Question about Tukey HSD (stat package)
On 14 Mar 2005 at 17:22, Peter Dalgaard wrote:
"JRG" <loesljrg at verizon.net> writes:
On 14 Mar 2005 at 16:13, Peter Dalgaard wrote:
Alice Le Bars <alice.le-bars at nantes.inserm.fr> writes:
Dear all, I would be glad if someone could tell me if Tukey HSD (function of stats library) accept the NA values and could correct the unbalanced design (different number of sample in each group)
I don't think the *theory* of the HSD allows this. There is some more or less well-founded speculation that if the imbalance is not too bad, then the HSD results are somewhat useful anyway, but please consider using the multcomp package, which is specifically designed to get these things right.
I believe that Hayter (JASA, 1984, 12, 61--75) proves that the so-called Tukey-Kramer method is guarnateed to be conservative for any set of cell sizes --- under the usual ANOVA model, of course.
Make that Annals of Statistics, not JASA. Hayter proves that simultaneous confidence intervals based on using the Studentized range distribution with (1/ni+1/nj)/2 instead of 1/n will have coverage at least (1 - \alpha). He doesn't say *how* conservative it is, though, and it doesn't change the fact that multcomp has a better idea.
Whoops, let it be Annals. However, there is a paper by Dunnett (ca 1981) which I again believe to be JASA which shows that the degree of conservativism is pretty slight, over a broad range of cell sizes. Your point about multcomp stands. ---JRG
-- O__ ---- Peter Dalgaard Blegdamsvej 3 c/ /'_ --- Dept. of Biostatistics 2200 Cph. N (*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918 ~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk) FAX: (+45) 35327907
John R. Gleason Associate Professor Syracuse University 430 Huntington Hall Voice: 315-443-3107 Syracuse, NY 13244-2340 USA FAX: 315-443-4085 PGP public key at keyservers