Even the graph weighted for population shows a U-shaped curve, with very liberal people having somewhat more children than moderately-liberal ones. Since you describe this as a "difference" between the two graphs, do you have reason to think that the weighted graph's U shape is some sort of statistical artifact?
Both graphs are weighted (and both are statistical artifacts…).
The equal-weighted graph counts the 312 45-59 year old survey respondents in Iceland as equally important as the 558 such respondents in France, even though it’s much smaller. The population-weighted graph is going to give much more weight to big countries. But that also means that any quirk of the survey in big countries will have an outsize effect. (For example, consider that ISSP survey modules have to be translated into different languages….)
It depends if you are thinking of Europe as “one place where the same basic process is happening everywhere” or as “a bunch of different countries which are running different experiments”. If the latter, then you want the equal-weighted version, you’re interested in the average country, not the average person! Though you probably also should divide up countries more and look at them individually, since they might have quite idiosyncratic processes at work.
I’d be a little careful of reading too much into the U shape. Sample sizes aren’t great here, once we restrict the age range and consider the effect of weighting. The effects I’ve focused on all seem to be big and robust.
Even the graph weighted for population shows a U-shaped curve, with very liberal people having somewhat more children than moderately-liberal ones. Since you describe this as a "difference" between the two graphs, do you have reason to think that the weighted graph's U shape is some sort of statistical artifact?
Both graphs are weighted (and both are statistical artifacts…).
The equal-weighted graph counts the 312 45-59 year old survey respondents in Iceland as equally important as the 558 such respondents in France, even though it’s much smaller. The population-weighted graph is going to give much more weight to big countries. But that also means that any quirk of the survey in big countries will have an outsize effect. (For example, consider that ISSP survey modules have to be translated into different languages….)
It depends if you are thinking of Europe as “one place where the same basic process is happening everywhere” or as “a bunch of different countries which are running different experiments”. If the latter, then you want the equal-weighted version, you’re interested in the average country, not the average person! Though you probably also should divide up countries more and look at them individually, since they might have quite idiosyncratic processes at work.
I’d be a little careful of reading too much into the U shape. Sample sizes aren’t great here, once we restrict the age range and consider the effect of weighting. The effects I’ve focused on all seem to be big and robust.