Racially Disparate Policy Responses to Mass Shootings

Racial Resentment and Gun Politics

Throughout American history, the relationship between gun policy and race has been intricate and evolving. In the past, White Americans could protect their right to bear arms while simultaneously disarming people of color, whom they perceived as threats to their lives and property (Cramer 1994; Satia 2019; Tahmassebi 1991). This strategy maximized White safety and rights but subjugated people of color. More recently, greater federal enforcement of the equal protection clause and changing social norms prevent the implementation of explicitly racist gun control laws. Still, racial sentiments shape gun policy, and I propose the following hypotheses, which I will proceed to defend:

• Hypothesis 1: White mass shooting fatalities lead to the implementation of new restrictive state gun laws.

After the Civil Rights Movement, explicitly racist beliefs based on ideas of innate racial superiority were labeled immoral and explicitly racist policies became illegal. However, research shows that race continues to shape politics through “symbolic racism” and racial resentment (Kinder and Sanders 1996; Sears 1988). This line of research argues that beliefs about race have become interconnected with ideals of individualism and traditional American values (Kinder and Sanders 1996; Sears 1988). For example, Whites with high levels of racial resentment believe that higher incidences of Black poverty and crime are due to individual and cultural failures, not structural or institutional factors (Kinder and Sanders 1996; Sears 1988). ¹ In turn, White Americans with latent racial prejudices gravitate towards positions presented as advancing individual liberties because they are implicitly perceived as self-beneficial and studies show that racial resentment strongly influences White opposition to policies perceived as primarily benefiting racial and ethnic minorities (Kinder and Sanders 1996; Kinder and Sears 1981; Mendelberg 2001; Sears 1988).

In the 1970s, gun rights movements engraved racial resentment into the politics of gun control. While organizations like the NRA supported gun control laws in the 1960s, including the Mulford Act and the Gun Control Act of 1968, a deep shift occurred in the late 1970s and the NRA adopted a language of rights and liberties to advocate for uncompromising gun rights (Filindra & Kaplan, 2016; R. B. Siegel, 2008). Simultaneously, the NRA and its members aligned themselves with the broader “New Right” movement that was categorized by White resistance to racially egalitarian policies (Filindra & Kaplan, 2016; R. B. Siegel, 2008). Today, racial resentment is a strong predictor of pro-gun attitudes even when controlling for confounds like conservative values and authoritarian inclinations (Filindra and Kaplan 2016). ²

Following the Civil Rights Movement and particularly after the rise in urban violence in the 1980s, pro-gun movements and gun manufacturers began framing guns as a tool of self-defense rather than one for hunting and sport (Yamane, Yamane, and Ivory 2020). The propagation of negative stereotypes, racially biased media frames, and racist dog-whistles have framed gun control measures as a threat to White safety, preventing “law-abiding gun owners” (i.e., White Americans) from owning guns but doing nothing to prevent “criminals” (i.e., Black Americans) from acquiring weapons (Parham-Payne 2014; Schutten et al. 2022). Furthermore, significant racial disparities in gun violence victimization rates facilitated right-wing narratives that paint racial minorities as the primary beneficiaries of gun control and White Americans as those most burdened by them. ³ The gun rights movement advances narratives that make guns a symbol of liberty and label gun control advocates, including “undeserving” minorities, enemies of freedom seeking “special” state protection from violence (Filindra and Kaplan 2016).

These narratives frame gun control measures as benefiting “undeserving” groups and burdening “deserving” groups (Kreitzer and Smith 2018; Schneider and Ingram 1993; Schutten et al. 2022). Powered by effective public and private lobbying campaigns, these frames became engraved in the collective consciousness. Therefore, policymakers are unwilling to pass laws that they and their constituents believe will benefit an “undeserving” and non-powerful group (racial minorities) while burdening a “deserving” and politically powerful group (White gun owners) (Kreitzer and Smith 2018; Schneider and Ingram 1993).

Threat to White Lives and Narrative Change

Theories of punctuated equilibrium suggest that mass shootings, a relatively uncommon but salient form of gun violence, may lead to stricter gun laws by serving as focusing events. Focusing events, like mass shootings, provide an urgent, symbol-rich example of purported policy failures, opening policy windows where policy change is more likely (Baumgartner, Jones, and Mortensen 2018; Birkland 1997). But mass shootings disproportionately affecting racial and ethnic minorities fail to significantly shift deeply engraved collective narratives regarding “targeted groups.” In the aftermath of a mass shooting that mostly kills people of color, gun control efforts are still seen as primarily benefiting racial minorities and burdening “law-abiding” White Americans. Unless those shootings shift social perceptions of racial minorities, we have limited reasons to expect a strong and compassionate policy response (Kreitzer and Smith 2018; Schneider and Ingram 1993).

