The Rasch model and other latent trait models encounter some difficulty when faced with an appreciable amount of missing data or omitting behavior. The present note assumes that some reasonable missing data model has been formulated which does not involve the parameters associated with the latent ability of interest. A maximum likelihood function is used that is based on probabilities which are conditional on the occurrence of a response. (Author/DB)