Data error. If you look in the excel datafile, you can see that there is only crime data for years 2002 and 2003 for Mongolia. 156 and 109 charges. The population of Mongolians in these two years are 10 and 15, giving a charge per capita of 15.6 and 7.3! (cells M54 and N54 before removal). The second highest country had a charge per capita of .79 (Georgia) which as noted is already about 71 times that of the lowest country with .01 (Thailand)....
Let me reply to a few points.You said: "The stats agency only reports crime rates for the countries with larger populations, so sampling error should not be so large."
Ok. There are two sources of sampling error with this type of study. First, the migrant sample rates (i.e., number of criminals caught ) as an index of the true population rates in country X. So, if you have 500 Syrians you are less likely to get a true estimate of e.g., the true migrant population murder/criminal rates than if you have 10000 Syrian, because one uncaught Syrian murder/criminal will have a much greater impact if the sample size is low. Dealing with low probability outcomes like crime/murder greatly worsens this problem. As for the second source, the migrant population in country X as an index of true migrant performance, were migrants unselected. Large migrant samples will not eliminate this second type of sampling error, it will just reduce the chances of it-- since the greater the migrant population the more likely that the migrants are unselected (taking into account national population size).
Because of these two types of sampling error, I suggested weighting. You said: I'm not familiar with any cross-country comparable crime statistic.
What I mean is that you should briefly mention the correlation between homicide rates and IQ on the national level to give some perspective. See LV 2012 page 271. You should do this since your NIQ- MigrantHomicide correlation is limited by the NIQ-NHomicide correlation, which is between 0.2 and 0.8, methodology depending. Also doing this gives some background. You said: Can you explain? in response to my statement "You could just compute the cross year joint probability."
You were saying that it is improbably that the rates would be consistently negative across years if there was no true association. I replied that you could compute the joint probability to test this point. You can use Fischer's test:https://en.wikipedia.org/wiki/Fisher%27s_method
It's not a big deal -- don't bother. You said: What is wrong with the current formulation [next up]?
This was too informal.