However, mass shootings that disproportionately impact White Americans can (temporarily) change gun control narratives. Research shows that Republicans are more likely to support restrictive gun laws when framed as a public safety issue instead of an individual rights issue (Haider-Markel and Joslyn 2001), and proximal exposure to mass shootings increases support for restrictive gun laws (Newman and Hartman 2019). However, perceptions of victims affect these responses because they change who is seen as the primary beneficiaries of a policy. Experimental research finds that White Americans are more likely to support gun control when they are primed to think about White people as gun violence victims versus Black people (Berryessa, Sierra-Arevalo, and Semenza 2022; Walker, Collingwood, and Bunyasi 2020). These patterns parallel racialized responses to other crises where White Americans, especially those with high racial resentment, support government interventions only when the perceived beneficiaries are White (Fong and Luttmer 2009; Iyengar and Hahn 2007; Lee, Seo, and Leahey 2022; Stephens-Dougan 2023). This research suggests that mass shootings will increase public support for restrictive gun laws when victims are White but not racial and ethnic minorities.

Beyond Public Opinion

The extensive and rich literature cited above illustrates how gun policy preferences are shaped by racial resentment and perceptions of what groups benefit from gun control. However, it is focused on public opinion—particularly White public opinion. And, while we expect public opinion—particularly White public opinion—to drive policy decisions, we know this is not always the case (Lax and Phillips 2012). State gun policies are particularly incongruent with public opinion, and mass shootings do not appear to increase voter turnout or affect vote choice (Hassell and Holbein forthcoming; Lax and Phillips 2012). Gun policy is largely shaped by legislators' personal preferences and interest group lobbying efforts (Goss 2010; Spitzer 2023).

This is not to say that legislators do not care about what constituents want in terms of gun policy, but that mass shooting-induced changes in public opinion alone are unlikely to lead to more restrictive gun laws as they do not translate to significant shifts in electoral behavior (Hassell and Holbein forthcoming; Lax and Phillips 2012; Newman and Hartman 2019). However, legislators are overwhelmingly White, their behavior is shaped by perceptions of targeted groups, and implicit racial biases lead them to discriminate against people of color even in low-cost situations (Butler 2014; Gell-Redman et al. 2018; Hansen and Clark 2020; Schneider and Ingram 1993). Therefore, I contend that mass shootings that disproportionately impact White Americans are more likely to increase legislators' willingness to pass gun restrictions than those that disproportionately impact racial and ethnic minorities.

Racial biases in legislators' responses to mass shootings do not just matter because they shape legislators' willingness to vote in favor of gun control legislation; legislators' immediate rhetorical responses to mass shootings affect the post-shooting political environment. If legislators are more likely to discuss gun violence and politicize the issue more often after a specific type of mass shooting, this rhetoric shapes legislative agendas, influences fellow legislators, and invites interest group lobbying and donations (Canes-Wrone 2001; Grimmer 2013; Iliev 2021).

I propose and test the following hypotheses:

• Hypothesis 3: Legislators are more likely to discuss gun violence in response to White mass shootings fatalities but not REM fatalities.

○ Hypothesis 3B: Legislators will discuss gun violence in more politicized ways in response to White mass shooting fatalities but not REM fatalities.

Signals from legislators illustrating an openness to address state gun laws are likely to shape how interest groups behave (Iliev 2021; Ray 2018). If interest groups believe legislators will address gun policy soon, they are more likely to compete for influence through lobbying and political donations. However, these groups are also shaped by the preferences of their leaders and members, who themselves may be prone to racial biases. The NRA and other gun rights groups are known to foster racial resentment as part of their policy strategy (Filindra 2023; Filindra, Kaplan, and Buyuker 2021; Schutten et al. 2022). So, resistance to efforts that are perceived to benefit White Americans may alienate members. And, while pro-gun control interest group leaders and members are less likely to harbor racial resentment, these organizations have been accused of internally marginalizing the voices of people of color (Clayton 2023; Reed et al. 2023). Therefore, gun policy interest groups might respond to mass shootings in racially biased ways for two reasons: 1) because they are responding rationally to shifts in public opinion and legislators' public statements, and 2) because of latent racial biases among their leaders and members. Given that gun policy is heavily influenced by interest group lobbying, divergent interest group responses may also shape post-shooting policy responses (Goss 2010; Spitzer 2023).

