Thanks very much to everyone who sent me replies - I have included them
below. I'm particularly keen to find out about packages offering mixed
models for survival analysis or for analysis categorical data as these are
not covered in SAS - I'd be grateful if anyone could add any more on this.
Thanks
Helen
Principal Statistician
Room B028
Information and Statistics Division
Trinity Park House
South Trinity Road
Edinburgh
EH5 3AQ
Tel: 0131 551 8304
0131 554 1097
07718 932096
[log in to unmask]
Dear Helen,
Obviously SPlus and Genstat are alternatives. These are covered in the
second edition of my book Cross-over Trials in Clinical Research. You will
also find software examples at
http://www.senns.demon.co.uk/cticr2.html
I am glad to see that plans for the second edition are progressing.
Regards
Stephen
Helen
The multilevel project has a webpage with a review of several software,
please see: http://multilevel.ioe.ac.uk/softrev/index.html
Hope this helps
Andrea
By the way, I found your book excellent!
PLEASE NOTE NEW ADDRESS & CONTACT TELEPHONE NUMBER
*******************************************************
Andrea Manca
Research Fellow
Centre for Health Economics
ALCUIN A Block - A/116
University of York
York, UK
YO10 5DD
Tel: +44 (0)1904 321430
Fax: +44 (0)1904 433 644
E-mail: [log in to unmask]
http://www.york.ac.uk/inst/che/manca.htm
Economic Evaluation Workshop Seminar Series
http://www.york.ac.uk/inst/che/seminars.htm
[log in to unmask]
Helen:
The multi-level model website (http://multilevel.ioe.ac.uk - seems to be
down currently) has links to a number of reviews on this. I've certainly
used R and S-plus (the nlme package) to fit multilevel models successfully.
I've had less joy with SPSS 11, which also claims to fit multilevel models,
but I think is restricted to two levels - I struggle to produce useful
graphics.
Best wishes
Dr Ian Bradbury
School of Biomedical Sciences
University of Ulster at Coleraine
[log in to unmask]
For multilevel models try some of these software packages:
MlWin contact the Institute of Education about this.
HLM was written by Bryk, Raudenbash and Congdon (1996).
VARCL was written by Longford (1993).
MIXREG, MIXOR, MIXNO and MIXPREG are all multilevel modules that have
been constructed by Hedeker.
Hope this helps.
Paula Hammond
Senior Statistician
Statistics Research and Analysis Group
National Foundation for Educational Research
The Mere, Upton Park, Slough SL1 2DQ United Kingdom
Main line: +44 (0)1753 574123
Direct line: +44 (0)1753 637391
Main fax: +44 (0)1753 691632
[log in to unmask]
Web: http://www.nfer.ac.uk
[log in to unmask]
There is the lme function in S-PLUS
Brian Everitt
Professor Brian Everitt
[log in to unmask]
Box 20, Biostatistics Department,
Institute of Psychiatry,
Denmark Hill,
London SE 5 8 AF
UK
phone +44 (0) 20 7848 0309
fax +44 (0) 20 7848 0281
[log in to unmask]
R is a good one.
www.r-project.org
Cheers
Huan
Hi Helen,
I know a little about this, from having to fit linear and generalized linear
models to various ecological data lately, though I'm certainly not an expert
(and have probably been doing it all wrong...).
Apart from SAS, I've used GenStat and R. GenStat is certainly worth a look;
in recent versions, I'd say the mixed models stuff is one of its strengths.
