I have been asked to perform a MMRM analysis using the the Kenward-Roger degrees of freedom approximation. The analysis is performed to compare 5 different treatments and is requested to include pairwise comparisons of their marginal means with Tukey-Kramer multiple comparison adjustment.
I have already done some MMRM analyses in SPSS, but I am not sure how to request Kenward-Roger df and how to perform Tukey-Kramer multiplicity adjustment. Can this be done in SPSS? Or is there an R program that I can plug in? Best, Andreas |
In SPSS, the Tukey-Kramer test is available for between-subject
factions -- see the following page on the IBM SPSS website:-- but not for within-subject factors. https://www-304.ibm.com/support/docview.wss?uid=swg21478616 SPSS appears to contain an option for the Kenward-Roger adjustment; see the IBM SPSS webpage below as well as looking in the SPSS syntax manual: http://www-01.ibm.com/support/docview.wss?uid=swg21477296 In R there is a package called pbkrtest that does the Kenward-Rogers approximation. An article that describes this package is: Halekoh, U., & Højsgaard, S. (2014). A kenward-roger approximation and parametric bootstrap methods for tests in linear mixed models-the R package pbkrtest. Journal of Statistical Software, 59(9), 1-30. A copy can be accessed here: https://www.jstatsoft.org/htaccess.php?volume=59&type=i&issue=09&paper=true The abstract for the article follows: |Abstract | |When testing for reduction of the mean value structure in linear |mixed models, it is common to use an asymptotic 2 test. Such |tests can, however, be very poor for small and moderate sample |sizes. The pbkrtest package implements two alternatives to such |approximate 2 tests: The package implements (1) a Kenward-Roger |approximation for performing F tests for reduction of the mean |structure and (2) parametric bootstrap methods for achieving the |same goal. The implementation is focused on linear mixed models |with independent residual errors. In addition to describing the methods |and aspects of their implementation, the paper also contains several |examples and a comparison of the various methods. Note that the article was published in 2014 and it is likely that the pkrtest has been modified (some are suggested in the article for future releases). The authors compare their R version of the rest with that provided by the SAS Mixed procedure. -Mike Palij New York Univeristy [hidden email] ----- Original Message ----- On Monday, September 12, 2016 8:12 AM, "Andreas Voelp" wrote: >I have been asked to perform a MMRM analysis using the the >Kenward-Roger > degrees of freedom approximation. The analysis is performed to compare > 5 > different treatments and is requested to include pairwise comparisons > of > their marginal means with Tukey-Kramer multiple comparison adjustment. > > I have already done some MMRM analyses in SPSS, but I am not sure how > to > request Kenward-Roger df and how to perform Tukey-Kramer multiplicity > adjustment. Can this be done in SPSS? Or is there an R program that I > can > plug in? > > Best, Andreas > > > > -- > View this message in context: > http://spssx-discussion.1045642.n5.nabble.com/Questions-on-mixed-effect-model-for-repeated-measures-MMRM-tp5733116.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > To manage your subscription to SPSSX-L, send a message to > [hidden email] (not to SPSSX-L), with no body text except > the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In reply to this post by Andreas Voelp
Hi Andreas,
The GENLINMIXED procedure offers the Satterthwaite approximation for computing df, but not Kenward-Roger (see the DF_METHOD keyword): http://www.ibm.com/support/knowledgecenter/SSLVMB_24.0.0/spss/advanced/syn_genlinmixed_build_options.html Similarly, it offers the sequential Bonferroni and sequential Sidak methods for multiple comparison adjustments, but not Tukey-Kramer (see the PADJUST keyword): http://www.ibm.com/support/knowledgecenter/SSLVMB_24.0.0/spss/advanced/syn_genlinmixed_emmeans_options.html It may be possible to compute these given the SPSS output, but I don't immediately see how. There are more gory details in the algorithms doc (pages 373-408 for generalized linear mixed models): ftp://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/24.0/en/client/Manuals/IBM_SPSS_Statistics_Algorithms.pdf Best, Alex ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In reply to this post by Mike
Quick clarification; in the technote Mike