I propose and test the following hypothesis:

• Hypothesis 4: Gun control interest groups are more likely to lobby state legislatures in response to White mass shooting fatalities but not REM fatalities.

Victims' Race and Ethnicity Shapes State Gun Policy Change: Data

To assess potential racial bias in state policy responsiveness, this study analyzes data on mass shootings primarily obtained from The Violence Project, a nonpartisan research center funded by the National Institute of Justice. Their comprehensive coding of victims' race and ethnicity, based on a rigorous multi-coder process, is a reliable source for the analysis. The definition of mass shootings utilized by The Violence Project follows that of the Congressional Research Service, requiring a multiple homicide incident with four or more victims murdered with firearms within one event, occurring in public locations or nearby settings, and not attributable to other criminal activity or commonplace circumstances. While this definition is relatively conservative, it best captures the salient public events that serve as potential focusing events.

From 1990 to 2020, there were 137 mass shootings across 38 states. There are 973 victims across these mass shootings. Victims' racial and ethnic information primarily comes from The Violence Project. As a verification check, I hand-coded the race and ethnicity of 475 mass shooting victims between 2010 and 2019 using pictures found online in news reports and obituaries. My coding matches The Violence Project's coding in 91% of cases. The Violence Project is missing race and ethnicity data for 114 victims, mainly from older years when less media information was available. I use Bayesian Improved Surname Geocoding (BISG) to predict the race and ethnicity of the remaining victims ⁴ and aggregate individual racial and ethnic minority fatalities into a broader umbrella category. ⁵ Six hundred thirty-three victims are White, and 340 are racial and ethnic minorities.

The main independent variables in this study are mass shooting fatalities disaggregated by victims' race and ethnicity (in these broader buckets) at the state-year level. ⁶ Figure 1 plots mass shooting fatalities by victims' aggregated race and ethnicity over time, showing that mass shooting fatalities have increased over time, though the racial and ethnic distribution is relatively constant.OPEN IN VIEWER

To measure the dependent variable, I use state panel data from the State Firearm Laws Project (Siegel et al., 2017). This dataset provides a well-established measurement of gun policy, minimizing the risk of researcher-induced measurement bias. The State Firearm Laws Project utilizes Thomson Reuters Westlaw data to track historical state statutes and codes the presence or absence of 133 firearm-related laws in all states over the 30-year period. The coding involves a dichotomous approach, where a value of 1 indicates a more restrictive stance on a specific policy, and 0 indicates a less restrictive stance. Table 1 provides a summary of the broad gun policy areas tracked by the State Firearm Laws Project, along with the specific number of policies within each domain. As a robustness test, an alternative dataset, the RAND State Firearm Law Database (Cherney et al., 2018), is utilized to construct a similarly structured dependent variable. The results from the main study remain largely robust when this alternative dataset is used (see SI-D).

OPEN IN VIEWER

Table 1.

State Firearm Laws Policy Areas.

Policy Area	Number of Policies
Ammunition Regulations	7
Assault weapons and large-capacity magazines	8
Background checks	11
Buyer Regulations	17
Child access prevention	11
Concealed carry permitting	7
Dealer Regulations	17
Domestic violence	21
Gun trafficking	7
Immunity	1
Possession Regulations	12
Preemption	3
Prohibitions for high-risk gun possession	10
Stand your ground	1

The primary dependent variable is the number of new restrictive gun laws implemented in time T AND T+1. In other words, if a shooting happened in State A in 2018, how many restrictive firearm laws does State A implement in 2018 and 2019? ⁷ This measurement recognizes fluidity in the policy change process, capturing policy changes during the year of the mass shooting and the year after. While some folks may take issue with this modeling strategy, the findings are robust to wide set of modeling specifications, including 1) setting the dependent variable to the total number of gun laws at time T+1 and including a lagged dependent variable to capture the dynamic effects (See SI-E), 2) setting the dependent variable to the total number of gun laws at time T+1 and the main independent variables to cumulative fatalities over time (See SI-F), and 3) setting the dependent variable to total number of gun laws at time T and the main independent variables to the number of fatalities over the last two years (See SI-G).

State partisan control, whether a state is under unified Democratic control, unified Republican control, or divided partisan control, ⁸ is used as a control and a moderating variable in robustness checks. ⁹ State partisan control is included as a covariate because gun policy is a particularly partisan issue, partisan control of government is closely linked with the direction of policy change, and state partisan control is a time-variant state-specific confound (Oliphant, 2017; Caughey, Xu, and Warshaw, 2017).

