The structure of information matrices in latent-variable models is explicated, and the degree to which missing information can be recovered by exploring collateral variables for respondents is characterized. Results are illustrated in the context of item-response-theory models, and practical implications are discussed. (SLD)