Methods for assessing a likelihood by demographic category Specific question: How do statisticians determine the likelihood that such-and-such type of person will commit a crime? What is their methodology?
To illustrate what I'm talking about, people will say things like, "Latinos are 5% more likely to steal a car" (not a real statistic).
How could they possibly know something like that? Specifically, what means do they employ to come to such a conclusion?
 A: In essence one extrapolates from past experience: So you looked at a 100 latinos and a 100 WASPs, and you found that 10 latinos stole a care and 5 WASPs, so the chance that a latino steals a car is 10% and the chance that a WASP steals a car is 5%, So the chance that a latino steals a car is 5 percentage points larger than a WASP.
Things are a bit more complicated: We want to make sure that we are comparing like with like. Younger people are much more likely to steal than older people, and the latino population in the USA is younger than the WASP population. If we don't control for age, then the difference between latinos and WASPs is in part due to the age difference. Similarly, you may want to control for other characteristics, like income.
Another problem is how do you find out if someone stole a car: do you ask them "did you steal a car in the last year?". That happens, but you have to be careful that you don't offend your respondent and that you get reasonably truthful answers. Some research use the trick to ask "How many people do you know that have stolen a car". If you also asked them how many people they know overall and their composition, you can estimate the car theft rates from that.
Alternatively some look police statistics, but if some group is under much stronger scrutany (racial profiling) than another, then the difference you find says more about how the police works then about differences in criminal behaviour between groups. 
