Table 9.

Differential returns to education by cSES, with cSES operationalized as father’s education (N = 5,417)

	PCS						MCS
	Base model			cSES interaction			Base model			cSES interaction
	Beta	95%CI	p-value	Beta	95%CI	P-value	Beta	95%CI	p-value	Beta	95%CI	p-value
Constant (high cSES)	50.60	(50.07, 51.14)	< 0.0005	50.81	(50.25, 51.37)	< 0.0005	53.88	(53.39, 54.37)	< 0.0005	53.82	(53.29, 54.35)	< 0.0005
Education	0.93	(0.79, 1.06)	< 0.0005	0.81	(0.67, 0.96)	< 0.0005	0.83	(0.51, 1.16)	< 0.0005	0.95	(0.45, 1.44)	< 0.0005
Low cSES	−1.17	(−1.85, −0.49)	0.001	−1.38	(−2.13, − 0.64)	< 0.0005	−0.31	(−0.95, 0.32)	0.337	−0.28	(−0.94, 0.38)	0.400
Low cSES*education				0.37	(0.05, 0.68)	0.022				−0.19	(−0.83, 0.46)	0.570

PCS is physical health component summary score; MCS is mental health component summary score.

Education is centered at 12, and birth year is centered at 1960 so the constant is interpretable.

For PCS, education is coded linearly as years of education; a 1-unit increase in education is a 1-year increase in schooling.

For MCS, education is coded linearly as years of education until 13 years, and flat thereafter (i.e. everything after 13 years of schooling is recoded as 13 years because found this specification was the best fit for the data). A 1-unit increase in education is a 1-year increase in schooling until 13 years. Models adjusted for cSES, birth place, and rural residence at age 14; we modeled main effects and interactions for Other race men and Other race women but did not display these results due to ambiguity in interpretation.

There were N = 741 individuals missing father’s education, so the sample size for this analysis was smaller than the main analysis. Those missing father’s education were disproportionately Black (55%).