WORK IN PROGRESS
Mass Forced Displacement and Incarceration: A Comparative Case Study using Augmented Synthetic Controls
Kırdar, Lopez Cruz and Türküm recently published a study in the Journal of Economic Behavior & Organization and concluded that the incarceration rates for various violent and non-violent crime subcategories fell due to the arrival of Syrian refugees. Using the exact, but backward-extented data on incarceration rates used by Kırdar et al. (2022) and incorporating staggered treatment timing of refugee settlement across provinces via the augmented synthetic control, this study, in stark contrast, finds that the treatment effects are always positive (i.e. crime inducing). However, these estimates are of low precision such that following the onset of refugee influx, nineteen categories of crime rates would not have been any different in the absence of mass displacement across Syrian border. This indicates that the negative estimates reported by Kırdar et al. (2022) are of drastically wrong quantities no matter how precise they may be. Our results suggest that the method of choice of the researchers is one and possibly not the only source of bias.
​
Keywords: Augmented synthetic control, crime, refugees, Syrians, Turkey
Causal Interpretability of Treatment Effects in Spatial Panel Models
Spatial panel models are indispensable tools for analyzing data where relationships extend across geographic or social spaces and evolve over time. These models have found widespread application in economics, environmental science, regional science, public health, and they help researchers explore crime spillovers, regional economic development, and disease transmission. Despite their popularity, causal identification and inference in spatial panel models remains a significant challenge because they violate a number of identifying assumptions. Understanding the causal structure of spatial panel models helps clarify long-standing debates about how spatial relationships affect outcomes.
While causal inference is well-developed for non-spatial models, the added spatial dimension introduces unique challenges, such as spatial confounding, interference and the need for spatial instrumental variables (IVs). Moreover, many applied researchers employ spatial panel models without fully understanding their identifying assumptions. This project also aims to offer practical guidance on which covariates one should control for or stay away from and how identification can be achieved to ensure valid inference.
Structural causal models (SCMs) and causal graphs provide a powerful set of tools for identifying causal effects. While they are well-established in other areas of causal inference, their application to spatial econometrics remains underexplored. This gap represents an opportunity to bridge theoretical advancements in causal inference with context-specific applications of spatial econometric models. By building SCMs and making use of Pearl’s causal framework, this paper analyzes the causal structures of frequently employed spatial panel models and several variants to illuminate their underlying assumptions, explore how causal effects can be identified, and evaluate the conditions under which valid inferences can be drawn.
The paper aims to align this integration with pressing debates in several fields such as the economics of crime, health economics, regional science, epidemiology and political economy. Prominent case studies include community policing and crime; gun control laws and crime, abortion and clinic access, vaccine rollout and immunization; and economic shocks, gerrymandering and voting behavior.