Parents prefer boys

[Mad Dr Jeffe]

Id say the data is bogus as well, of course that all depends on their sample, but the numbers dont seem weighty enough. For example if the margin of error is 4% then it suddenly it could become 1% of marriages that fail because they only have girls.

25- 30 percent now that would be something.

this is a charged topic for me, so I think I'll stay out of it. But my parting shot is that I think these economists used their data to prove what they wanted rather than than show the truth.

[The Holy Saint, Grand High Poobah, Master of Monkeys, Ehlers]

I don't see there's any way to say whether the data is good or not just from that article. We have no way of judging the margin of error or what a significant variation is in that area. You can make guesses and say would would please you, but it doesn't alter what makes something statistically and mathematically significant.

The key, whatever the margin for error is, would be how repeatable the data is. If multiple surveys get the same or very similar data, well then, there's something there.

[stacer]

I'm not saying the numbers are bogus (necessarily). I'm just saying that correlations are not causes. And I was trying to remember the ANOVA that Mustard talked about, about how to determine statistical significance. Oh well.

[House of Mustard]

The way ANOVA works  -- well, I don't know how it works, but what it does is:  It gives you a percentage for every variable that shows you how much of your hypothesis is directly attributable to that variable.  For example, (and these numbers are made up) let's say that you're trying to figure out why some people do well on a test.  Your variables might be Time Studied, IQ, GPA, etc....  ANOVA goes through each variable and tells you 1) if it actually affects the outcome and 2) how much it affects it.  It might say that yes, Time spent studying affects your grade on a test, and it accounts for 45% of how well you do, or something like that.

In this situation, there would certainly be a pile of variables tossed in, and all that it would take for them to make their claim is for ANOVA to tell them that gender of children IS statistically significant.  It might only account for 5% of the reason parents get divorced, but it does have an effect.

[House of Mustard]

Alright - the glories of the internet.

I found the actual article.  Appendix Table A1 is the ANOVA matrix.  If you want to take a look, the R-squared, listed at the bottom, is the number that determines significance.

http://gsbwww.uchicago.edu/labor/dahl.pdf

The significance is certainly low, showing that (in the case of families with only one child - a girl) the child's gender only accounts for 2.5% of the divorce.  Still, though, that's something.

Edit:  I've been reading through the paper, and, whether you agree with the conclusions or not, the numbers are really fascinating.

[Mistress of Darkness]

I wasn't so worried about the numbers as how the numbers were being presented. The way that article was written implied that a child's gender plays a greater role in divorce than it actually does.

Stupid publicity seeking reporters! Bad! No Donut! Or Coffee, or Whatever!

But 2.5%, I can believe that.

[House of Mustard]

Yeah.  It makes it sound bigger than it is.

But:

It's still interesting.  It's hard to argue with the numbers - on a lot the samples they had one or two million people.

The fact that there is a measurable difference between male and female children in so many different issues (shotgun weddings, decisions to have more kids, decisions to get divorces, etc...) shows that there is a very real parental preference.