It is generally very difficult to decide which models are best since they are alternative ways of specifying the problem and you cannot decide between them on liklelihood grounds. It is generally only possible to approach the problem from which approach you like. I will email you some slides about this on Monday. I usually use Splus or R to approach the problem, but it woukld not be hard to write it in SAS using a minimisation routine, maybe in IML. Do you have an application where you think that non-ignorable non-response is likley?
Gillian -----Original Message----- From: [email protected] [mailto:[email protected]] Sent: 15 December 2000 14:52 To: [email protected] Subject: IMPUTE: Thank you for responding. But these models are needed for multiple imputation. Selection models and (pattern) mixture models have been proposed to use for multiple imputation when the data are subject to nonignorable nonresponse. But how can be decided which model is the best approach? Another issue is are there SAS macro's available for selection or mixture models? kind regards
