Two Problems with Profiling

Profiling is a method of observing characteristics, actions and circumstances to draw conclusions about a person.  It is a useful tool that we all use every day.  Doctors build health profiles based on age, weight, race, sex and personal habits such as drinking and smoking.  Amazon books uses profiling to suggest books you might be interested in based on past purchases and current browsing. 

Profiling becomes controversial when race is one of the observed characteristics or when the profiler can legally use coercion on the profiled even when the profiled characteristic is statistically based.  Suppose Officer Holmes of the Arcadia Police Department observes that drug use has crept into his normally peaceful jurisdiction.  He travels to Gotham and notes that four times as many Hispanic males are arrested on drug charges than Asian males.  Based on that observation, he focuses his time and other police resources on Hispanic males to reduce drug related crime.  The problem could be one of reverse causality.  Perhaps more Hispanic males were arrested in the past because the police harbor prejudices against Hispanic males.  They are arrested more frequently because they are investigated more frequently.

Now assume that Officer Holmes has read scientifically conducted research based on surveys that finds that 12% of Hispanic males acknowledge illegal drug compared to 3% of Asian males.  He is scientifically justified in using race as a profiling characteristic, but problems remain.  In the course of his work, using race, sex, location and other behaviors to profile potential drug users, he observes 1,000 Hispanic males.  Of these, he stops and questions 270, and of these, he arrests and charges 60 with a drug related crime and all are convicted.  The remaining 210 Hispanic males were innocent and while they were only stopped and questioned, may resent Holmes’ behavior and grow to distrust the police.

Profiling becomes more objectionable as the ability to correctly identify offenders decreases and the level of coercion increases.  On January 8, 2011, Jared Loughner made himself notorious by killing six and wounding twelve in a failed assassination attempt of Congresswoman Gabrielle Giffords.  Some called for more stringent gun control.  Second Amendment supporters and gun owners argued that it would be wrong to inconvenience many gun owners for the actions of a single deranged man.  Others noting that friends, family, students and teachers had observed Loughner’s erratic behavior and on several occasions had called the police for assistance suggest that a stronger enforcement mechanism be developed to help youths with mental disorders and protect the public.

Let’s say that Arizona imposes a new law that that allows authorities to medicate youths profiled by behavioral characteristics as potentially violent.  Assume that researchers followed the behavior of 10,000 young adults and one in a thousand (10 young adults in our sample) have a disorder that will lead to acts of violence.  Now assume that we implement a system that correctly identifies 80% of those with potentially violent mental disorders (8 of the ten potentially violent young adults are identified.  Two are not.).  But there is a tradeoff.   In our example, there are 9,990 young adults who are not potentially violent yet one time in a hundred, or 99.9 times a young adult will be wrongly identified as being potentially violent and subject to forced mediation.  Is catching 8 of 10 potentially violent worth the cost of forcefully medicating nearly 100 who are not?  If this tradeoff is too high, how many wrongful forced medications will you tolerate to stop the violence these youths will cause?