When the causal graph has more than two nodes, something can have a negative correlation (when measured with no controls) despite having a positive causative effect (which would show a positive correlation in an RCT), or vice versa. People who get chemotherapy are way more likely to die of cancer than people who don't.
I can't imagine the education/fertility relationship being an example of that, though. Nerds go to college more and have fewer kids, but not as many fewer as they'd have had without going to college? Sounds like a stretch.
p
When the causal graph has more than two nodes, something can have a negative correlation (when measured with no controls) despite having a positive causative effect (which would show a positive correlation in an RCT), or vice versa. People who get chemotherapy are way more likely to die of cancer than people who don't.
I can't imagine the education/fertility relationship being an example of that, though. Nerds go to college more and have fewer kids, but not as many fewer as they'd have had without going to college? Sounds like a stretch.
More options
Context Copy link