I treat people a certain way. Someone suggested that the way I treat people is not well-tolerated by women. I said, among other things, that treating women differently to accomodate them might not be appreciated by women. Alicorn suggested that (to paraphrase), men like to be treated with c=3 and women like to be treated with c=7, so I should treat everyone with c = something in-between 3 and 7.
I asked whether that should be weighted (by the number of men and women present, or the number of men and women desired, or not at all); but more importantly, replied that that is inconsistent: Either
There is no significant mean difference between the way men and women like to be treated, or
There is a difference, and therefore I should treat men with c=3 and women with c=7.
Saying there is a difference, and I can use my knowledge of that difference to compute c = (9x3 + 1x7)/10 = 3.4 (assuming LW is 90% male); but I can’t use my knowledge of that difference to treat men with c=3 and women with c=7, seems illogical. If the preferences of men are distributed normally around c=3, and the preferences of women are distributed normally around c=7, using gender as a factor in my model will result in lower error, no matter how great the variance in these distributions is.
I can imagine possible justifications for the c=3.4 approach; but I can’t imagine any justifications strong enough not just to rule out even considering the c(gender) approach for all applications, but also to impugn the morality of those who consider it.
I am harping on this because it is an error in reasoning that, in American intellectual society, is not only committed routinely, but is generally assumed to be morally superior to any rational approach.
People derive lots of negative utility from noticing that they are being treated differently according to gender. If you do so, you are unlikely to be able to keep that a secret. Therefore, you should treat both genders the same.
That might be a good reason, but IMHO not good enough to take the option off the table.
Also, it seems to me the opposite may be more true: People like being treated in ways that indicate they are successfully conforming to gender stereotypes. Boys and men like being given the signal that they are masculine; girls and women, that they are feminine.
This is true for some stereotypes but not for others. In addition, people don’t want to feel that you are treating them differently because of their gender, even if they’d feel good if you treated them differently because you noticed their conformance to stereotypes. People will probably be able to tell what you are doing.
I treat people a certain way. Someone suggested that the way I treat people is not well-tolerated by women. I said, among other things, that treating women differently to accomodate them might not be appreciated by women. Alicorn suggested that (to paraphrase), men like to be treated with c=3 and women like to be treated with c=7, so I should treat everyone with c = something in-between 3 and 7.
I asked whether that should be weighted (by the number of men and women present, or the number of men and women desired, or not at all); but more importantly, replied that that is inconsistent: Either
There is no significant mean difference between the way men and women like to be treated, or
There is a difference, and therefore I should treat men with c=3 and women with c=7.
Saying there is a difference, and I can use my knowledge of that difference to compute c = (9x3 + 1x7)/10 = 3.4 (assuming LW is 90% male); but I can’t use my knowledge of that difference to treat men with c=3 and women with c=7, seems illogical. If the preferences of men are distributed normally around c=3, and the preferences of women are distributed normally around c=7, using gender as a factor in my model will result in lower error, no matter how great the variance in these distributions is.
I can imagine possible justifications for the c=3.4 approach; but I can’t imagine any justifications strong enough not just to rule out even considering the c(gender) approach for all applications, but also to impugn the morality of those who consider it.
I am harping on this because it is an error in reasoning that, in American intellectual society, is not only committed routinely, but is generally assumed to be morally superior to any rational approach.
Of course, the ideal is for each individual to find out c(individual) and use that.
People derive lots of negative utility from noticing that they are being treated differently according to gender. If you do so, you are unlikely to be able to keep that a secret. Therefore, you should treat both genders the same.
(Isn’t the mean a least-squares estimator?)
That might be a good reason, but IMHO not good enough to take the option off the table. Also, it seems to me the opposite may be more true: People like being treated in ways that indicate they are successfully conforming to gender stereotypes. Boys and men like being given the signal that they are masculine; girls and women, that they are feminine.
D’oh! Yes. Removed my comment.
This is true for some stereotypes but not for others. In addition, people don’t want to feel that you are treating them differently because of their gender, even if they’d feel good if you treated them differently because you noticed their conformance to stereotypes. People will probably be able to tell what you are doing.