More formally, this is the origin story for Simpson's paradox: when you combine subgroups, trends can disappear or even reverse. The more dimensions you analyze, the close to 'truth' you can get.
This is why you have such varying claims about stuff like women earning equal pay with men. People who want social change (activists, politicans) group together the entire population, and come to the conclusion that women make 83 cents on the dollar. But when you break out along other dimensions (age/YOE, occupation, hours worked), the gap is reduced. But never eliminated!
This is why you have such varying claims about stuff like women earning equal pay with men. People who want social change (activists, politicans) group together the entire population, and come to the conclusion that women make 83 cents on the dollar. But when you break out along other dimensions (age/YOE, occupation, hours worked), the gap is reduced. But never eliminated!
What I'd really like to do some day is be smart enough to replicate https://www.metafilter.com/126704/with-numbers-like-these-wh... and see how more recent census data compares.