Estimating the economic costs of organized crime by synthetic control methods

Abstract

Summary The economic costs of organized crime have been estimated for the case of southern Italy by Pinotti (Economic Journal 2015; 125, F203–F232, 2015): using synthetic control methods, he finds that, due to the advent of the Italian Mafia in the regions Apulia and Basilicata, GDP per capita dropped by 16%. Replicating this study in a narrow sense by estimating the same model with the same data, but using different software implementations, we observe minor differences stemming from the different implementations. By identifying the correct implementation, we find that the loss in GDP per capita due to the presence of the Mafia has been slightly overestimated.

Elevated police turnover following the summer of George Floyd protests: A synthetic control study

“…The use of synthetic controls is not new in criminal justice research (e.g., Becker & Klößner, 2017; Donohue et al., 2019; Mourtgos et al., 2021; Mourtgos & Adams, 2021; Saunders et al., 2015). Historically, however, the selection of comparison groups has often been made “on the basis of subjective measures of affinity between affected and unaffected units” (Abadie et al., 2010, p. 493).…”

“…Donohue and colleagues (2019) follow a similar approach to test the effect of right‐to‐carry laws on violent crimes, as do Saunders and colleagues (2015) to test the efficacy of focused deterrence on crime but at the neighborhood (instead of state) level. While this method has been shown to be robust (Becker & Klößner, 2017), it still requires comparing the experimental unit to a synthetic control constructed from nonequivalent controls. That is, there is an assumption that unknown confounding variables between the experimental unit and nonequivalent controls (e.g., states, neighborhoods, and police departments) are being accounted for in the statistical model.…”

“…Police agency culture, practices, surrounding communities, and political idiosyncrasies vary greatly (Bottoms & Tankebe, 2012;Crank, 2014;Paoline, 2003). And while the more conventional synthetic control method (Abadie et al, 2010) has been shown to be robust (Becker & Klößner, 2017) and can arguably account for this variation, we believe one advantage of our methodological approach is not having to invoke this assumption. That is, we are able to build the synthetic counterfactual using the time series behavior of the experimental group prior to and after the intervention, as well as predictors specific to the agency being studied.…”

“…The synthetic control is constructed using the experimental group's time series behavior before and after the intervention and combining control variables predictive of the target series before the intervention (Brodersen et al, 2015). The use of synthetic controls is not new in criminal justice research (e.g., Becker & Klößner, 2017;Donohue et al, 2019;Saunders et al, 2015). Historically, however, the selection of comparison groups has often been made "on the basis of subjective measures of affinity between affected and unaffected units" (Abadie et al, 2010, p. 493).…”

Combining Matching and Synthetic Control to Tradeoff Biases From Extrapolation and Interpolation

“…Abadie et al (2010Abadie et al ( , 2015 view the weights as tuning parameters and choose them using cross-validation. We could do this as well with our criterion ( 14), but we have elected not to in the current paper because optimizing over the weights introduces computational issues that, while solvable, are not the main focus of our paper (Becker andKlößner, 2017, 2018).) For the matching estimator, τ is the number of matches, m. The MASC estimator has both m and the model average parameter, φ.…”