On the datasets I've been working with, it is considerably faster and more
flexible than SAS, and it will fit a very wide range of models, including
mutivariate and generalized ones, and including Lee and Nelder's
hierarchical generalized linear models (which go beyond the usual Normal
assumptions for the random terms. Doesn't mean it's perfect in every way of
course --- e.g. it's been criticized for using chi-squared distributions for
Wald tests, for instance, rather than F. (I know this point is somewhat
controversial but I don't entirely understand the issues myself really, to
be honest.) If you don't know it you can download a trail version (lasts a
month) from their website at www.vsn-intl.com
R has an available package called nlme wich fits linear and nonlinear mixed
effects models, using the approach in the book by Pinheiro and Bates. The
package available for R is, as far as I know, a port to R of the S-Plus
package, and it doesn't do everything that the S-Plus version does, I
believe. (I haven't used the S-Plus version myself.) In my very limited
experience, it's very good at what it's specifically designed to do, but
that is in some ways is a bit limited. (For instance, if the random effects
are hierarchical it's easy to set up and runs very fast, but if, as in one
thing I tried to use it for, there are crossed random effects, it is clunky
to use and runs very slowly.) Fitting generalized linear mixed models in R
(or S, I think) is more of a pain, and though there's a good section on how
to do it (and on available packages) in the latest edition of Venables and
Ripley's book, I personally haven't had much success in getting it to work.
Hope this helps.
Regards,
Kevin McConway
Senior Lecturer in Statistics
Department of Statistics
The Open University
Walton Hall
Milton Keynes MK7 6AA, UK
Phone: +44-1908-653676
Fax: +44-1908-652140
email: [log in to unmask]
[log in to unmask]
Dear Mrs Brown,
I am sure you are aware that both S+ and STATA are implementing
mixed models methodology. From a Bayesian perspective BUGS is the
recommended package. There are also a few standard textbooks available
explaining in some detail how to use these programs to apply the mixed
model techniques.
Please keep me posted in case you find more info regarding this topic
Best wishes
Dr Dimitris N Lambrou
Statistician
Athens-Greece
[log in to unmask]
What about LISREL, MLwiN, and Stata? RH.
[log in to unmask]
Dear Helen,
You might try joining the Multilevel list (another JISCmail list) and posing
the question there. Multilevel has been in existence for many years and is
a discussion list specifically for multilevel modelling. It currently has
1661 subscribers though not all are active, and includes some of the leading
lights like Harvey Goldstein, Judith Singer, Joop Hox, Tom Snijders, Antony
Fielding, et al. You can see the list at
http://www.jiscmail.ac.uk/lists/multilevel.html
and access its archives, even as a non-member. You can join via the link on
the above page or see my help page at
http://www.nursing.teaching.man.ac.uk/staff/mcampbell/multilevel.html
Despite being the list owner, I know little about multilevel modelling (!),
although I am aware that MlWin, HLM, Stata and even SPSS support mixed or
multilevel models.
Best wishes,
Malcolm Campbell
Multilevel list owner
Dear Helen,
In answer to your request, I think my colleague Geert Verbeke and I, who
prepared a Springer text on longitudinal data (Verbeke and Molenberghs 2000,
Linear Mixed Models for Longitudinal Data, New York: Springer-Verlag), have
done a bit of work on (testing for) variance components. We thought that the
statement in your book as if a zero variance in the variance-covariance
matrix of the random effects would require some updating. We would like to
refer to our text, as well as to a paper on score testing in this respect.
Please do not hesitate to contact us for further information,
best wishes,
Geert
_____________________________
Geert Molenberghs
Biostatistics
Center for Statistics
Limburgs Universitair Centrum
Universitaire Campus, Building D
B-3590 Diepenbeek
Belgium
Tel: +32 11 26 8238
Secr: +32 11 26 8202
Fax: +32 11 26 8299
mobile: +32 476 35 4512
Email: [log in to unmask]
Web site: http://www.luc.ac.be/censtat
[log in to unmask]
Hi,
I use the package nlme for R (it is also available for S-plus). The
key reference is:
Mixed-Effects Models in S and S-PLUS (Springer, 2000), by Pinheiro
and Bates. It is a great package and the book is fantastic.
Cheers,
Simon.
[log in to unmask]
Helen
I don't have a copy of your book to hand today, so don't know what you
mention already. I hope that you cover GenStat, R and S-Plus as well as
SAS: I have used GenStat, SAS and S-Plus myself. GenStat has had a
directive called REML for a long time, probably longer than most packages
because it was developed in conjunction with the original inventors of
this approach (Desmond Patterson & Robin Thompson). It has been
continuously developed at Rothamsted in conjunction with the agriculture
people like Brian Cullis in New South Wales; the latest addition is the
VPREDICT directive in the new Seventh Edition of GenStat, which provides
the formation of predictions (or LSMEANS as SAS would refer to them) from
mixed models. I'm sure that Stata, Statistica, SPSS and MLWin must also
have facilities, though I heven't tried them myself.
