Does Torture Stop Terror? Nope.

Torture is in the news. Bush-era memos justifying the use of violence against detainees have just been released. The Bush people that wrote or acted on these memos justify such violence in the name of stopping terrorism. The claim is that the violence led the detainees to give up information that allowed the authorities to foil terrorist attacks in the planning stages.

As many have pointed out in recent days, it's difficult to know with any certainty that this is the case. It's possible that the detainees would have given up the intel even if gentler techniques of questioning had been used (as professional interrogators for the FBI and other agencies have long claimed). It's also possible that the authorities would have found about the impending attacks from other sources, or that the attacks would have been called of for any number of reasons. So the debate is stuck in a circle, with supporters of the methods claiming they worked, and opponents saying that they did not.

One partial solution is to look at different types of data. In particular, do countries that practice torture heavily experience less terrorism? The answer is pretty clearly "no." Here's how I arrived at this conclusion:

My colleague Jim Piazza and I have a paper coming out in Comparative Political Studies that analyzes the influence of human rights abuses on terrorism. We find that, for a wide range of data sources, control variables, and statistical specifications that governments that abuse rights actually experience more terrorism.

But what about torture? Torture is only one of the broader range of human rights we looked at in the paper. Is the relationship of torture to terror different? It might be, since far more countries engage in torture than in other forms of violent human rights abuses. To answer this question, I re-analyzed the data from the paper. Details are in the following paragraph; skip down if you just want the punchline.

I re-estimated the three models described in the paper that use the MIPT measure of terrorism as the dependent variable. This counts the number of terrorist attacks in each country from 1998 to 2004 committed by domestic and transnational groups, and also combines these into a measure of all terrorism. I used a negative binomial regression with robust standard errors clustered on countries and the same independent variables as those reported in the paper (political participation, constraints in the executive, regime durability, international war, civil war, and the logs of population and GDP per capita). I replaced the independent variable measuring human rights with two new variables. The first is a measure of torture from the CIRI project. The second is the measure of human rights used in the paper minus torture. This is meant to capture the possibility that torture and other human rights abuses are substitutes for each other; a regime might not torture, say, but could still have a bad record of respecting other rights.

Torture has a negative and statistically significant relationship to terrorism in all three models. In other words, countries that engage in more torture (and thus have a lower score on the torture variable) consistenly experience more, not less, of both domestic and transnational terrorism. This mirrors the more general finding reported in the paper that respect for human rights is associated with less terror as well.

What are the implications for the debate in the US today? The clearest is that torture does not work, at least in reducing terorism. It's another nail in the coffin for those who justify torture as a tool of counterterrorism. It also suggests that we don't need to worry about how revealing the details of the US torture program will provide terrorists with the skills to avoid providing information to interrogators. Intead, it suggests that a suprisingly easy and morally unambiguous counterterrorism strategy is to be nice to people. Being mean (like, say, torturing) seems to annoy some victims, who go on to become or serve as examples to new terrorists.