Emailed for the study two days ago. Got it this morning.

So the NLSY79 cohort was given the CES-D depression scale in '91, again in '94, and then not again until '98 and '08. The NLSY79 cohort was also queried about share of family income yearly. In '87, the cohort was queried "about their opinion towards women’s roles in work and family domains using eight-item on a four-point scale (1 = strongly agree, 2 = agree, 3 = disagree, and 4 = Strongly Disagree). They limited their sample to 2-parent households in a married or committed relationship where at least one parent worked full time that had at least one child under 13 living at home.

They ended up with 1463 fathers and 1769 mothers. CES-D is 20 questions. This study picked 7 of those 20 and discarded the other 13. Then they looked at the '87 numbers for gender roles. Then they looked at the money numbers and did this:

That's for depressive symptoms and relative earnings. stay-at-home status is another 30 coefficient table. For those keeping track at home, that's 15 questions, and 76 cofactors to find 2 correlations in a sample size of 3k. I'm no statistician so maybe what they're doing is totally rigorous and makes absolute sense but the results aren't exactly jumping out at me. I would also argue that the amount of massaging they're doing of the data makes my spidey senses tingle regardless of what they're looking for.