Figure 2 plots the number of restrictive gun laws over time by state with smoothed trends by partisan control. ¹⁰ The distribution of firearm laws follows logical partisan divisions. States under unified Democratic control have the most restrictive gun laws, and states under unified Republican control have the fewest restrictive gun laws. One important observation is that the gap between Democratic and Republican-controlled states has grown. To account for this, I add state-specific time trends to the main models as a robustness check.OPEN IN VIEWER

Victims' Race and Ethnicity Shapes State Gun Policy Change: Methods

This study primarily uses a “non-traditional” staggered difference-in-differences (DID) design. Models have a dual “dosed treatment” (White fatalities and REM fatalities) that are contemporaneous (treatment reversal is inherent). This method estimates the dynamic effect of mass shooting fatalities on the implementation of new restrictive gun laws by comparing states with and without shootings at different levels of deadliness while controlling for state characteristics and temporal effects using two-way fixed effects (TWFE).

While some literature supports the potential outcomes framework, indicating that predictors of mass shootings are individual-level attributes of shooters, not location-specific characteristics (Madfis, 2017; Paolini, 2015; Wike & Fraser, 2009), recent work suggests that mass shootings may be correlated with some relevant political variables at the county level (Hassell and Holbein forthcoming), potentially violating the parallel trends assumptions necessary for well-defined difference-in-differences designs. I take various steps to address this issue and show models are robust to a wide set of specifications.

In the main analysis, I present three models addressing confounds in increasingly rigorous ways. One model specification uses only state and year-fixed effects. This model shows that the results are not an artifact of specific hand-picked control variables, which may bias results (Achen 2005). In another model, I include a series of time-variant state-specific controls. ¹¹ In the third model, I add state-specific linear time trends to control for the state-level underlying trends shaping variation in the outcome variable over time, as suggested by scholars engaged in similar research (Hassell and Holbein forthcoming). This helps capture the state-specific effects and ensure that any observed changes in the outcome are not due to the underlying trends but rather to the treatment (in this case, the mass shooting fatalities).

Victims' Race and Ethnicity Shape State Gun Policy Change: Findings

The results in Figure 3 consistently support Hypotheses 1 and 2. ¹² Across all three models in Figure 3, White mass shooting fatalities have a positive and statistically significant effect on the number of new restrictive gun laws a state implements over the next two years, supporting Hypothesis 1. Model 1, which relies only on two-way fixed effects, suggests that each White mass shooting fatality predicts 0.154 additional gun laws over the next two years (p < 0.01). On average, a state with ten White mass shooting fatalities responds by implementing 1.54 new gun laws. When time-variant state-specific controls are added in Model 2, the model predicts that ten White mass shooting fatalities will lead to 1.25 new restrictive gun laws (p < 0.01). While the estimates are slightly smaller, they are largely consistent and robust when state-specific linear time trends are added in Model 3, suggesting that ten White mass shooting fatalities are associated with about one new restrictive gun law (p = 0.01).OPEN IN VIEWER

On the other hand, Figure 3 suggests that REM mass shooting fatalities have a negative effect on new restrictive gun laws, though this finding is not statistically significant across all three models, lending support to Hypothesis 2. Model 1 implies that ten REM mass shooting fatalities predict about 0.27 fewer restrictive gun laws, though the effects are not statistically significant (p = 0.25). Models 2 and 3 indicate that ten REM mass shooting fatalities predict about 0.65 fewer restrictive gun laws and predicted effects are statistically significant (p < 0.01). Importantly, robustness checks show that the difference in effects between White fatalities and REM fatalities are statistically significant (See SI-C). However, within these models and robustness checks included in the appendix, the coefficient for REM mass shooting fatalities is sometimes not statistically significant and is dependent on model specification (but it is always negative when statistically significant). Furthermore, the next analysis also questions whether this latter finding is driven by a violation of the parallel trends assumption.

Bias in Difference-in-Differences Estimates

Recent scholarly work has raised concerns about estimated average treatment effects in staggered DID models due to potential biases caused by "forbidden comparisons" (Callaway and Sant’Anna 2021; Liu, Wang, and Xu 2022). ¹³ Existing econometric tools usually only address this issue for a limited set of "traditional" DID models with a single bivariate treatment without reversal. However, the models presented in Figure 3 include two continuous treatments (Number of White Fatalities; Number of REM Fatalities) that involve treatment reversal. Unfortunately, to the best of my knowledge, no econometric tool developed thus far can deal with all of these non-standard applications simultaneously. However, Liu et al. (2022) developed an estimation technique to handle treatment reversal if using a single bivariate treatment. Although this technique does not cover all the non-standard applications used in the previous models, I can adjust the data and employ two separate models to estimate the effect of majority White mass shootings and majority REM mass shootings individually, testing the robustness of the main findings in Figure 3 using a different independent variable but without the risk of TWFE-induced biases. To do so, I measure whether a state had one or more majority White mass shootings or one or more majority REM mass shootings. The dataset includes 77 majority White shootings and 38 majority-minority shootings.

