topic Restrict statement equivalent in PROC MIXED? in SAS Programming

Restrict statement equivalent in PROC MIXED?

jvgatto89 — Tue, 13 Sep 2022 15:55:38 GMT

I am writing code to run a repeated measures linear mixed model. One of the linear predictors needs a fixed slope set to 1. The other predictor variable and the autocorrelation coefficient are estimated by the model. I have done something similar in PROC REG where there is a RESTRICT statement. Is there an equivalent to RESTRICT in PROC MIXED where I can restrict the slope to a constant?

Below is the code:

ODS OUTPUT FITSTATISTICS=FIT_RICKER2;
ODS OUTPUT CovParms=RICKER_PARMS2;
PROC MIXED METHOD=REML ALPHA=.05 NOITPRINT NOINFO DATA=FISH_DENSITY;
BY REGION SITE SPECIES;
CLASS TIME;
MODEL LOG_Nt=LOG_Nt_Minus1 Nt_Minus1/ s outp=pred_RICKER;
RANDOM TIME;
REPEATED / TYPE=AR(1) SUBJECT=INTERCEPT;
ESTIMATE 'INTERCEPT' INTERCEPT 1;
RUN;

Re: Restrict statement equivalent in PROC MIXED?

PaigeMiller — Tue, 13 Sep 2022 16:28:56 GMT

If you have to restrict the slope on a variable (let's just arbitrarily pick x2) to exactly 1, then you could write the fixed portion of the model

y = x0 + beta1*x1 + 1*x2 + beta3*x3 + ...

which is equal to

y–x2 = x0 + beta1*x1 + beta3*x3 + ...

so it would seem that you could fit a model to y–x2

Re: Restrict statement equivalent in PROC MIXED?

jvgatto89 — Tue, 13 Sep 2022 18:25:24 GMT

I cannot change the response variable. I am comparing it to other models and the response must be consistent across all model types.

Re: Restrict statement equivalent in PROC MIXED?

PaigeMiller — Tue, 13 Sep 2022 18:30:16 GMT

@jvgatto89 wrote:

I cannot change the response variable. I am comparing it to other models and the response must be consistent across all model types.

You can fit the model, and then "un-transform" the response variable, so you can compare it to other models.