Hi I was asked if lme can use FIML (Full Information Maximum Likelihood) instead of REML or ML but I don't know the answer. Does anybody know if this is implemented in R? Thanks Francisco
F Z wrote:> Hi > > I was asked if lme can use FIML (Full Information Maximum Likelihood) > instead of REML or ML but I don't know the answer. Does anybody know if > this is implemented in R?To the best of my knowledge, FIML is ML so the answer is yes. For example, the phrase "Full Information Maximum Likelihood" is used in Singer and Willett (2004) "Applied Longitudinal Data Analysis" (Oxford University Press) as a synonym for maximum likelihood.
Thanks to Douglas Bates and Christopher Lawrence for their responses. Christopher is right, that is what I was asking about. I guess that there is no implementation of FIML in R. Would this be a worthy method to include in R? I don't really use this method so I would say no but maybe some people think in a different way (For example SAS users trying ot move to R)? Respectfully Francisco>From: Chris Lawrence <chris at lordsutch.com> >To: R-Help <r-help at r-project.org> >Subject: Re: [R] FIML in lme >Date: Sat, 28 Aug 2004 02:29:08 -0500 > >On Aug 27, Douglas Bates wrote: > > F Z wrote: > > >I was asked if lme can use FIML (Full Information Maximum Likelihood) > > >instead of REML or ML but I don't know the answer. Does anybody know >if > > >this is implemented in R? > > > > To the best of my knowledge, FIML is ML so the answer is yes. > > > > For example, the phrase "Full Information Maximum Likelihood" is used in > > Singer and Willett (2004) "Applied Longitudinal Data Analysis" (Oxford > > University Press) as a synonym for maximum likelihood. > >I have seen FIML used to refer to a type of ML estimation where a >missing data treatment is included in the estimation procedure >(parameter estimates are derived from incomplete cases for only the >variables present in the case, rather than simply discarding the >cases), at least in the latent-variable SEM context, specifically in >AMOS. This may be what Francisco is getting at. > >To my knowledge, no R packages implement this sort of "FIML", for any >class of models, although there are other available missing data >treatments (EM, MCMC estimation). > > >Chris >-- >Christopher N. Lawrence, Ph.D. >Visiting Assistant Professor of Political Science >Millsaps College >1701 N. State St >Jackson, MS 39210 >(601) 974-1438 / lawrecn at millsaps.edu > >______________________________________________ >R-help at stat.math.ethz.ch mailing list >https://stat.ethz.ch/mailman/listinfo/r-help >PLEASE do read the posting guide! >http://www.R-project.org/posting-guide.htmlSecurity. http://clinic.mcafee.com/clinic/ibuy/campaign.asp?cid=3963
You might want to look into multiple imputation methods as an alternative (see work by Don Rubin and Joe Schafer). I know Schafer has a library for S-plus, but not sure about R.> -----Original Message----- > From: F Z [mailto:gerifalte28 at hotmail.com] > Sent: Monday, August 30, 2004 12:52 PM > To: chris at lordsutch.com; r-help at r-project.org; bates at stat.wisc.edu > Subject: Re: [R] FIML in lme > > Thanks to Douglas Bates and Christopher Lawrence for their responses. > Christopher is right, that is what I was asking about. I guess thatthere> is no implementation of FIML in R. Would this be a worthy method to > include > in R? I don't really use this method so I would say no but maybe some > people think in a different way (For example SAS users trying ot moveto> R)? > > Respectfully > > Francisco > > > >From: Chris Lawrence <chris at lordsutch.com> > >To: R-Help <r-help at r-project.org> > >Subject: Re: [R] FIML in lme > >Date: Sat, 28 Aug 2004 02:29:08 -0500 > > > >On Aug 27, Douglas Bates wrote: > > > F Z wrote: > > > >I was asked if lme can use FIML (Full Information MaximumLikelihood)> > > >instead of REML or ML but I don't know the answer. Does anybodyknow> >if > > > >this is implemented in R? > > > > > > To the best of my knowledge, FIML is ML so the answer is yes. > > > > > > For example, the phrase "Full Information Maximum Likelihood" isused> in > > > Singer and Willett (2004) "Applied Longitudinal Data Analysis"(Oxford> > > University Press) as a