thr3ads.net - R help - [R] question on multilevel modeling [Nov 2006]

If this information is useful, please help other people find it:
Share via:

Christine A. Calmes

2006-Nov-07 03:31 UTC

[R] question on multilevel modeling

Hi,
 
I am trying to run a multilevel model with time nested in people and
people nested in dyads (3 levels of nesting) by initially running a
series of models to test whether the slope/intercept should be fixed or
random.  The problem that I am experiencing appears to arise between the
random intercept, fixed slope equation AND.
 
(syntax:
rint<-lme(BDIAFTER~BDI+WEEK+CORUMTO, random=~1|DYADID/PARTICIP,
data=new) 
summary(rint))
 
the random slope, random intercept model 
 
(syntax: 
rslint<-lme(BDIAFTER~BDI+WEEK+CORUMTO, random=~CORUMTO|DYADID/PARTICIP,
data=new) 
summary(rslint))
 
at which point I obtain the exact same results for each model suggesting
that one of the model is not properly specifying the slope or intercept.
 
Or, I receive the following error message when I try to run the random
slope/random intercept model.
 
Error in solve.default(pdMatrix(a, fact = TRUE)) : 
        system is computationally singular: reciprocal condition number
= 6.77073e-017
 
(whether I receive an error message or the same results depends on the
specific variables in the model).
 
It has been suggested that I may need to change the default starting
values in the model because I may be approaching a boundary-is this a
plausible explanation for my difficulties?  If so, how do I do this in R
and can you refer me to a source that might highlight what would be
reasonable starting values?
If this does not seem like the problem, any idea what the problem may be
and how I might fix it?
 
Thank you so much for your assistance,
Christine Calmes
 
 
Christine A. Calmes, MA
Dept of Psychology
University at Buffalo: The State University of New York
Park Hall 216
Buffalo, NY 14260
ccalmes@buffalo.edu
(716) 645-3650 x578
 

	[[alternative HTML version deleted]]

Christoph Buser

2006-Nov-07 07:59 UTC

head link

[R] question on multilevel modeling

Dear Christine

I think the problem in your second model is that you are
including "CORUMTO" both as a fixed effect and as a random
effect. 

Regards,

Christoph 

--------------------------------------------------------------

Credit and Surety PML study: visit our web page www.cs-pml.org

--------------------------------------------------------------
Christoph Buser <buser at stat.math.ethz.ch>
Seminar fuer Statistik, LEO C13
ETH Zurich	8092 Zurich	 SWITZERLAND
phone: x-41-44-632-4673		fax: 632-1228
http://stat.ethz.ch/~buser/
--------------------------------------------------------------


Christine A. Calmes writes:
 > Hi,
 >  
 > I am trying to run a multilevel model with time nested in people and
 > people nested in dyads (3 levels of nesting) by initially running a
 > series of models to test whether the slope/intercept should be fixed or
 > random.  The problem that I am experiencing appears to arise between the
 > random intercept, fixed slope equation AND.
 >  
 > (syntax:
 > rint<-lme(BDIAFTER~BDI+WEEK+CORUMTO, random=~1|DYADID/PARTICIP,
 > data=new) 
 > summary(rint))
 >  
 > the random slope, random intercept model 
 >  
 > (syntax: 
 > rslint<-lme(BDIAFTER~BDI+WEEK+CORUMTO, random=~CORUMTO|DYADID/PARTICIP,
 > data=new) 
 > summary(rslint))
 >  
 > at which point I obtain the exact same results for each model suggesting
 > that one of the model is not properly specifying the slope or intercept.
 >  
 > Or, I receive the following error message when I try to run the random
 > slope/random intercept model.
 >  
 > Error in solve.default(pdMatrix(a, fact = TRUE)) : 
 >         system is computationally singular: reciprocal condition number
 > = 6.77073e-017
 >  
 > (whether I receive an error message or the same results depends on the
 > specific variables in the model).
 >  
 > It has been suggested that I may need to change the default starting
 > values in the model because I may be approaching a boundary-is this a
 > plausible explanation for my difficulties?  If so, how do I do this in R
 > and can you refer me to a source that might highlight what would be
 > reasonable starting values?
 > If this does not seem like the problem, any idea what the problem may be
 > and how I might fix it?
 >  
 > Thank you so much for your assistance,
 > Christine Calmes
 >  
 >  
 > Christine A. Calmes, MA
 > Dept of Psychology
 > University at Buffalo: The State University of New York
 > Park Hall 216
 > Buffalo, NY 14260
 > ccalmes at buffalo.edu
 > (716) 645-3650 x578
 >  
 > 
 > 	[[alternative HTML version deleted]]
 > 
 > ______________________________________________
 > 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
 > and provide commented, minimal, self-contained, reproducible code.

