ABSTRACT
This article seeks to clarify the relationship between Black Lives Matter and the enactment of state-level police reform by engaging with a broader discussion surrounding policy innovation that has taken shape in recent decades. We ask, what contributes to the differences in state responsiveness to the BLM movement? Moreover, is there a link between the protests, themselves, and state-level police reform enactments? We find, a state’s response to demands for police reform is heavily dependent on a combination of both the conditions within their state as well as their position in the overall police reform diffusion space. More importantly, we find that the BLM movement, itself, played dual roles in applying pressure on states to enact reforms from August 2014 – December 2020: (1) reform efforts on the part of lawmakers were proportionate to the frequency that BLM protesters take their grievances to the street, and (2) policy adoptions are largely shaped by the state-to-state diffusion network where states with the highest frequencies of protests are most influential. The results of this analysis should serve as a sharp rebuttal for those that question and downplay the effectiveness of Black-led social movements in achieving substantive policy change.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 The one noteworthy disadvantage to this database is the fact that their count is almost certainly an undercounting of the true number of protests within a given state, as a function of the authors’ choice to restrict the cataloging of protests to include only those conducted in cities with populations larger than 30,000 people. Please see the article for a more thorough conversation of the limitations (Williamson, Trump, and Einstein Citation2018).
2 We acknowledge the debate that the performance of the Poisson and Negative Binomial models is far superior to the OLS model when handling zero-bounded, count-dependent variables because the least-squares model does not constrain the expected number of events to be zero (King Citation1988). Additionally, employing a least-squares measurement for count-level data introduces a great deal of inefficiency and bias into the analysis. We include all three measures for two reasons. First, the LNAM model operates on assumptions that are found within the OLS model. Second, there is no substantive difference in the findings of the three models when it comes to the key variable of interest.
3 Tests for overdispersion and log-likelihood comparison test revealed that the negative binomial added very little – if anything – in terms of model fit. In fact, models 2 and 3 provided identical covariate measures. The only difference between the two comes in the goodness-of-fit measures, where the Poisson model performed slightly better. Nevertheless, we opt to include both count-level models in our reporting of the findings alongside the OLS model to serve as robustness checks.
4 The process of identifying inferred diffusion networks with the NetInf package is best accomplished with larger numbers of policies over longer periods of times. Where most applications package examine dozens of policies that diffuse over decades, this application examines 12 policies that spread much more rapidly – many times, over the course of days, or weeks. Because of this, we lax the parameter on the confidence of edge formation from p < 0.05 to p < 0.1.