topic How to do customized hypothesis tests in proc rmstreg in Statistical Procedures

How to do customized hypothesis tests in proc rmstreg

yahousas — Wed, 22 Dec 2021 01:58:42 GMT

Using the RMSTREG procedure, I would like to know the group effect adjusting for sex. In the model, the interaction term is also included. The group has two categories, A and B.

The model is intercept+ beta1*group+beta2*sex+beta3*group*sex. For the overall group effect test, the hypothesis is beta1 (group effect in male) = beta1+beta3 (group effect in female)=0. which is same as beta1=beta3=0.

Is there any way to test the hypothesis ( beta1=beta3=0.) in Proc RMSTREG or manually in SAS? Does the estimate statement work for this? 'estimate' is colored in red in my code... I am not sure about the reason. and the ESTIMATE output showed nothing. My SAS version is 9.4.

This is the dataset

data rmst;
input Obs	time	status	group$	Sex$;
cards;
3	49	1	A	M
4	70	0	A	M
5	31	1	B	M
6	28	1	B	M
7	56	1	A	M
8	70	0	A	M
9	35	1	A	M
10	35	1	A	M
11	42	1	B	M
12	70	0	B	M
17	35	1	B	M
18	35	1	B	M
19	45	1	A	M
20	63	0	A	M
23	33	1	A	M
24	33	1	A	M
27	31	1	A	M
28	45	1	A	M
31	45	1	A	M
32	35	1	A	M
35	63	0	A	M
36	63	0	A	M
39	28	1	A	F
40	54	1	A	F
41	52	1	A	F
44	42	1	B	F
45	47	1	B	F
46	28	1	B	F
47	28	1	B	F
48	70	0	A	F
49	70	0	A	F
50	70	0	A	F
53	70	0	B	F
54	28	1	B	F
;
run;

Here is the code I tried.

proc rmstreg data=rmst;
class Sex group ;
lsmeans group*Sex/e diff;
estimate group 1 0 Sex*group -1 0 0 0;
model time*status(0) = group|Sex;
run;

Here is the ESTIMATE output.

Thank you!!!

Re: How to do customized hypothesis tests in proc rmstreg

sbxkoenk — Wed, 22 Dec 2021 09:02:20 GMT

Hello,

Could you sort it out meanwhile?

I don't think you need the ESTIMATE statement

, you need the TEST statement.

Something like :

TEST age1ht=0, age2ht=0;

Koen

Re: How to do customized hypothesis tests in proc rmstreg

yahousas — Wed, 22 Dec 2021 17:18:21 GMT

Thank you. I sorted out the data and tried the test statement but I got a syntax error message. Is there any way to fix it? or any other suggestions?

Here is the code I used. (I think you meant I need to dummy variables manually and use the test?, here is the reference, https://stats.oarc.ucla.edu/sas/faq/how-can-i-compare-regression-coefficients-across-three-or-more-groups/ )

data rmst2;
set rmst;
sex_f=.;
sex_m=.;
if Sex='F' then sex_f=1; ELSE sex_f=0;
if Sex='M' then sex_m=1; ELSE sex_m=0;
group_a=.;
group_b=.;
if group='A' then group_a=1;ELSE group_a=0;
if group='B' then group_b=1;ELSE group_b=0;

sfga=sex_f*group_a;
sfgb=sex_f*group_b;
smga=sex_m*group_a;
smgb=sex_m*group_b;
run;

proc rmstreg data=rmst2;
class sex_f sex_m group_a group_b sfga sfgb smga smgb;
model time*status(0) = sex_f sex_m group_a group_b sfga sfgb smga smgb ;
*lsmeans group*Sex/e diff;
test group_a=0,sfga=0 ;
run;

Here is the relevant log.

369  proc rmstreg data=rmst2;
NOTE: PROCEDURE RMSTREG used (Total process time):
      real time           0.00 seconds
      cpu time            0.00 seconds

NOTE: The SAS System stopped processing this step because of errors.
370  class sex_f sex_m group_a group_b sfga sfgb smga smgb;
371  model time*status(0) = sex_f sex_m group_a group_b sfga sfgb smga smgb ;
372  *lsmeans group*Sex/e diff;
373  test group_a=0,sfga=0 ;
                 -
                 22
                 76
ERROR 22-322: Syntax error, expecting one of the following: a name, ;, (, *, -, /, :, @,
              _CHARACTER_, _CHAR_, _NUMERIC_, |.
ERROR 76-322: Syntax error, statement will be ignored.
374  run;