Dave Atkins

2006-Nov-07 14:50 UTC

head link

[R] question on multilevel modeling

Christine--

You have two and only two individuals per dyad; when you try to fit random 
slopes at level-2 (within couples), you are attempting to estimate an intercept,
slope, and covariance for each individual.  Basically, you don't have enough
data to do it.  If you restrict yourself to a single random-effect at level-2, 
you shouldn't have any problems.  So, it's not a code problem; it's
an issue of
the amount of data (or nature of your data) relative to the model you're 
attempting to fit.

You might want to see a paper I wrote on HLM with dyadic data (which has an 
appendix with R code); I discuss this issue some and various strategies for 
dealing with it:

Atkins, D. C. (2005).  Using multilevel models to analyze marital and family 
treatment data: Basic and advanced issues.  Journal of Family Psychology, 19, 
98-110.

Hope that helps.

cheers, Dave


Hi,

I am trying to run a multilevel model with time nested in people and
people nested in dyads (3 levels of nesting) by initially running a
series of models to test whether the slope/intercept should be fixed or
random.  The problem that I am experiencing appears to arise between the
random intercept, fixed slope equation AND.

(syntax:
rint<-lme(BDIAFTER~BDI+WEEK+CORUMTO, random=~1|DYADID/PARTICIP,
data=new)
summary(rint))

the random slope, random intercept model

(syntax:
rslint<-lme(BDIAFTER~BDI+WEEK+CORUMTO, random=~CORUMTO|DYADID/PARTICIP,
data=new)
summary(rslint))

at which point I obtain the exact same results for each model suggesting
that one of the model is not properly specifying the slope or intercept.

Or, I receive the following error message when I try to run the random
slope/random intercept model.

Error in solve.default(pdMatrix(a, fact = TRUE)) :
         system is computationally singular: reciprocal condition number
= 6.77073e-017

(whether I receive an error message or the same results depends on the
specific variables in the model).

It has been suggested that I may need to change the default starting
values in the model because I may be approaching a boundary-is this a
plausible explanation for my difficulties?  If so, how do I do this in R
and can you refer me to a source that might highlight what would be
reasonable starting values?
If this does not seem like the problem, any idea what the problem may be
and how I might fix it?

Thank you so much for your assistance,
Christine Calmes


Christine A. Calmes, MA
Dept of Psychology
University at Buffalo: The State University of New York
Park Hall 216
Buffalo, NY 14260
ccalmes at buffalo.edu
(716) 645-3650 x578

-- 
Dave Atkins, PhD
Assistant Professor in Clinical Psychology
Fuller Graduate School of Psychology
Email: datkins at fuller.edu
Phone: 626.584.5554

David Duffy

2006-Nov-08 21:19 UTC

head link

[R] question on multilevel modeling

"Christine A. Calmes" <ccalmes at buffalo.edu>
wrote:> 
> [...]  Basically, I am trying to test a
> psychological model in which a person's level of passive, negative
> communication with others in their environment (CORUMTO) on one week
> predicts their level of depression on the following week (BDIAFTER)
> controlling for their level of depression on the same week that that
> they co-ruminate (BDI).  The WEEK variable accounts for the fact that
> there are 7 weeks worth of data in the model and that scores on adjacent
> weeks are more closer related than scores on weeks that are further
> apart.
The alternative would be to fit using the sem package, explicitly
modelling the time-series and estimating the lagged effects of CORUMTO
and BDI -- it may be possible to disentangle the separate effects (a la
instrumental variables type analysis).  This would be easiest if there
are 7 rounds of observations for everyone.  The many time series packages may
also be usable for this.

David Duffy.

-- 
| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v

Maybe Matching Threads

Search for more reasonably related threads

R help - Nov 2006 - question on multilevel modeling

[R] question on multilevel modeling

[R] question on multilevel modeling

[R] question on multilevel modeling

[R] question on multilevel modeling

Maybe Matching Threads