I meant something like this: Imagine that there is a thing T that you want to study. Correlation between T and X is 0.9. Correlation between T and Y is 0.6. Let’s assume that there are no other known factors besides X and Y which would correlate significantly with T.
If you start your research by asking (if you are primed to ask) “is there a significant correlation between T and Y?”, your research will continue like this “yes, we have measured that correlation between T and Y is 0.6, end of story” and you will publish this. There is a risk that you will miss X completely, because you will focus only on Y. But if your goal is to find a good predictor of T, it would be better to discover X.
I think there is a lot of motivated “research” about violence, where the bottom line is: men are evil, women are victims. This has some relation to the territory: certainly men commit much more violent crimes than women. Though even in this situation, why stop at the male sex? Why not also evaluate the impact of e.g. education, social class, previous criminal record, or (political correctness forbid!) ethnicity? Maybe there is some correlation here, too.
If we move from physical violence to other kinds of abuse, the results may change. Not just the correlation with male sex can be weaker, maybe even negative, but more importantly, there may be a significant correlation with something else, which we completely ignore, because we focus only on correlation with sex.
So generally, is is better to ask “what causes this kind of abuse?” than “how is this kind of abuse related to sex?”. If the correlation with sex is significant (yes, sometimes it is), let it come freely as an answer to the first question, but let’s not start with assumption that it is significant.
I meant something like this: Imagine that there is a thing T that you want to study. Correlation between T and X is 0.9. Correlation between T and Y is 0.6. Let’s assume that there are no other known factors besides X and Y which would correlate significantly with T.
If you start your research by asking (if you are primed to ask) “is there a significant correlation between T and Y?”, your research will continue like this “yes, we have measured that correlation between T and Y is 0.6, end of story” and you will publish this. There is a risk that you will miss X completely, because you will focus only on Y. But if your goal is to find a good predictor of T, it would be better to discover X.
I think there is a lot of motivated “research” about violence, where the bottom line is: men are evil, women are victims. This has some relation to the territory: certainly men commit much more violent crimes than women. Though even in this situation, why stop at the male sex? Why not also evaluate the impact of e.g. education, social class, previous criminal record, or (political correctness forbid!) ethnicity? Maybe there is some correlation here, too.
If we move from physical violence to other kinds of abuse, the results may change. Not just the correlation with male sex can be weaker, maybe even negative, but more importantly, there may be a significant correlation with something else, which we completely ignore, because we focus only on correlation with sex.
So generally, is is better to ask “what causes this kind of abuse?” than “how is this kind of abuse related to sex?”. If the correlation with sex is significant (yes, sometimes it is), let it come freely as an answer to the first question, but let’s not start with assumption that it is significant.
Thank you; I edited the question to eliminate the (selection bias?) privileged hypothesis.