Do you also cover nonlinear mixed-effect models? GenStat doesn't offer
these, but SAS and S-Plus do (Proc NLMIXED and Function nlme). PK
modellers have been fitting them for years using a package called NonMEM,
and this has a more modern version called WinNonMEM. I understand that
MLWin also offers them, but I don't have any first-hand experience.
I'm organizing a half-day meeting for the RSS Stat Comp Section in May on
nonlinear mixed-effects models, and I'm still looking for a speaker
knowledgeable about the approach used in R and/or S-Plus. I would be
grateful for any suggestions.
Best wishes
Peter
Peter Lane
Research Statistics Unit, GlaxoSmithKline
Hi Helen,
Thought I'd reply as a namesake (except for the 'e'). I work/have worked on
the MCMC functionality in the software package MLwiN and was going to
suggest 3 good avenues for you to learn more about other packages.
Firstly of course the multilevel mailing list is probably a better avenue
for your question than allstat.
Secondly the multilevel modelling project where I used to be based have been
doing a software review exercise (basically reviewing the multilevel models
functionality of many packages) with details underneath the multilevel
homepage (http://multilevel.ioe.ac.uk/) Note I just tried this link at it
may be down this morning.
Finally and specifically to MLwiN all our manuals/books can be downloaded
freely either from the above website or for some of the books my own
publications page (see footer).
Note I do not have the latest version of the user's guide on my page yet.
Hope this stuff will be helpful.
Best wishes,
Bill.
Dr William Browne work phone : (0115) 9514940
Lecturer in Statistics home phone : (0115)9513373
Mathematical Sciences mobile : (0779) 1577701
University of Nottingham (Note this no. is also new)
University Park E-mail :
[log in to unmask]
Nottingham NG7 2RD
web: http://www.maths.nott.ac.uk/htbin-local/staff.info?pmzwjb
web (MLwiN): http://multilevel.ioe.ac.uk/
web (Publ.): http://www.maths.nott.ac.uk/personal/pmzwjb/bill.html
[log in to unmask]
Helen
STATISTICA includes some mixed modelling capability in both the Variance
Components and GLM modules.
If you are interested, I would be happy to send you an evaluation copy of
STATISTICA so that you can consider including some details in the book.
Matt Coates
Technical Services Department
StatSoft Ltd.
Tel: +44 (0) 1234 341226
e-mail: [log in to unmask]
[log in to unmask]
Or you could use R or S-plus - see
J S Pinheiro and D M Bates, Mixed-Effects Models in S and S-PLUS, New
York, NY: Springer-Verlag 2000
Peter M Lee
[log in to unmask]
>From: "Brown, Helen" <[log in to unmask]>
>I'm interested in finding out more about packages other than
>SAS that can used for fitting mixed (or multilevel) models.
- Pinheiro and Bates' NLME library in R and S-Plus will fit linear and
nonlinear models and it has just been extended to handle generalized
linear mixed models.
- Rabe-Hesketh, Pickles and Skrondal's gllamm Stata program covers a
wide range of mixed models.
- Lee and Nelder's HGLM package in Genstat.
- Aitkin and Francis' GLIM 4 macros will fit exponential family 2-level
models with discrete random coefficients using nonparametric maximum
likelihood, and random intercept models with Normal random effects.
- SPSS now fits linear mixed models.
The Centre for Multilevel Modelling has a webpage at the Institute of
Education, London which reviews the features of many of these as well as
the specialist packages such as MLWin, HLM, Mplus, etc.
Nick
--------------------------------
Dr. Nick Sofroniou
Educational Research Centre
Saint Patrick's College
Drumcondra
Dublin 9
Republic of Ireland
--------------------------------
Helen,
Further to our conversation I can confirm that GenStat offers the capability
you describe. Indeed this is one of GenStat's major strengths.
