Desired fertility, the "up to God" response, and sample selection bias

An unresolved question in the analysis of survey data relating to fertility attitudes and beliefs is how non-numeric responses to questions on ideal family size should be treated. This paper demonstrates that simply dropping "up to God" responses will bias regression results. An unbiased estimator is presented which explicitly models the way in which observations are selected into the sample. The estimator is then employed on Guatemalan and Indian data. No support for the notion that women who answer "up to God" are women who would have given relatively large numeric answers is found in these two samples.