Heckman-type selection models to obtain unbiased estimates with missing measures outcome: theoretical considerations and an application to missing birth weight data

Table 2 Estimated selection probabilities from probit models

Probit models
Variable	Unadjusted coeff		Adjusted coeff
Variable	dy/dx	95% CI	dy/dx	95% CI
IFAS: No⁽⁺⁾
Yes	0.171***	0.150; 0.192	0.243***	0.221; 0.265
Distance
Distance	−0.024***	-0.027; -0.021	− 0.000	− 0.006; 0.005
Educational attainment: No schooling⁽⁺⁾
Primary			0.025**	0.002; 0.048
Coranic			0.014	−0.033; 0.062
Secondary or higher			0.080***	0.044; 0.117
Marital status: Unmarried⁽⁺⁾
Common-law union			−0.019	−0.057; 0.019
Married			0.064***	0.024; 0.103
Divorced/widowed			0.055	−0.061; 0.170
Maternal age (years): 20–34⁽⁺⁾
15–19			0.016	−0.014; 0.047
35–49			0.051***	0.021; 0.081
Socioeconomic status: Most poor⁽⁺⁾
Poor			0.018	−0.013; 0.048
Middle			0.097***	0.065; 0.129
Rich			0.100***	0.068; 0.131
Most rich			0.105***	0.066; 0.145
Anaemia: No⁽⁺⁾
Yes			0.050***	0.021; 0.079
Lack of appetite: No⁽⁺⁾
Yes			0.039***	0.014; 0.064
Previous births: 0 child⁽⁺⁾
1–4			−0.029	-0.086; 0.028
≥ 5			−0.092***	-0.154; -0.029
Twin births: No⁽⁺⁾
Yes			0.092***	0.043; 0.140
observations	7325		7325

*** p < 0.01. ** p < 0.05.* p < 0.1;⁽⁺⁾ Reference category; dy/dx = Marginal effect is a change in the probability that Y = 1 with a specific change in X. Adjusted model controls village fixed effects; CI Confidence interval

ISSN: 1471-2288