November 05, 2006

Actual vs. Observable Criminality

Inspired from this Telegraph story:

Race watchdogs are to investigate the national DNA database over revelations that up to three quarters of young black men will soon have their profiles stored.

Trevor Phillips, the chairman of the Commission for Racial Equality (CRE), vowed to examine whether the database breached race relations laws following the findings by The Sunday Telegraph. "This is tantamount to criminalising a generation of young black men," he said.

An estimated 135,000 black males aged 15 to 34 will be entered in the crime-fighting- database by April, equivalent to as many as 77 per cent of the young black male population in England and Wales. By contrast, only 22 per cent of young white males, and six per cent of the general population, will be on the database.
We should remember that a group's observed criminality is a function of their propensity for criminality as well as the authorities' propensity to investigate members of the group. I suspect that differences in average criminality (perhaps related to personality differences between groups) result in a (correct) perception that an individual of group A has a higher probability of engaging in some type of criminal activity than an individual of group B. This perception then leads one to investigate more thoroughly persons of group A, thus leading to a greater observable criminality for group A.

Suppose that 60% of car thefts in a city are done by Blues. The police is then justified in thinking of Blues as more likely to steal cars than Greens. If a car theft takes place, and given two suspects, a Blue and a Green, and given a limited number of police resources, the smarter choice is to investigate the Blue individual.

Suppose that the police has enough resources to investigate 1 suspect 50% of the time, and both suspects 50% of the time. Then, 50% of the time they would investigate both suspects and catch the perpetrator (30% Blues and 20% Greens), but 50% of the time they will only investigate the Blue one leading to an arrest of a Blue one in 30% of the cases.

Thus, in total there will be 60% Blues and 20% Greens that were arrested, i.e., a difference in actual criminality of 1.5x was transformed into a 3x observed one. Moreover, 80% of the total criminals would be arrested with a race-conscious policy.

Suppose now that a race-blind policy is enacted. Now, for the 50% of the time where they can only investigate 1 suspect, they pick his race randomly and with equal probability. Now they will expect to catch a Blue criminal whenever a Blue committed the crime and they happened to investigate the Blue suspect (15%) and similarly a Green one (10%). So, with a race-blind policy 75% of the total criminals would be arrested.

This illustrates the problem: using the information encoded in the "race" variable leads to more efficient crime fighting: more criminals are caught. However, it also leads to a upwardly biased estimate of the criminality of the more criminal group.

In my opinion, society should not throw away information encoded in sex, race, ethnicity, etc. in fighting crime. That is the rational choice to maximize justice. However, it should also be aware that this rational group-conscious policy biases our perception.

PS: The relevant statistics for particular human group differences, including those cited by the Telegraph, are not important for the argument presented above. I am agnostic about both the magnitude in actual criminality between different groups (e.g., men vs. women, Muslims vs. Buddhists, blacks vs. Mongoloids, etc.) as well as the potential racial biases (above and beyond the rational group-conscious choice), or racial political correctness that may further inflate/deflate observable differences.

No comments: