UPDATE: The conversation with Whiskey is being continued at Mangan’s. Jason Malloy isn’t there, but Peter Frost has two more posts at his own blog continuing the theme, and Jason has contributed to the first of those.
Novaseeker has responded to Jason Malloy’s claims which I highlighted recently. I responded in the comments there with some references to the GSS & Whiskey’s claims, but I didn’t lay out much data. Whiskey said that Jason’s analysis of college campuses as an area where a low male-female ratio produces loutish behavior also describes the post-college world, since women do not marry down when it comes to class (primarily defined by education). The “effective population size” thus remains low for men. Novaseeker similarly thinks the effective population size is limited to the few with certain traits, but those claims can’t be easily checked by the GSS while Whiskey’s can. Not that Whiskey would bother doing something like that.
The first analysis I mentioned was SPEDUC against EDUC in the GSS, filtering for SEX(2). In English, that’s highest grade completed for the spouse of the respondent, against the education of the respondent, filtering for female respondents. There’s a huge heaping of data there, so I’ll just present the mean spouse education matched for each grade of the wife’s education.
|Wife’s highest grade of education completed||Mean years of spouse’s highest grade completed||Spouse’s standard deviation|
Before finishing grade 14 (sophomore year of college) married women will tend to be less educated than their spouse. At 14 and beyond that reverses. Nothing terribly surprising, just assortative mating with some spousal regression to the mean. Among those who have graduated from highschool, the standard deviation tends to increase with grade. As fewer men are available even at a lower relative education, women expand their range.
You may object that highest grade of education is an imperfect measure of class, as occupational prestige is also important. Fortunately, the GSS covers such socio-economic and status indicators as well. SEI gives the respondent’s socioeconomic index, SPSEI gives the respondent’s spouse’s index and PASEI gives the respondent’s father’s index (try and guess the code for mother’s). There is also SPPRES for spouse’s occupational prestige score (PRESTIGE for respondent) as well as other occupation and industry codes. Look for them under RESPONDENT BACKGROUND VARIABLES -> Socio-Economic and Status Indicators and PERSONAL AND FAMILY INFORMATION -> Respondent’s [or Spouse’s or Father’s] Employment. If you think an analysis with those variables will conflict with the story I give here, go to the GSS web application, test them out and let us know the results!
Because the SPEDUC variable filters out respondents who are unmarried, I ran MARITAL against EDUC for SEX(2) to check the marital status of women by years of education.
|Woman’s highest grade of education completed||Probability of currently being married|
Probability of being married at the time of interview peaks for those who stopped with high-school graduation, but remains high for all subsequent years and has secondary peaks at 16 and 18 (if you run this yourself you’ll see the weighted N also has local maxima there, indicating that dropouts are unusual weirdos). Steven P. Martin’s Growing Evidence for a “Divorce Divide”? has a good exploration of how divorce rates have changed over time and the impact of factors like education, age at marriage and both marital & non-marital births on the probability of divorce. Education delays age of marriage but makes marriage more durable. Since the 70s there have been great declines in the probability of divorce for women with college degrees, but not for those with less education. Women with a professional or masters degree show the most pronounced decline. Martin states counterfactually that divorce rates would have risen for men and women with no college education had age of marriage not increased for them since the 70s. Martin selects from people who are already married at one point in time to find the probability of divorce later, but estimating the probability of marriage over a lifetime is trickier. Suqin Ge and Fang Yang attempt to do so in Accounting for the Gender Gap in College Attainment. According to them, before 1990 having attended college implied a lower probability for women, after that it implied a higher one.
Whiskey prefers relying on anecdotes, and his favorite illustrative one for the topic of divorce is Sandra Tsing Loh. His message is that even if women deign to marry betas at all, they can’t stand them and will seek divorce. This is the opposite of reality, as the Inductivist shows. “Alpha” men (I assume determined with the variable NUMWOMEN) are the most likely to be divorced or separated. In the comments, Jason Malloy pipes up to suggest that they also have less happy marriages than betas. This should not be that surprising as the Inductivist’s previous post noted that betas are more religious, and religious marriages are happier.