I will present a project on a model for nonresponse which I am currently working on. Some preliminary results will be given. To have a good perspective of the topic, I will start by giving some short comments about the current methodology for finite population sampling from three different approaches: design based (randomization), model based (non-Bayesian) and model based (Bayesian).

Then I will focus on Rubin’s method to handle nonresponse: multiple imputation. The model follows some ideas from Rubin’s approach but employs the idea of using linearity of probability measures to construct a bivariate distribution which links the parameters of the respondents and the non-respondents.