I begin by testing the effects of majority White mass shootings on new restrictive gun laws in Figure 4, employing the bias-robust estimator developed by Liu et al. (2022). This model estimates the average treatment effect of a majority White mass shooting on the number of new restrictive gun laws implemented during the year of the shooting and the following year by imputing counterfactual outcomes for treated observations (Liu et al., 2022). Additionally, I conduct a placebo test, comparing the post-shooting effect to pre-treatment trends. The results from this analysis are robust, indicating that a majority White mass shooting leads states to implement approximately one additional restrictive gun law during the year of the shooting or the subsequent year (p = 0.03). The placebo test confirms that pre-treatment trends do not explain the post-shooting effect, and the results hold up well when unit-specific linear time trends are added to the model (ATT = 0.75; p = 0.07).OPEN IN VIEWER

Next, I examine the effects of majority REM mass shootings on new restrictive gun laws in Figure 5, employing the same estimator and placebo test. The results from the previous model are not as robust when using this estimation strategy, although it is crucial to recognize that both the estimation strategy and the measurement of the independent variable change. Figure 5 indicates that a majority REM mass shooting leads a state to implement 0.5 fewer restrictive gun laws over the next two years, but this effect is not statistically significant (p = 0.48). It is essential to note that this estimate may be influenced by pre-shooting trends, which indicate that states tend to loosen gun laws in the years immediately preceding majority REM mass shootings. The effects do not substantially differ when unit-specific linear time trends are included (ATT = – 0.58; p = 0.35). Given these findings, including the placebo test of pre-shooting trends, we should approach the interpretation of the results from the previous analysis with caution. It is plausible, and perhaps likely, that racial and ethnic minority mass shooting fatalities do not significantly affect new restrictive gun laws instead of having a statistically significant negative effect.OPEN IN VIEWER

Controlling for Confounding Demographic Characteristics

When considering one of the motivating examples comparing Florida's policy response to the Parkland, Florida, and Orland, Florida, shootings, victims' characteristics differ in many ways. Among them are age, sexual orientation, location, gender, and income. The Stoneman Douglas High School shooting (Parkland) happened at a school in an upper-income neighborhood, and the victims were children, while the Pulse nightclub shooting (Orlando) occurred at a nightclub in a low-income neighborhood, and the victims were primarily young LGBT adults. We might expect legislators to see children, students, women, higher income, and heterosexual individuals as more “deserving” of protection than young adults, men, low-income earners, and LGBT individuals for a wide range of reasons (Bridges 2016; Israel-Trummel and Streeter 2022; Padamsee 2020). If these characteristics are correlated with race, previous findings may be biased.

While I was unable to collect data on victims' sexual orientation, income, or wealth, I was able to collect data on victims' age, expressed gender, and location. Therefore, in Figure 6, (i) control for the number of mass shooting victims that are under 18 years old and female presenting and the number of victims killed while in school or in a high-income neighborhood (defined as in the top 1/3^rd of income tertials). ¹⁴ OPEN IN VIEWER

The findings from the main models are robust when these controls are included. Model 1 in Figure 6 controls for the number of mass shooting fatalities under 18 years old. According to this model, ten White mass shooting fatalities lead to about one additional restrictive gun law on average (p = 0.02). Controlling for the number of female victims in Model 2 suggests that ten White mass shooting fatalities lead to about 1.2 restrictive firearm laws (p = 0.24). While the coefficient is not statistically significant, this might be because of high collinearity in the independent variables, as the coefficient for female victims is small (0.004) and standard errors are large (p = 0.98). In Model 3, I control for the number of mass shooting fatalities on school campuses. This model suggests ten White mass shooting fatalities lead to about one additional restrictive gun law. Model 4 controls for the number of mass shooting fatalities in the highest tertial census tracts based on median income that year. Once again, the results are robust to this model specification, which predicts that ten White Mass shooting fatalities lead to 1.1 additional restrictive gun laws (p = 0.01). The control variables are not statistically significant in any of the models.

