topic Re: SAS code for tukey tests in SAS Studio

SAS code for tukey tests

palominm — Thu, 12 Jul 2018 19:35:31 GMT

Hi.

I am trying to run a tukey test after the annova

with this code:

data new; set mpa;
proc glm;
class fabric laundry time temperature;
model deltae=fabric|laundry|time|temperature;
means fabric*laundry*time*temperature; means fabric*laundry*time*temperature;

proc glm;
class fabric laundry time temperature;
model deltae=fabric|laundry|time|temperature;
means fabric*laundry*time*temperature/tukey lines;
run;

I think i am mistaking the second part..

Can anyone please give me a light on that?

thanks

Re: SAS code for tukey tests

OliviaWright — Fri, 13 Jul 2018 19:43:01 GMT

Hi @palominm!

One issue that I see with your PROC GLM is that you have the pipes | in your model statement. Those should not be necessary. In all of the examples I see they just list their dependent variables with spaces in between the variables.

That could be an issue, however it may not be the only one. Did you received an error in the log when you ran this code?

Re: SAS code for tukey tests

BrianGaines — Fri, 13 Jul 2018 20:07:54 GMT

I may be missing something but the use of the bar (pipe) operator should not make a difference since it is equivalent to listing the factors separately.

Instead, I believe the issue is that the tukey option can only be used with main effects. As per the PROC GLM documentation for the MEANS statement, "multiple comparisons tests apply only to main effects," and for the TUKEY option it states that it "performs Tukey’s studentized range test (HSD) on all main-effect means in the MEANS statement."

So the code

means A B / tukey;
means A*B;

produces the same results as

means A B A*B / tukey;

since the interaction effect will be ignored by Tukey's test.

Re: SAS code for tukey tests

OliviaWright — Fri, 13 Jul 2018 20:09:42 GMT

Ah, I see that now. I never seen anyone use the pipes for this but you are correct!

Re: SAS code for tukey tests

BrianGaines — Fri, 13 Jul 2018 20:12:07 GMT

Yes, they are very convenient for factorial models! 🙂