points to for Kenward-Roger in SPSS:
http://www-01.ibm.com/support/docview.wss?uid=swg21477296 ... the reference to KENWARDROGER is in the context of SAS command syntax. Alex From: Mike Palij <[hidden email]> To: [hidden email] Date: 09/12/2016 09:32 AM Subject: Re: Questions on mixed-effect model for repeated measures (MMRM) Sent by: "SPSSX(r) Discussion" <[hidden email]> In SPSS, the Tukey-Kramer test is available for between-subject factions -- see the following page on the IBM SPSS website:-- but not for within-subject factors. https://www-304.ibm.com/support/docview.wss?uid=swg21478616 SPSS appears to contain an option for the Kenward-Roger adjustment; see the IBM SPSS webpage below as well as looking in the SPSS syntax manual: http://www-01.ibm.com/support/docview.wss?uid=swg21477296 In R there is a package called pbkrtest that does the Kenward-Rogers approximation. An article that describes this package is: Halekoh, U., & Højsgaard, S. (2014). A kenward-roger approximation and parametric bootstrap methods for tests in linear mixed models-the R package pbkrtest. Journal of Statistical Software, 59(9), 1-30. A copy can be accessed here: https://www.jstatsoft.org/htaccess.php?volume=59&type=i&issue=09&paper=true The abstract for the article follows: |Abstract | |When testing for reduction of the mean value structure in linear |mixed models, it is common to use an asymptotic 2 test. Such |tests can, however, be very poor for small and moderate sample |sizes. The pbkrtest package implements two alternatives to such |approximate 2 tests: The package implements (1) a Kenward-Roger |approximation for performing F tests for reduction of the mean |structure and (2) parametric bootstrap methods for achieving the |same goal. The implementation is focused on linear mixed models |with independent residual errors. In addition to describing the methods |and aspects of their implementation, the paper also contains several |examples and a comparison of the various methods. Note that the article was published in 2014 and it is likely that the pkrtest has been modified (some are suggested in the article for future releases). The authors compare their R version of the rest with that provided by the SAS Mixed procedure. -Mike Palij New York Univeristy [hidden email] ----- Original Message ----- On Monday, September 12, 2016 8:12 AM, "Andreas Voelp" wrote: >I have been asked to perform a MMRM analysis using the the >Kenward-Roger > degrees of freedom approximation. The analysis is performed to compare > 5 > different treatments and is requested to include pairwise comparisons > of > their marginal means with Tukey-Kramer multiple comparison adjustment. > > I have already done some MMRM analyses in SPSS, but I am not sure how > to > request Kenward-Roger df and how to perform Tukey-Kramer multiplicity > adjustment. Can this be done in SPSS? Or is there an R program that I > can > plug in? > > Best, Andreas > > > > -- > View this message in context: > http://spssx-discussion.1045642.n5.nabble.com/Questions-on-mixed-effect-model-for-repeated-measures-MMRM-tp5733116.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > To manage your subscription to SPSSX-L, send a message to > [hidden email] (not to SPSSX-L), with no body text except > the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Alex, thanks for the correction. I had
to re-read the paragraph
to see that most of the content applied to the
SAS procedure
and not SPSS (indeed, the point was the one
should compare the
Satterhwaite results in SPSS and SAS).
That will learn me.
-Mike Palij
New York University
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
In reply to this post by Andreas Voelp
As other have pointed out, the KR df approximation is not available in MIXED or GELINMIXED. It is available in SAS. It is generally superior the Satterthwaite's method (available in SPSS), especially for repeated measures design and complex designs. Technically, the KR df approximation corrects for the downward bias of the covariance matrix of the fixed and random effects, and then computes a Satterthwaite's df based on the inflated covariance matrix. Although it is computationally expensive, often times it is worth specifying the KR because it corrects for inflated Type I error rates. Ryan On Mon, Sep 12, 2016 at 8:12 AM, Andreas Voelp <[hidden email]> wrote: I have been asked to perform a MMRM analysis using the the Kenward-Roger |
Free forum by Nabble | Edit this page |