topic Re: Standard Error and Confidence Intervals in SAS Programming

Standard Error and Confidence Intervals

tompatmck — Fri, 13 Apr 2018 21:17:38 GMT

Hello,

I'm wondering why the standard errors (lsmeans/stderr) for proc glm are the same for means of different groups. The confidence limits (lsmeans/CL) are equal as well. In other statistical programs it seems that the errors are different for each group. Does it have something to do with how the variance is pooled?

Thanks.

Tom

Re: Standard Error and Confidence Intervals

art297 — Fri, 13 Apr 2018 21:24:38 GMT

Post some example data (in the form of a datastep) and the proc glm that you ran.

Art, CEO, AnalystFinder.com

Re: Standard Error and Confidence Intervals

tompatmck — Mon, 16 Apr 2018 14:27:52 GMT

Hello,

Thanks for the reply.

I have a randomized complete block design with treatment as a fixed effect and block as a random effect. There are 4 levels of treatment and 4 levels of block.

proc glm;

class block treatment;

model biomass = block treatment;

random block;

lsmeans treatment/stderr cl;

run;

quit;

The standard error for the lsmean is the same for all levels of the treatment effect. Also the confidence limits are equal for each treatment level.

Any more insight would be appreciated.

Thanks

Re: Standard Error and Confidence Intervals

PaigeMiller — Mon, 16 Apr 2018 14:37:52 GMT

@art297 clearly requested that you "Post some example data (in the form of a datastep)". I don't see the data.

If the number of observations is the same in each group (and other assumptions are met), you could get standard errors that are the same for each group.

Re: Standard Error and Confidence Intervals

art297 — Mon, 16 Apr 2018 14:46:40 GMT

In addition to what @PaigeMiller said, you may want to take a look at https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_glm_sect037.htm

PROC GLM may not be the proc that you want to be using for your model.

Art, CEO, AnalystFinder.com

Re: Standard Error and Confidence Intervals

tompatmck — Mon, 16 Apr 2018 14:47:12 GMT

Sorry about that. The design is balanced so there are equal reps in each treatment. I natural log transformed the data to meet assumptions. Here is some data and the code:

options formdlim = '-';
data GH_AC2015;
input Blk$ Pot$ Species$ Treatment$ Biomass Rootbiomass;
biot=LOG(Biomass);
roott=LOG(Rootbiomass);
cards;
1 6 AC Oven 0.2718 0.0703
1 7 AC Sterile 0.087 0.0459
1 13 AC Control 0.2551 0.0583
1 14 AC Rescue 0.117 0.0493
2 21 AC Sterile 0.072 0.0658
2 24 AC Control 0.1514 0.0413
2 26 AC Rescue 0.207 0.0398
2 29 AC Oven 0.1304 0.0485
3 33 AC Control 0.087 0.0204
3 38 AC Rescue 0.2012 0.0576
3 41 AC Sterile 0.0191 0.0196
3 44 AC Oven 0.1434 0.0797
4 52 AC Oven 0.2601 0.0976
4 55 AC Control 0.3554 0.0791
4 59 AC Rescue 0.2981 0.0652
4 64 AC Sterile 0.0834 0.0975

;

proc glm ;
class blk treatment ;
model biomass = blk treatment;
random blk;
lsmeans treatment/stderr cl adjust = tukey;
run;
quit;

Re: Standard Error and Confidence Intervals

PaigeMiller — Mon, 16 Apr 2018 14:52:23 GMT

So equal number of observations in each cell will produce equal stadnard errors in each cell.

You may have done a log transformation, but your GLM code doesn't use the log-data.