SEMINARIESCHEMA FÖR MATEMATISK STATISTIK 

Onsdag    19/6  10.15         Ann-Cecilia Karlsson
          1996                Matematisk statistik, Lund
                              presenterar sitt examensarbete:


                              Analysis of truncated data by means of a
                              stochastic EM algorithm
                              

Abstract:

 We have used two iterative algorithms, the EM and SEM algorithms,
 to calculate estimates of the parameters from truncated normally
 distributed data sets. 
 
 In the EM algorithm the conditional expectations involved are
 computed analytically but in the SEM algorithm they are estimated
 by means of a simulation procedure. After that, maximization of
 thelikelihood gives an estimate of the included parameters.

 The result is that the two algorithms give quite similar estimates
 of the parameters for one-dimension and in the case of two-dimensional
 data, when the EM algorithm is very cumbersome, the SEM algorithm gives
 reliable estimates.
 





Björn Holmquist 046-2228546