In comparison, all four models find that ten racial and ethnic minority mass shooting fatalities lead to about 0.6 fewer restrictive gun laws and these findings are statistically significant in three models. However, the previous analysis suggests being cautious with over emphasizing these findings because they may be partially driven by pre-shooting trends. While the point estimates for White fatalities are slightly smaller across these models than the effects observed in previous models, the findings are largely robust to this set of controls. These models show that systematic differences in victims' age, gender, or location do not drive racial disparities in state policy responses to mass shootings.

Racial Biases in Legislators' Post-Shooing Rhetoric: Data

Thus far, the evidence demonstrates that victims' race and ethnicity shapes state gun policy responses to mass shootings. Because various mechanisms may help explain these disparities, I will explore some I believe to be important. I contend that victims' race and ethnicity shapes state legislators' support for gun control post-shooting and that legislators' racially biased responses to mass shootings do not just influence their willingness to vote for restrictive gun legislation but also affect the way they communicate with the public which in turn shapes legislative agendas, public opinion, other legislators' views, and interest group behavior (Canes-Wrone 2001; Grimmer 2013; Iliev 2021).

I study legislators' post-shooting communication using Twitter. Twitter is an excellent data source for conducting this research because it is a direct source of communication between an elected official and the public, mirrors their policy agenda in other communication mediums, cannot be micro-targeted, and is used by most legislators during the period under observation (Casas & Morar 2015; Straus et al., 2013).

Using data from Butler, Kousser, and Oklobdzija (2023), which includes 3,580,727 tweets from 3399 legislators (63%) and mass shooting data from The Violence Project, I focus on mass shootings occurring between January 1st, 2012 and April 23rd, 2019, covering 40 shootings across 16 states. ¹⁵ I focus on tweets posted in the seven days before and after an in-state mass shooting. ¹⁶ In these 16 states, 826 legislators tweeted at least once in that time frame. Empty tweets or those only containing UTF-encoded objects were removed from the dataset. After subsetting tweets to the relevant dates and removing non-usable tweets, the remaining 29,188 tweets were analyzed.

I code two dependent variables. First, I code the number of tweets a legislator posted discussing gun violence using a supervised naïve Bayes learning model. ¹⁷ I then corrected false positives in the predicted dataset by hand but was unable to correct false negatives by hand given the size of the dataset. This resulted in 1532 tweets classified as discussing gun violence. ¹⁸ These are examples of tweets classified as discussing gun violence by the supervised training model:

• RT @NevadaDPS: #VegasShooting Should you decide to donate blood, here's a link to make a reservation: https://t.co/csHTbEyAOM

Next, I hand-coded whether tweets discussing gun violence used politicized frames. This process involved a research assistant and I going through tweets classified as discussing gun violence and analyzing their content. Tweets classified as using politicized frames prime gun policy or call for legislative action to reduce gun violence. These tweets often reference gun legislation or otherwise bring specific attention to the institutional causes of gun violence. ¹⁹ Some tweets call for legislative action but do not directly mention gun policy. The research assistant and I hand-coded the same set of Tweets separately to ensure intercoder reliability. Our coding matched in about 88% of cases. Instead of adjudicating between differences, I analyze both sets of codes separately. Only 538 tweets were classified as discussing gun violence using a political frame. ²⁰ Some examples of tweets classified as using a political frame include:

• Our “leaders” will say absolutely anything to avoid a real conversation about common sense gun legislation.

Table 2 summarizes the most common words used in each classification category. Words derived from "prayer" stand out in the general corpus of tweets discussing gun violence but not in the subset of politicized frames. The subset of tweets using political frame were more likely to include the words "gun,” “violence,” "safety,” “need," “must,” and “action” than the general corpus of gun violence-related tweets.

OPEN IN VIEWER

Table 2.

Ten Most Common Words in Gun Violence Tweets Overall vs Politically Framed Gun Violence Tweets.

Frame	Top 10 Words (In Order of Frequency)
All gun violence mentions	Shooting, victims, families, prayers, today, Dallas, gun, officers, praying, violence
Political frame	Gun, violence, shooting, mass, school, need, safety, must, action, families

Racial Biases in Legislators' Post-Shooing Rhetoric: Methods

I employ a pre-post within-unit design, analyzing tweets posted seven days before and seven days after an in-state mass shooting. The unit of analysis is legislator pre- or post-shooting. In other words, I aggregate the data up to the legislator-level and compare the number of tweets they posted discussing gun violence the week before an in-state mass shooting and the number they posted the week after the shooting. The dependent variables are the number of tweets discussing gun violence and the number of tweets discussing gun violence that use a political frame (a subset of the former). I investigated the independent effects of both types of victims (White and REM) using multivariate regression. Fatality counts are treated as a continuous treatment, influencing legislators in the post-shooting period with the pre-shooting period as a baseline. State-fixed effects, year-fixed effects, and controls for legislators' party identification or legislator-fixed effects and year-fixed effects are included in the models. ²¹

