A predictor in a poorly specified model can perform just fine until you look a bit closer and identify stronger predictors, or address confounding factors explicitly. eg.
IFF women are more risk-averse THEN less women entrepreneurs
IFF time lost to childbirth is a factor THEN less " "
In the former case, risk aversion drives the predictor, and linkage with gender makes it appear as though gender is a strong predictor. After breaking the predictor in two, a much more informative fit is generated. Et cetera. I make no claims as to the correctness of the above. But I am dubious that vagina possession is the kiss of death for starting one's own business. Confounding seems likely.
Ratcheting up or down the sample size does not address confounding. Are women incapable of becoming successful entrepreneurs? Oprah would probably suggest not. Are there numerous factors working against them? Yes. Far more interesting is to probe which characteristics are overwhelmingly female and which are shared among sexes.
IFF women are more risk-averse THEN less women entrepreneurs IFF time lost to childbirth is a factor THEN less " "
In the former case, risk aversion drives the predictor, and linkage with gender makes it appear as though gender is a strong predictor. After breaking the predictor in two, a much more informative fit is generated. Et cetera. I make no claims as to the correctness of the above. But I am dubious that vagina possession is the kiss of death for starting one's own business. Confounding seems likely.
Ratcheting up or down the sample size does not address confounding. Are women incapable of becoming successful entrepreneurs? Oprah would probably suggest not. Are there numerous factors working against them? Yes. Far more interesting is to probe which characteristics are overwhelmingly female and which are shared among sexes.