Re: Local Maximum

[Spencer Graves <spencer.graves@PDF.COM>]

1.  A normal prior just adds a parabola opening down to the 
log(likelihood).  I don't know what software you are using, so I don't 
know how easy it would be to maximize a "penalized" likelihood.
2.  Alternatively, what do you know about the location of the undesired 
local maximum and it's zone of attraction?  If you can characterize 
this, you may be able to select starting values to avoid that.
hope this helps.  spencer graves

Dimitris Rizopoulos wrote:
> Dear S+ users,
>
> I am fitting a latent trait model (factor analysis model for binary
> data) using the EM algorithm!!
>
> After fitting the model with many different sets of initial values, I
> observed that there is a local maximun in the log-likelihood. For some
> sets of initial values the EM algorithm does not "fall" in the local
> maximum but for others it does!!
>
> Is there any way the log-likelihood to overcome that local maximum, in
> the EM algorithm?
>
> Thanks in advance for your suggestions,
> Dimitris
>
> =====
> Dimitris K. Rizopoulos 
> ERASMUS Graduate Student
> International Study Program in Statistics
> Katholieke Universiteit Leuven
> Belgium
> http://www.kuleuven.ac.be/ucs/isps/
>
> ____________________________________________________________
> Do You Yahoo!?
> Αποκτήστε τη δωρεάν @yahoo.gr διεύθυνση σας στο http://www.otenet.gr
> --------------------------------------------------------------------
> This message was distributed by s-news@lists.biostat.wustl.edu.  To
> unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with
> the BODY of the message:  unsubscribe s-news