Figure 7 provides a visual representation of the data, showing the proportion of tweets referencing gun violence and using political framing before and after an in-state shooting. Tweets referencing gun violence peak on the day of the mass shooting and the day after, gradually decreasing over time. Around 17% of tweets address gun violence on the day of the shooting, compared to only one percent the day before. Within six days of the shooting, only three percent of tweets discuss gun violence. Approximately three percent of tweets use a political frame during the post-shooting period.OPEN IN VIEWER

Racial Biases in Legislators' Post-Shooing Rhetoric: Findings

The first set of models in Figure 8 (left panel) tests whether the number of White mass shooting fatalities and REM mass shooting fatalities affect how much legislators discuss gun violence on Twitter. ²² This analysis aims to determine whether victims' race and ethnicity impact the overall quantity of gun violence-related tweets. The coefficient for the Number of White Fatalities is positive and statistically significant. The models suggest that ten White mass shooting fatalities led legislators to post approximately one additional gun violence-related tweet, on average, in the week after the shooting compared to the week before. Similarly, the Number of REM Fatalities coefficient is also positive and statistically significant. The models indicate that ten racial and ethnic minority fatalities led legislators to post about 0.5 additional gun violence-related tweets, on average. Although both coefficients are statistically significant, the impact of racial and ethnic minority fatalities on legislators' gun violence tweets is about half that of White fatalities, and robustness checks modeling that main independent variables as an interaction indicate that the differences are statistically significant (See SI-I).OPEN IN VIEWER

The second set of models in Figure 8 (right panel) explores how the number of White and REM mass shooting fatalities affects legislators' politicized discussion of gun violence on Twitter. The Number of White Fatalities coefficient is positive and statistically significant in all three models. The models suggest that ten White mass shooting fatalities lead legislators to post about 0.3 additional tweets discussing gun violence using political frames on average the week after the shooting compared to the previous week. Although this effect may appear small, this is about 1/3^rd of a standard deviation in the pre-treatment period and the impact of racial and ethnic minority fatalities on politicized gun violence tweets is even smaller. The Number of REM Fatalities coefficient is positive but not statistically significant in any model. The models estimate that ten racial and ethnic minority mass shooting fatalities led legislators to post 0.1 more tweets discussing gun violence through a political frame the week after a shooting compared to the week before. However, when modeled as an interaction, the differences in effects are not statistically significant (See SI-I)

These findings support Hypothesis 3 but provide only mixed support for Hypothesis 3B, indicating that legislators pay more attention to gun violence after mass shootings when victims are White versus people of color. There is only suggestive evidence that legislators politicize gun violence, blaming institutional failures more after mass shootings when victims are White but not when they are racial and ethnic minorities.

Racial Biases in Interest Group Lobbying Responses: Data

I explore one more potential mechanism, gun policy interest group lobbying, because of its purported importance in gun politics (Goss 2010; Spitzer 2023). To examine whether gun policy interest groups respond to mass shootings in racially biased ways, I utilize state lobbying expenditure data from Follow the Money, covering the years 2012 through 2020. Lobbying expenditures serve as a measure of latent lobbying efforts. The dataset includes 11 states with lobbying expenditure data starting from 2012, which expanded to 19 states in 2014, encompassing states with multiple mass shootings, like California, Colorado, Florida, and Texas. Using all available data since 2012 results in an unbalanced panel of 19 states observed over seven to nine years. The lobbying data explored in this study is less complete and, therefore, less reliable than the gun policy data explored previously, so it is important to take extra care when interpreting and extrapolating from these findings.

I analyze lobbying expenditures for gun control and gun rights interest groups separately. Mass shooting data comes from The Violence Project, and the unit of analysis is state-year.

To better visualize the lobbying expenditure data, Figure 9 illustrates lobbying expenditures by interest group positions in states based on their partisan control. Interestingly, gun policy interest groups seem to lobby at higher rates in states with unified Democratic governments, and their spending fluctuates significantly from year to year within states.OPEN IN VIEWER

Like policy, mass shootings may influence interest group lobbying efforts that same year or the following year, with the timing partially affected by when the shootings occur within the calendar year. To address this possibility, the dependent variable is the total combined lobbying spending in years T and T+1, matching the primary empirical models in the first study. Therefore, the analysis estimates the effects of mass shooting fatalities by race in time T on lobbying spending that year and the following year.

