working in the energy industry tend to skew conservative, whereas corporate elites working in technology tend to skew liberal. Additionally, in hiring new individuals, firms are often constrained to hire those working in the same industry. Together, these stylized facts suggest that the average individual ideology of other firms in the same industry will influence a firm’s own average individual ideology.
Section I of the Supplementary Material contains details about the statistical models used to implement the IV design and reports the results, as well as certain diagnostics and robustness checks. In short, the IV results support the hypothesis that conservative average individual ideology decreases the probability that a firm signs the HRC statement. The key second-stage coefficient is negative, statistically significant, and substantively large, with a one-point increase in average individual ideology in the conservative direction decreasing the probability that a firm signs the HRC statement by 76 percentage points. These results are robust to the introduction of state fixed effects and alternative inferential methods. Of course, as with any IV design, an exclusion restriction is necessary to give these results a causal interpretation. In this context, that means that average individual ideology for other firms in the same industry cannot directly influence the likelihood that a firm signs the HRC statement. As discussed in the Supplementary Material, this restriction could be violated if, for example, there are peer effects driving firms to sign the statement, with companies following the lead of others in the same industry. To the extent that such peer effects exist, individual ideology would still be driving firms to the sign the statement—the instrument is based on ideology, after all—but the point estimates may combine the ideological effects of the firm’s own directors and executives and those of other firms in the same industry. It is important to keep this possibility in mind when interpreting the substantive magnitude of the IV coefficients.
DISCUSSION
This article has documented important shifts in the political landscape of corporate America over the first two decades of the new millennium. Between 2001 and 2022, the average observed ideology of the corporate elites in my sample moved meaningfully to the left, starting from a modest conservative skew and ending close to the center. This descriptive finding contrasts sharply with previous research, which relied on earlier periods or different data to find a pronounced conservative skew (Bonica Reference Bonica2016; Cohen et al. Reference Cohen, Hazan, Tallarita and Weiss2019; Fos, Kempf, and Tsoutsoura Reference Fos, Kempf and Tsoutsoura2023). But I also find significant heterogeneity across roles, industries, and firms, reflecting an ideologically fractured landscape.
These developments may help to explain observed changes in the ways that firms have engaged with the political system in recent years. A number of observers have remarked that firms have increasingly taken progressive stances on various social and political issues (e.g., Barari Reference Barari2024). Some critics have attributed this shift to insincere, profit-driven motivations, reflecting an attempt by companies to fool or distract consumers, politicians, or both (Douthat Reference Douthat2018; Ramaswamy Reference Ramaswamy2021).Footnote 20 My findings suggest a different explanation for this phenomenon: taking progressive stances on social and political issues may satisfy demand from stakeholders, such as mid-level executives who now skew left. My analysis of corporate positions on LGBTQ-related legislation provides evidence consistent with this explanation.Footnote 21 To the extent this is true—and thus firm-level political activity reflects the sincere preferences of firm stakeholders—so-called woke capitalism may not be as empty as its critics claim. An important task for future research will be to study whether and how stakeholder preferences influence corporate political activity in other domains, such as PAC giving (Li Reference Li2018) or lobbying on more traditional economic issues, which may be harder cases.Footnote 22
Of course, the possibility that corporate elite preferences are driving firm-level political activity raises additional positive and normative questions about whether corporate involvement in politics is good for shareholders, society, or both, as well as enduring questions about the purpose of the corporation (e.g., Berle Reference Berle1931; Dodd Reference Dodd1932; Friedman Reference Friedman1970; Hart and Zingales Reference Hart and Zingales2017; Lund and Pollman Reference Lund and Pollman2023; Strine and Walter Reference Strine and Walter2015). The perceived growth in corporate political activity in recent years has prompted renewed scholarly interest in these longstanding issues (e.g., Fan Reference Fan2019; Fisch and Schwartz Reference Fisch and Schwartz2024; Lin Reference Lin2018). I leave a deeper examination of these questions to future research.
The potential role of increased gender and racial diversity in pushing corporate America to the left also has implications for the political conflict over diversity, equity, and inclusion. To the extent that increased gender and racial diversity indeed shifts average ideology to the left, this may create political incentives for each side to support or subvert corporate diversity programs, with liberal and conservative politicians, respectively, benefiting from increased and decreased diversity. But the null results of the analysis of the impact of California’s gender diversity mandate—despite increasing the representation of women on boards, the law does not appear to have moved average director ideology—complicates this story. One possible explanation for the lack of a political impact from the law may be that incumbent boards strategically nominate new directors with a view to maintaining an ideologically diverse group of directors with connections to politicians on both sides of the aisle, muting any impact on average ideology. Such an explanation would be consistent with the finding that the mandate appears to have shifted average ideology for men to the right.
Taking a broader perspective, the transformation documented in this article has profound implications for American politics. Corporate elites have traditionally been a key source of support for the Republican Party, but my findings suggest that they are now increasingly ideologically fractured and in some industries may even be more aligned with Democrats. These changes have also coincided with broader shifts in political behavior, especially among whites, as more educated voters have increasingly supported Democrats over Republicans (Grossmann and Hopkins Reference Grossmann and Hopkins2024; Sides, Tesler, and Vavreck Reference Sides, Tesler and Vavreck2018) and “knowledge economy” professionals have similarly moved to the left (Short Reference Short2024). Indeed, given that about half of the leftward shift in average ideology in my sample cannot be attributed to discrete changes in the distribution of individual characteristics, such broader changes may explain a fair amount of the leftward shift among corporate elites. Ultimately, in altering both the interest group environment and the voter bases, these shifts could interact to fundamentally reshape the political coalitions supporting each party.
To the extent that such coalitional changes can also prompt changes in the policies championed by each party (Mayhew Reference Mayhew2002), this could fundamentally alter the policy environment as well. For example, survey evidence has shown that technology entrepreneurs tend to be fairly conservative on regulatory policy, even though they are liberal on redistributive economic policy and social issues (Broockman, Ferenstein, and Malhotra Reference Broockman, Ferenstein and Malhotra2019), raising questions about whether the leftward shift in the technology industry could push Democrats to the right on regulatory policy. Going forward, researchers should be attentive to the potential policy effects of these shifting coalitions.
SUPPLEMENTARY MATERIALS
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055425100993.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/ES0YIJ. Limitations on data availability are discussed in Section A of the Supplementary Material.
ACKNOWLEDGEMENTS
For helpful comments and suggestions, I am grateful to Ryan Bubb, Haodi Dong, Jeff Gordon, Gleason Judd, Nolan McCarty, Leo E. Strine Jr., audiences at Columbia Law School, INSEAD, and Princeton University, two anonymous referees, and the editor, Andrew Eggers.
CONFLICT OF INTEREST
The author declares no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The author affirms this research did not involve human participants.
Open access
Comments
No Comments have been published for this article.