synonym for maximum likelihood. > > > >I have seen FIML used to refer to a type of ML estimation where a > >missing data treatment is included in the estimation procedure > >(parameter estimates are derived from incomplete cases for only the > >variables present in the case, rather than simply discarding the > >cases), at least in the latent-variable SEM context, specifically in > >AMOS. This may be what Francisco is getting at. > > > >To my knowledge, no R packages implement this sort of "FIML", for any > >class of models, although there are other available missing data > >treatments (EM, MCMC estimation). > > > > > >Chris > >-- > >Christopher N. Lawrence, Ph.D. > >Visiting Assistant Professor of Political Science > >Millsaps College > >1701 N. State St > >Jackson, MS 39210 > >(601) 974-1438 / lawrecn at millsaps.edu > > > >______________________________________________ > >R-help at stat.math.ethz.ch mailing list > >https://stat.ethz.ch/mailman/listinfo/r-help > >PLEASE do read the posting guide! > >http://www.R-project.org/posting-guide.html > > > Security. http://clinic.mcafee.com/clinic/ibuy/campaign.asp?cid=3963 > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting- > guide.html
I'm not sure you are correct on this. Other texts on multilevel models (e.g., Raudenbush and Bryk, Kreft and Deeuw, and Singer & Willett) all use FiML as a synonym for ML. In fact, Kreft and Deleeuw go as far to even state they are the same thing (see page 131). When you run a model in HLM selecting "Full Maximum Likelihood" and method="ML" in lme, the results, including all fixed effects, variance components, empirical bayes residuals, degrees of freedom are exactly the same. So, I think Doug is correct in that ML == FiML. Harold -----Original Message----- From: r-help-bounces at stat.math.ethz.ch [mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of F Z Sent: Monday, August 30, 2004 12:52 PM To: chris at lordsutch.com; r-help at r-project.org; bates at stat.wisc.edu Subject: Re: [R] FIML in lme Thanks to Douglas Bates and Christopher Lawrence for their responses. Christopher is right, that is what I was asking about. I guess that there is no implementation of FIML in R. Would this be a worthy method to include in R? I don't really use this method so I would say no but maybe some people think in a different way (For example SAS users trying ot move to R)? Respectfully Francisco>From: Chris Lawrence <chris at lordsutch.com> >To: R-Help <r-help at r-project.org> >Subject: Re: [R] FIML in lme >Date: Sat, 28 Aug 2004 02:29:08 -0500 > >On Aug 27, Douglas Bates wrote: > > F Z wrote: > > >I was asked if lme can use FIML (Full Information MaximumLikelihood)> > >instead of REML or ML but I don't know the answer. Does anybodyknow>if > > >this is implemented in R? > > > > To the best of my knowledge, FIML is ML so the answer is yes. > > > > For example, the phrase "Full Information Maximum Likelihood" isused in> > Singer and Willett (2004) "Applied Longitudinal Data Analysis"(Oxford> > University Press) as a synonym for maximum likelihood. > >I have seen FIML used to refer to a type of ML estimation where a >missing data treatment is included in the estimation procedure >(parameter estimates are derived from incomplete cases for only the >variables present in the case, rather than simply discarding the >cases), at least in the latent-variable SEM context, specifically in >AMOS. This may be what Francisco is getting at. > >To my knowledge, no R packages implement this sort of "FIML", for any >class of models, although there are other available missing data >treatments (EM, MCMC estimation). > > >Chris >-- >Christopher N. Lawrence, Ph.D. >Visiting Assistant Professor of Political Science >Millsaps College >1701 N. State St >Jackson, MS 39210 >(601) 974-1438 / lawrecn at millsaps.edu > >______________________________________________ >R-help at stat.math.ethz.ch mailing list >https://stat.ethz.ch/mailman/listinfo/r-help >PLEASE do read the posting guide! >http://www.R-project.org/posting-guide.htmlSecurity. http://clinic.mcafee.com/clinic/ibuy/campaign.asp?cid=3963 ______________________________________________ R-help at stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html