Summary of Week 3
This week, we looked at the effect of police on crime.
The rational choice model predicts that having more police officers should reduce crime by increasing the probability of punishment, but testing this prediction using actual data is not straightforward. The model predicts that having more police officers should cause crime to fall, holding all else constant.
However, in real life, when we observe that one city has more police officers than another, we suspect that the city with more officers may have more crimes, more budgets, and/or government officials who are more concerned about public safety than the other city. Since each of these differences can have an important effect on city crime rates, it is difficult to disentangle the causal effect of police on crime from data.
Ideally, we would like to run a randomized experiment in which some places receive more police officers than other places with similar characteristics, but running such an experiment is usually very difficult. Economists overcome this problem by exploiting a sudden, unanticipated increase in the level of police presence, such as the significant increase in police presence following a terrorist attack. Findings from such quasi-experimental studies allow researchers to recover the causal relationship of interest from data, without having to run the randomized experiment.
This week, we also learned about how to the difference-in-differences analysis to investigate the causal effect of more police presence on crime.
Following a terrorist attack in Buenos Aires, the Argentinean government quickly provided additional police protection to all Jewish and Muslim institutions in Buenos Aires. Although the number of crimes fell near these institutions after the increase in police presence, we cannot view the before-after difference in crime as the causal effect of police on crime because many other crime-relevant factors may have also changed following the terrorist attack.
In order to disentangle the causal effect of additional police presence on crime, we compute the difference in auto thefts in the protected areas before and after the attack, relative to the difference in auto thefts in other parts of the city before and after the attack. Under the assumption that the crime rates in the protected areas of the city would have followed the same trend as the crime rates in other parts of the city if the increased police presence did not take place, this difference-in-differences captures the causal effect of additional police presence on crime.
Next week, we will learn about the effect of prison on crime. As we saw this week, ideally we would like to run a randomized experiment to find out the causal effect of longer prison sentences on crime, but running such an experiment is probably even more infeasible. However, economists have come up with clever strategies to find the causal effect of longer prison sentences on crime without having to run the experiment. We will learn more about these findings next week.
Lastly, I want to mention that the empirical strategies that we have seen and will see next week (panel data regression, difference-in-differences, and regression discontinuity) can be easily applied to other research questions as well. Many economists use these empirical strategies to investigate important questions in economics, such as whether having more education causes people to have higher income and live healthy and whether more generous unemployment benefits discourage people from working.
See you next week.
© Songman Kang, Hanyang University