Implications for Empirical Estimation:
Before proceeding further let us pause to look at some distinguishing characteristics of this model. First suppose the variance of E, is multiplied by some positive constant k2. The latent regressioi will now be, Yi* = xi,β +kεr. But, Y*/k = xi,β + εr is the same model with the same data. The observed data will remain as it is. Y, is still 0 or 1, depending only on the sign of Yi* not on its scale. In other words the data have no information about k and so it cannot be estimated. Secondly, if the model contains a constant term, say a, then,
Since α is unknown, the difference (a - c) remains an unknown parameter. Therefore, we may write
The problem becomes one of choosing an appropriate probability distribution for εi. Successful estimation of the model ,outlined above is dependent on making the correct choice of functional fdnh for G(xi,β) or the appropriate probability distribution for εr. Choices made myst be such that the predicted probabilities are consistent with the underlying theory. Choosing G(xi,β) to be any proper, continuous probability distribution, say F(xi,β), defined over the real line will ensure logical consistency.