Table 6:

OLS, IV and Upper Bounds Estimates for the Effect of BMI on CES-D Score

	Mean (1)	OLS (2)	IV (3)	IV(1) (4)	UB of β under (A4) (5)	UB of β under (A4 & A5) (6)
Panel A: Add health
All	5.79	0.028*** (.010)	0.020 (.038)	0.032** (.016)	0.020 [.052]	0.020 [.066]
Women	6.18	0.062*** (.014)	0.019 (.054)	0.075*** (.020)	0.019 [.094]	0.019 [.118]
Men	5.33	−0.028** (.014)	0.020 (.053)	−0.045* (.023)	−0.028 [−.003]	−0.027 [.010]
Panel B: HRS
All	1.26	0.029*** (.004)	0.051*** (.015)	0.022*** (.006)	0.029 [.037]	0.022 [.035]
Women	1.43	0.030*** (.005)	0.053*** (.019)	0.023*** (.008)	0.030 [.041]	.023 [.040]
Men	1.03	0.026*** (.006)	0.046*** (.023)	0.019** (.010)	0.026 [.041]	0.018 [.042]

Notes: Add Health regressions control for age, age squared, gender, birth order, mother’s education, picture vocabulary score, PGSs for depression and education, and the first 20 ancestry-specific principal components of the genetic data. HRS regressions control for age, age squared, gender, mother’s education, PGSs for depression and education, and the first 10 ancestry-specific principal components of the genetic data. Heteroscedasticity-robust standard errors in (.)

^***significant at 1%

^**significant at 5%

^*significant at 10%.

The Nevo & Rosen (2012) approach is implemented using the imperfectiv command in Stata. The upper endpoint of the 95% confidence interval on the bounded parameter is given in [.].