Comment by nomilk

> that would still leave up to 49% .... how did you conclude?

This fundamentally misunderstands how predictive models work. A parameter is a potentially useful predictor if it's better than excluding it 50.000001% of the time (high frequency trading is good evidence of this).

> conclude

Conclusions? Absolutely not. Higher statistical probability? Yes. Based on evidence. To state your point, which is bleeding obvious, of course you cannot know how recently someone or their family came to America based only on their behaviour with regard to the law. But their behaviour with regard to the law absolutely can be a useful predictor.

(in fact, it's precisely this rationale that justifies in some cases giving foreigners lighter sentences where 'their culture' allowed for xyz but the local jurisdiction doesn't - roadrules in the UK is a pretty good example: local truck drivers get the full penalty; those from continental Europe often do not, since road rules are less likely to be known by them)

Alternate example: ask a Western drug smuggler busted in Indonesia or Vietnam how long they expected their jail sentence to be if caught (spoiler - it's a trick question: they'll say 3-5 years and instead are met with the death penalty; whereas most locals are well aware of this) - ignorance to local laws and customs does correlate with how long someone (or their family) has lived in the area, if at all.