Hat-tip to Vanessa: the latest report on global gender inequality is out, providing statistics from international surveys about gender gaps in wages, political representation, education, health, and employment. Vanessa emphasized that the US only ranked 22nd; in fact, I’m fairly certain it should rank lower, because the report says it ranks third in closing the income gap, behind Moldova and Tanzania, even though in reality Sweden does better.
A good rule of thumb is that if an international study conflicts with an intranational one, then the intranational one is better. There are exceptions, for example if the country that takes the survey fudges data, but usually it holds. So when the US Census Bureau says the USA has a full-time wage gap of 23% and Statistics Sweden says Sweden has a full-time wage gap of 15%, they’re almost certainly right.
An international study usually doesn’t have high enough a sample size to compete with a national one. For a good comparison, let’s talk about national/local instead of international/national. The USA’s National Crime Victimization Survey has a sample size of 150,000. It’s hard to improve on it if you want to, say, find out about national rates of assault. But if you wanted to find the rates of assault in New York, you’d have to do so based on only 4,800 people, of whom 4 or 5 will have been victims of assault with injury. The standard error for that would be so huge the study would be meaningless. If, however, your study were limited to just New York, you’d probably have enough funds to have a larger sample size, say 50,000, which would give you ample data.
The US Census Bureau’s determination of wage gaps in the US is to my knowledge not based on a sample. The official Swedish figures cite a standard error figure of less than 1% of each gender’s income. Somehow, the global study says the occupation-controlled gap in Sweden is 29%, even though occupation controlling always decreases the gap. Color me unimpressed.