Re: st: Repeated Measures ANOVA and Missing Values

["John L. Woodard, Ph.D." <john.woodard@wayne.edu>]

Thank you Cameron and Phil,

Phil, with xtmixed, can I analyze a split-plot type of design that includes a between-groups variable (e.g., group, but unequal sample sizes across groups) and the session repeated measures variable? Would the unbalanced design present problems?  I was also wondering what the syntax might look like for that design (one between groups variable, one repeated measures variable)?  Many thanks! 

John


On Oct 16, 2011, at 7:27 PM, Philip Ender wrote:

> I'm trying to run a oneway repeated measures ANOVA (the variable is =
> session, with three levels).  I have organized my data in long format, =
> with one column for subject ID, one column for session number and one =
> column for the dependent variable.  Here is a listing of part of my data =
> file:
> ...
> There are obviously several cases that are missing observations at =
> session 2 or session 3.  Does wsanova or Stata's anova command =
> automatically drop cases with missing values at any of the three =
> sessions?  My first pass with the analysis seems to suggest it does not.
> 
> What is the best way to deal with these missing values (either deleting =
> cases with missing values or telling wsanova or anova to ignore them in =
> the analysis)?
> -------------------------------
> John,
> 
> You are correct, -wsanova- does not drop cases that have missing
> observations.  If
> want complete case analysis you will have to drop observations manually.
> 
> However, if you want to use all of the observations I would recommend
> -xtmixed-.  It
> would look something like this in Stata 12:
> xtmixed DV i.session || ID:, reml
> testparm i.session
> 
> If you have lots of subjects you can use the chi-square value as is.
> If your sample is on
> the small size, you can rescale the chi-square to F by dividing it by
> its df (in your case 2).
> You can use the residual df from the -wsanova- as denominator df.  The
> -Ftail- function
> will get you the p-value.
> 
> This approach is not universally approved depending upon what
> discipline you come from.
> My Chicago Stata Users Group presentation on this topic can be found at
> http://www.stata.com/meeting/chicago11/materials/chi11_ender.pdf
> 
> -- 
> Phil Ender
> UCLA Statistical Consulting Group
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/