Racial Biases in Interest Group Lobbying Responses: Methods

Like in the first study, I ran three sets of models. The first model specification includes only state and year-fixed effects. The second model incorporates a similar series of time-variant state-specific controls. In the third model, I add state-specific linear time trends. Results are largely robust when using new DID estimators developed by Liu et al. (2022) (See SI-J).

Racial Biases in Interest Group Post-Shooting Lobbying Response: Findings

The first set of models in Figure 10 (left panel) tests the effect of White fatalities and REM fatalities on gun control interest groups' lobbying spending. ²³ These findings lend support to Hypothesis 4. White mass shooting fatalities have a consistent positive and statistically significant effect on gun control interest group lobbying spending, while REM fatalities do not. The models predict that ten White mass shooting fatalities lead gun control interest groups to spend between $61,000 and $98,000 more lobbying the state over the next two years. The results are robust to new DID estimators (See SI-J).OPEN IN VIEWER

The Number of REM Fatalities coefficients are negative and nearly statistically significant in all three models. These models suggest that ten racial and ethnic minority fatalities lead gun control interest groups to spend between $15,000 and $30,000 less over the next two years lobbying the state. However, this latter finding falls short of conventional levels of statistical significance and is not robust to new DID estimators (See SI-J).

The second set of models in Figure 10 (right panel) tests the effect of White mass shooting fatalities and racial and ethnic minority mass shooting fatalities on gun rights interest groups' lobbying spending. These findings do not support Hypothesis 5. The Number of White Fatalities coefficient is generally small and not statistically significant across all three models. Similarly, REM mass shooting fatalities do not appear to impact gun rights interest groups' lobbying spending over the next two years.

Overall, this analysis suggests that gun control interest groups spend more money lobbying state legislators in responses to mass shootings when victims are White but not when victims are racial and ethnic minorities. Gun rights interest groups' lobbying spending is not affected by either type of mass shooting fatality.

Discussion

This paper contends that victims' racial and ethnic identities shape state policy responses to mass shootings, and the empirical findings support this claim. White mass shooting fatalities and majority White mass shootings lead to stricter state gun laws, but racial and ethnic minority fatalities and majority–minority mass shootings have either a negative effect or no substantial effect on the implementation of new restrictive state gun laws. The results are robust to a wide range of modeling specifications (See SI-C through SI-G). Neither victims' age nor location explains these disparities. Robustness checks further suggest that the shooter's race and ethnicity are not biasing results or that the effects are strictly partisan (See SI-K; SI-L).

Elite behavior may help explain these disparate responses. State legislators discuss gun violence more in the days after an in-state mass shooting as White fatalities increase, but fatalities of color have a substantially weaker positive effect. Gun control interest groups spend more lobbying states in response to more White mass shooting fatalities but not more racial and ethnic minority mass shooting fatalities. Mass shootings do not appear to influence gun rights interest groups' lobbying efforts, irrespective of victims' race. We do not know if race affects gun control interest groups' lobbying because of internal implicit biases or because they respond strategically to legislators' signals and public opinion change. However, these racially disparate lobbying responses may ultimately impact the policy process.

Better understanding the causes of racial disparities in state gun policy responses to mass shootings presents a crucial opportunity for future research. Future studies may want to explore more carefully whether racially biased responses to mass shootings are driven by voters' demands, legislators' biases, media frames, interest group lobbying, or other causes.

This paper pushes us to think beyond conventional explanations, like NRA lobbying or political polarization, for gun policy inaction in the United States. It contends that the racialization of gun politics also matters. Gun rights narratives have leveraged racial resentment and disparities in gun violence victimization rates to frame gun control as a “special” benefit for people of color and a threat to White Americans, creating political barriers to gun restrictions. When mass shooting victims are White, these narratives can temporally shift, increasing political support for gun control. But when victims are people of color, racialized gun control narratives that impede change are unaffected. These findings indicate that victims' race and ethnicity may partially explain legislators' inaction on gun laws, even in the face of 45,000 gun deaths annually.

Moreover, the paper reveals that legislators tend to respond more compassionately to tragic events and public health crises when the victims are White, compared to when they are racial and ethnic minorities. The racially biased responses discovered in this paper are not unique to gun policy. For instance, differing responses to Hurricane Harvey and Hurricane Maria and the different policy tools used to address the opioid crisis compared to the crack epidemic reflect racial and ethnic inequities in government responsiveness. Further investigation into the causes of these disparities is essential to promote a more equitable and just democracy.