As promised information on GenStat 7 and url linking to our download page:
You may download a 30 day trial version of GenStat from
http://www.nag.co.uk/local/downloads/downloadinfo.asp
This takes you to a page requesting information so we can identify you. You
need to accept an online license before downloading.
I am sure the developers of GenStat - VSN International, who will be very
keen to work with you. Please feel free to contact me again or Roger Payne:
Dr Roger W Payne
[log in to unmask]
Chief Science and Technology Officer,
VSN International,
5 The Waterhouse, Waterhouse Street,
Hemel Hempstead, Herts HP1 1ES, UK
Tel: +44-(0)1442-450230
Fax: +44-(0)870-1215653
Best Regards,
John
John HOLDEN
Sales Executive - Education, Research & Distributors
The Numerical Algorithms Group Limited
Wilkinson House
Jordan Hill Road
OXFORD
OX2 8DR
UK
Direct Line: +44 (0) 1865 518 052 E-mail:
[log in to unmask]
Switchboard: +44 (0) 1865 511 245 Fax: +44 (0) 1865 310 139
http://www.nag.co.uk/
Interested in keeping up with the world of technical computing?
If so, sign up to receive "NAGNews", an electronic newsletter containing
technical tips for developers, user application stories and up-to-date
information on NAG products. To register, go to
http://www.nag.co.uk/Local/NAGNews/index.asp.
lme in R and S-PLUS and GLMM in R can be used to fit multilevel
models. There is an article in the December 2003 R News
http://cran.r-project.org/doc/Rnews
on using R to fit longitudinal achievement models in education.
--
Douglas Bates [log in to unmask]
Statistics Department 608/262-2598
University of Wisconsin - Madison http://www.stat.wisc.edu/~bates/
[log in to unmask]
Dear Helen,
ASReml, samm (S-Plus function that calls ASReml) and GenStat (with has
ASReml as part of the program) fit very general and extensive mixed
models. They are marketed by VSN International.
website: http://vsn-intl.co.uk/
Click on Products then ASReml
Regards
Ari
[log in to unmask]
You might want to take a look at the following sites
1. BMDP
http://www.statsol.ie/bmdp/bmdp.htm
Modules e.g., take a look at BMDP %V and 8V for mixed
models.
2. SYSTAT
http://www.statsol.ie/systat/systat.htm
Hope these are of help.
Paul Johnson
http://www.biostatsoftware.com
[log in to unmask]
Dear Helen,
I have written a Stata program called gllamm
that uses adaptive quadrature to estimate
generalized linear mixed models and various
extensions, including models for ordinal
and nominal responses. See
http://www.gllamm.org
I would be happy to send you any of the papers
referred to there that you cannot get hold of.
Best wishes,
Sophia
Hi Helen
I saw your mail on allstat re: mixed models software.
I use GenStat and ASREML, both available from VSN International
http://www.vsn-intl.com/.
There is also a review of mixed models software (including GenStat
but not ASREML) by the multilevel models project at
www.multilevel.ioe.ac.uk
Hope this helps
Sue Welham
Biomathematics Unit
Agriculture and the Environment Division
Rothamsted Research
Harpenden UK AL5 2JQ
E-mail: [log in to unmask]
Tel: +44 (0)1582 763133 ext 2278
Fax: +44 (0)1582 467116
As part of my learning about mixed models and R I went through most of the
book trying to replicate your analyses in R (and odd ones in the MIX suite
of programs). In some cases I succeeded in replicating in others not. My
failures may well reflect the fact that I was learning about mixed models
and about R simultaneously, but some of them do reflect differences between
the dataset on your web site and the ones used in the book, or typos in the
book. I can send you
a) a list of discrepancies
b) my notes on fitting your datasets in R (about 70 pages)
if you are interested.
I could try to summarise the differences between R and SAS but you must be
able to find someone better than me to do that.
Michael Dewey
[log in to unmask]
http://www.aghmed.fsnet.co.uk/home.html
_________________________________________________________________
Common Services Agency Disclaimer
The information contained in this message may be confidential or legally privileged and is intended for the addressee only. If you have received this message in error or there are any problems please notify the originator immediately. The unauthorised use, disclosure, copying or alteration of this message is strictly forbidden.
_________________________________________________________________
|