The rapid growth of social media over the past two decades has presented firms with a significant challenge: Users of social media can share and disseminate damaging information with potentially adverse consequences. Prominent examples that went viral include accounts of racist and sexist treatment of employees at Walmart, unsafe working conditions at Amazon, unfair pay practices at Chipotle, and environmental violations at Nestlé. As noted by social psychologist Takuya Sawaoka, “[t]he internet now allows […] thousands of people to participate in collective […] [monitoring], in a way that wasn’t possible before” (Meinch, 2021). These anecdotes suggest an important, yet unexplored, role for social media in monitoring corporate misconduct. In a new study, we examine whether and to what extent social media activity reduces corporate misconduct.

That examination poses several empirical challenges. Most important, many forms of social media activity are unobservable and endogenous to misconduct. We attempt to overcome these challenges through various complementary empirical approaches. Our primary methodology follows Guriev et al. (2021) and exploits the staggered introduction of third-generation (3G) mobile broadband networks across the United States to identify exogenous increases in social media activity. 3G access provides users with faster data transfers and the ability to share content on their mobile devices and has been a key driver of the rapid growth in the use of Twitter, Facebook, YouTube, and other applications. We thus examine whether 3G access is associated with reductions in local misconduct levels.

While focusing on 3G alleviates many endogeneity concerns, it admittedly provides a less direct link between social media activity and misconduct. Consequently, we also use Twitter as a setting for a rich set of additional analyses to strengthen the link between social media activity and misconduct. In particular, we utilize detailed data from Twitter to both validate the 3G instrument and assess the direct association between Tweet volume and misconduct. We also examine the effects of exogenous increases of Twitter activity following the 2007 South by Southwest (SXSW) festival, which has been shown to drive growth on the platform (Müller and Schwarz, 2020; Fujiwara et al., 2021). The Twitter setting is not without its limitations, as it represents activity on only one social media platform, thus sacrificing external validity. We therefore view our various empirical strategies as complementary and as facing orthogonal limitations.

The final notable feature of our empirical framework is the misconduct data we use from Violation Tracker. These data comprise a wide set of violations and penalties related to issues that are regularly exposed on social media, including financial issues as well as non-financial issues related to workplace safety, employee discrimination, labor relations, and environmental violations.

Our main results show that 3G access leads to a decline in both penalties and violations in the three years after a 3G rollout. As for economic significance, our estimates indicate that 3G access decreases facility-level penalties by nearly 13 percent and the number of violations by 1.8 percent. These estimates likely represent a lower-bound effect because they are based only on observed misconduct and do not capture reputational costs. Overall, these findings are consistent with the view that social media is an effective monitor of corporate misconduct.

Having established a robust relationship between 3G access and misconduct, we next introduce a set of analyses relying on Twitter data to better establish the social media mechanism underpinning our findings. These analyses help support our claim that 3G access is an appropriate proxy for increased social media activity and allow us to assess whether our effects vary meaningfully with social media activity. We use data from Twitter as it is one of the most popular social media applications, with more than 330 million monthly active users. In addition, Twitter is one of the only social media platforms that provide “geotags.” These data allow us to obtain the precise geographical origin of Tweets as of 2010 and thus map Tweets directly to local 3G rollouts and facility-level misconduct.

In our first analysis, we examine how Twitter activity changes in a locality following the introduction of 3G. This is an important step in our analyses as it is not clear whether 3G should have a meaningful impact on social media activity in our setting (compared with Guriev et al.’s (2021) international setting) given the prevalence of home broadband internet in the U.S. We examine regressions of the number of Tweets in a zip code on our 3G indicator and include zip code and year fixed effects, making this test akin to a generalized DiD model. Our results indicate a sharp increase in Tweets in a zip code following 3G access. This evidence validates a necessary underlying assumption in our earlier analysis as it establishes that 3G access does indeed increase social media activity.

Next, we attempt to establish a more direct link between social media activity and misconduct. As discussed above, our primary analyses use 3G access as an instrument, given that prior research provides a precedent for considering 3G as a plausibly exogenous shock to social media activity. To support the underlying social media mechanism, we repeat our main analysis and use an indicator for high social media activity (defined as above the median number of Tweets in a zip code) as an alternative treatment variable. We find that facilities located in zip codes with an above median number of Tweets are associated with reduced misconduct levels. This result further supports the argument that social media activity helps curtail misconduct.

The Twitter setting also provides an opportunity to consider an alternative identification strategy. Following Fujiwara et al. (2021) and Müller and Schwarz (2020), we use the SXSW festival as a shock that spurred the use of Twitter across the U.S. We find that facilities located in counties with interest in SXSW and that experienced an increase in the number of Twitter users following the festival (but not before the festival) significantly reduce misconduct compared with facilities located in counties with interest in SXSW that did not experience an increase in Twitter users following the festival.

Our evidence thus far suggests that social media activity is associated with reductions in misconduct. We next conduct a set of cross-sectional analyses that consider whether the effects of 3G vary based on firm visibility. In the case of social media, more visible firms are expected to respond more to 3G as social media content will be consumed by more viewers across the country. As per Becker (1968), increased viewership should be associated with higher expected costs, as it increases the chance of negative content going viral and generating reputational damage or attracting regulatory scrutiny. We use three proxies to capture firm visibility: firm size, firm Twitter followers, and firm media coverage. Consistent with our expectation, we find that 3G reduces misconduct more for facilities of large firms, facilities of firms with higher levels of Twitter followers, and facilities of firms with more media coverage. Our results indicate a significant and robust relationship between 3G rollout and facility-level misconduct and suggest an important role for social media in monitoring firm misconduct.

Our results can be of interest to policymakers and regulators concerned about the role of social media and corporate misconduct. While the current public debate primarily focuses on negative effects of social media, such as the spread of “fake news” and disinformation, our findings highlight a positive effect: its ability to reduce corporate misconduct by, in part, empowering citizens to monitor firms’ behavior.

This post comes to us from professors Jonas Heese and Joseph Pacelli at Harvard Business School. It is based on their recent paper, “The Monitoring Role of Social Media,” available here.