Customers’ response to firms’ disclosure of social stances: evidence from voting reform laws

For example, Senate Minority Leader Mitchell McConnell recently stated that corporate CEOs should “stay out of politics” and not “pick sides” in these discussions (Cowan 2021).

We use Black rather than African-American in our study to be more inclusive. Black includes individuals who are from other countries/continents who may not identify as African-American.

Exerting economic influence (e.g., relocating business activities, altering products or offerings, etc.) requires capital investment or organizational changes that may also relate to corporate performance or strategic objectives. Our study focuses on raising awareness and provides a cleaner setting to examine just the ESG activity.

For example, the clash between Florida Republican leaders and Disney has put CEOs on alert regarding the consequences of their firms’ social positions (Cutter and Glazer 2022). See “CEO activism in America is risky business” published by The Economist in April 2021 (https://​www.​economist.​com/​business/​2021/​04/​14/​ceo-activism-in-america-is-risky-business).

Appendix 1 of Jin et al. (2022) describes SafeGraph’s approach to measuring customer traffic, including adjustments to improve accuracy and some limitations of the data. While we believe the trends in these data represent the overall population, the raw numbers are understated, so we describe the economic magnitude of our findings in percentage form throughout the paper.

The list of all SpeakOut firms (including those not used in our study) and non-SpeakOut brands is available from the authors upon request.

To ensure we can identify appropriate non-SpeakOut stores, we manually examine all brands for all SpeakOut firms. We exclude the following brands from SpeakOut firms that we view as not having a good match in SafeGraph. The brands excluded are Amazon 4-Star, Amazon Books, Amazon Distribution, Amazon Fresh, Amazon Go, and Target Distribution.

While firms’ decisions to speak out against the Georgia voting reform law is probably not exogenous at the firm level, this decision is unlikely to be correlated with changes in short-term management decisions at the individual store-location level.

Since the dependent variables are log transformed, we calculate the economic magnitude as (1 – e^coefficient).

Recent research on difference-in-differences methods cautions against using statistical tests to assess the plausibility of the parallel trends assumption (Bilinski and Hatfield 2018; Roth 2022; Rambachan and Roth 2023). Traditional tests of pre-trends face power issues that can lead to Type I or Type II errors (Bilinski and Hatfield 2018). To empirically assess how potential violations of the parallel trends assumption could impact our inferences, we follow the method in Bilinski and Hatfield (2018) and Hou and Poliquin (2023). We include a linear trend variable, TimeTrend, to our baseline model and interact it with SpeakOut. The TimeTrend variable equals to one, two, three, and four for November, December, January, and February (the four months in the pre-period), respectively, and zero otherwise. The coefficient on SpeakOut*TimeTrend is negatively significant, suggesting that there could be a pre-trend between SpeakOut stores and non-SpeakOut stores. However, the coefficients of interest and the average treatment effect across the post-period coefficients remain economically and statistically significant and provide the same inferences (and in some cases stronger inferences) than results from the model without the time trend. These findings suggest that a pre-trend has little to no impact our inferences.

We classify news outlets as conservative leaning based on The Media Bias Chart at https://​adfontesmedia.​com/​. We concentrate on sources that are categorized as skews right, hyper-partisan right, or most extreme right and offers analysis or opinions. Conservative leaning sources are Fox News, New York Post, Breitbart News, TheBlaze, Washington Times, NewsMax.​com, and WashingtonExamin​er.​com.

To ensure that our results stem from mentions in conservative leaning media, we also include an indicator variable (and the applicable interaction terms) that equals one if a firm is mentioned by one of 34 mainstream media outlets, such as ABC, NBC, CBS, USA Today, and Yahoo! News. We continue to find a negative and significant (p < 0.01) coefficient on SpeakOut_i × Post × ConservativeMedia_i,l for daily visits, monthly visits, and unique visitors.

We construct our demographic indicator variables (BlackPopHigh, EducPopHigh, and IncomeHigh) using the median income, education, and Black population across all counties in our sample, instead of the median of all observations in our sample. Stores tend to locate in larger cities, where income, education, and diversity tend to be higher. Also, larger cities have more stores than smaller cities in our sample. Therefore, the mean of BlackPopHigh, EducPopHigh, and IncomeHigh is higher than 0.5.

The spending data are available for just under 50% of the observations in our full sample. In addition, the correlation between SafeGraph’s spending data for Target, Chipotle, and McDonald’s and reported quarterly earnings was 0.83, 0.77, and 0.89, respectively (Doan 2021).

The second and third quarters of 2020 overlapped with the outbreak of COVID-19. While we cannot rule out the possibility that COVID-19 might contaminate our inferences, the pandemic affected both SpeakOut and non-SpeakOut stores.

We require observations to have nonmissing firm characteristics from Compustat and nonmissing stock return data from CRSP. Therefore, the number of firms included in this analysis is smaller than the main analyses, where we only require nonmissing SafeGraph data.