Table 2.
Variables involved in missing data mechanisms and the consequent expected bias (compared to the true population values) of coefficient estimates from using the complete records to fit a linear regression and logistic regression of Y on X1, X2
Mechanism depends on | Biased estimation of parameters using complete records | ||||||
---|---|---|---|---|---|---|---|
Typical regression | Logistic regression | ||||||
Constant | Coefficient of X1 | Coefficient of X2 | Constant | Coefficient of X1 | Coefficient of X2 | ||
Y | Yes | Yes | Yes | Yes | No | No | |
X 1 | No | No | No | No | No | No | |
X 2 | No | No | No | No | No | No | |
X1, X2 | No | No | No | No | No | No | |
Y, X1 | Yes | Yes | Yes | Yes | Yes | No | |
Y, X2 | Yes | Yes | Yes | Yes | No | Yes | |
Y, X1, X2 | Yes | Yes | Yes | Yes | Yes | Yes |