[Application of tobit regression models in modelling censored epidemiological variables]

Many variables in epidemiological studies are continuous measures obtained by means of measurement equipments with detection limits, generating censored distributions. The censorship, opposite to the trucation, takes place for a defect of the data of the sample. The distribution of a censored variable is a mixture between a continuous and a categorical distributions. In this case, results from lineal regression models, by means of ordinary least squares, will provide biased estimates. With one only censorhip point the tobit model must be used, while with several censorship points this model's generalization should also be used. The illustration of these models is presented through the analysis of the levels of mercury measured in urine in the study about health effects of a municipal solid-waste incinerator in the county of Mataró (Spain).