An alternative to simply excluding people would be to include a "confidence index", possibly based off the number of people in each group. The more people in t

An alternative to simply excluding people would be to include a "confidence index", possibly based off the number of people in each group. The more people in the group in the US, the more likely that that group is representative of the entire group around the world. (Hand waving: the null hypothesis is that there are actually no differences between the religious groups. So, the more people you include in each group, the closer you will get to the overall average.) When you have lots of people in each group, relative to the total population, but still see differences, then that is interesting. This fits with what you are trying to exclude, because your immigrants are likely a small percentage of their respective religious groups, so we can be uncertain about the conclusions we draw from a small sample.