We consider a simple linear regression model that accommodates situations where both the dependent and the independent variables are interval censored. We obtain maximum likelihood estimators of its parameters and compare their performance with that of estimators derived under ordinary linear regression models. We also develop prediction intervals for the response and illustrate the results with data from an audiometric study designed to evaluate the possibility of prediction of behavioural thresholds from physiological thresholds.
Keywords: Linear models; interval censoring; prediction interval.