Prediction in the lognormal regression model with spatial error dependence

In the context of the lognormal regression model with spatial error dependence, the present study examines correction of a bias in prediction. If interest lies in the predicted mean value of the dependent variable, antilogarithmic transformation of the predicted mean value of the regressand produces a bias. In order to correct such a transformation bias, we derive several alternative predictors by extending some of the predictors suggested for the lognormal regression model with spherical disturbances. Behaviors of our predictors are described in a theoretical manner, and their performances are assessed in an experimental manner. Extension of an asymptotically unbiased predictor is shown to be useful.