topic Re: "Holm-Tukey" not "Tukey-Kramer" on GLIMMIX output in Statistical Procedures

"Holm-Tukey" not "Tukey-Kramer" on GLIMMIX output

jmorrison — Sat, 19 Aug 2017 16:58:10 GMT

Greetings:

I am doing multiple comparisons of least squares means in proc glimmix, with adjust=tukey, in SAS 9.4.

The documentation under 'LSMEANS Statement' for PROC GLIMMIX states the following:

"When you specify ADJUST=TUKEY and your data are unbalanced, PROC GLIMMIX uses the approximation described in Kramer (1956) and identifies the adjustment as "Tukey-Kramer" in the results."

In the output, however, the adjustment is labeled as "Holm-Tukey" not "Tukey-Kramer".

I can not find any reference for a Holm-Tukey test, and do not understand why it is labeled as such.

I need to report in a paper with a reference what test I used for the multiple comparison, and now I am not sure what it should be.

Does anyone have any helpful information about this discrepancy?

The data are proportion cover of plants in plots, analyzed as repeated measures of a factorial design with fixed and random effects.

Here are the first few lines of data (there are six forests, with 32 to 40 plots per forest, so it is unbalanced).

data cover;
input forest $ plot $ fencing $ miviadd $ year cover;

LBAL b1 fence addmivi 2012 0.11
LBAL b2 fence nomivi 2012 0.18
LBAL b3 nofence addmivi 2012 0.13
LBAL b4 nofence nomiv 2012 0.10
LBAL b1 fence addmivi 2015 0.13
LBAL b2 fence nomivi 2015 0.22
LBAL b3 nofence addmivi 2015 0.11
LBAL b4 nofence nomiv 2015 0.09

Here is the code for the LSMEANS step:

lsmeans forest*fencing*year / slicediff=(forest*year) CL ilink adjust=tukey plot=meanplot;

Here is the top of the output for this section:

Thanks!

Re: "Holm-Tukey" not "Tukey-Kramer" on GLIMMIX output

Reeza — Sat, 19 Aug 2017 17:37:26 GMT

1. What's your version of SAS STAT (proc setinit;run;)

2. What version of the documentation are you referring to?

Re: "Holm-Tukey" not "Tukey-Kramer" on GLIMMIX output

jmorrison — Sat, 19 Aug 2017 18:53:30 GMT

It is SAS 9.4...is there a different version within that for SAS/STAT?

The User Guide is SAS/STAT 12.3.

Thanks,

Janet

Re: "Holm-Tukey" not "Tukey-Kramer" on GLIMMIX output

Reeza — Sat, 19 Aug 2017 21:33:58 GMT

Yes, what does the proc setinit show for the SAS STAT.

It's likely 14.2 or 14.1. Check if the newer docs are the same.

Re: "Holm-Tukey" not "Tukey-Kramer" on GLIMMIX output

jmorrison — Sat, 19 Aug 2017 22:35:15 GMT

proc setinit just showed the expiration dates fot the components, but the sas support page for it said to use proc product_status to see the verison numners of each component. When I did that it said:

For SAS/STAT ...
Custom version information: 12.3 (maintenance)

So I would think the 12.3 doc I was looking at is correct for this version.

However, I looked anyway at the 14.2 User's Guide and it has the exact same language as 12.3.

Re: "Holm-Tukey" not "Tukey-Kramer" on GLIMMIX output

jmorrison — Mon, 21 Aug 2017 18:28:23 GMT

HI all,

I ended up asking SAS Tech Support about my question. FYI, here is the answer:

You've come across a known bug in the labeling of the effect comparison table. It should indeed be labeled as "Tukey-Kramer", and not "Holm-Tukey".

Janet