Getting confidence interval in PROC MIXED

[tinaning]

Hello all,

I'm running mixed models using PROC MIXED.

Here are my codes:

proc mixed data=try covtest;
	class month pid;
	model outc=age male month|treatment / s;
	random int / subject=pid;
run;

Results:

                                                              Standard
Effect                       month   Estimate      Error     DF   t Value   Pr > |t|

treatment                              4.8044     0.7133   12E3      6.74     <.0001
treatment*month                  4    -4.6793     1.1857   12E3     -3.95     <.0001
treatment*month                  5    -3.3753     0.8922   12E3     -3.78     0.0002
treatment*month                  6    -2.9591     0.8697   12E3     -3.40     0.0007
treatment*month                  7    -0.3250     0.8824   12E3     -0.37     0.7126
treatment*month                  8          0          .      .       .        .

I want to get the treatment effects in each month, i.e. 4.8 + (-4.7) = 0.1 in month 4, 4.8 + (-3.4) = 1.4 in month 5. Is there any way to get confidence intervals for the calculated values (0.1, 1.4, etc)?

[Reeza]

Have you tried adding CL to your MODEL statement?

model outc=age male month|treatment / s CL;

https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_mixed_sect0...

CL	displays confidence limits for fixed-effects parameter estimates

[tinaning]

It only gives CI for each fixed parameter estimates, I need CI for the linear combination, i.e. 4.8+(-4.7)=0.1

[SteveDenham]

Use the LSMEANS statement, with a CL option.

proc mixed data=try covtest;
	class month treatment pid; /*notice that I have added treatment as a class effect */
	model outc=age male month|treatment / s;
	random int / subject=pid;
LSMEANS month*treatment/cl;
run;

Steve Denham

[tinaning]

Thanks Steve.

I tried your codes, LSMEANS show CI for the means of the outcome variable in each specified groups, not exactly what I was trying to calculate.

I want to get CI for 4.8+(-4.7)=0.1. It won't be significant if CI spans zero.

[lvm · Accepted solution]

Just add the diff option to the lsmeans statement.

[tinaning]

Great! I got it right, thanks Steve and lvm.

Here are my complete codes for share.

ods output diffs=diffdata;
proc mixed data=try covtest;
	class month pid treatment;
	model outc=age male month|treatment / s;
	lsmeans month*treatment / cl diff;
	random int / subject=pid;
run;