Looking for better approach to do ttest

[Madhu6]

Hello Everyone,

I have two datasets Pr_table5 (test) and C_table5 (control).

Each table consists of CD3 tested on different time intervals (day 0 day3 day7 day14 day30) with different n on each day. For example, day0 N is 63, and day30 N is 50

I am calculating a t-test comparing Pr_table5 and C_table5 to get the following table:

T-Tests
Variable	Method	Variances	DF	t Value	Pr > |t|
mean	Pooled	Equal	XX	XX	XX
mean	Satterthwaite	Unequal	XX	XX	XX

So I calculated the mean for Pr_table5 (mean of day0, mean of day3...) and c_table5  (mean of day0, mean of day3...) and dropped n min max std so that it will have only mean from table1 and table 2.

proc means data = Pr_table5 noprint;
output out = a(drop = _type_ _freq_ );
run;
proc transpose data = a out = a (drop  = n min max std ) ;
id _STAT_;
run;

proc means data = c_table5 noprint;
output out = b(drop = _type_ _freq_ );
run;

proc transpose data = b out = b (drop  = n min max std ) ;
id _STAT_;
run;

Then I merged two tables  (combo) which contain means from two tables.

data combo_a;
set a ;
_name_ = 'mean_a';
run;
data combo_b;
set b ;
_name_ = 'mean_b';
run;
data combo;
set combo_a combo_b;
run;

Next, I did a t-test. 

proc means data=combo noprint;
var mean;
by _name_;
output out=new_combo;
run;
proc ttest data=new_combo nobyvar;
class _name_;
var mean;
run;

Please let me know if there is a better approach, from both a statistical and programming point of view.

Thank you,
Madhu

[Reeza · Accepted solution]

Can you do it directly on the raw data? SAS can analyze the difference between summarized data but this seems a bit weird to me. 

Are you certain the code posted works correctly because it seems like you'd only have a single observation and PROC TTEST expects multiple. Or it expects the more data from PROC MEANS, specifically the N and STD usually. 

data combo;
set pr_table5 c_table5 (indsname=source);
dset = source;
run;

*really don't think you need this step but if you think you do for some reason;
proc means data=combo nway stackods mean;
class dset;
ods output summary = combo_means;
run;

proc ttest data=combo_means;
class dset;
var mean;
run;

Here's an example of how to do PROC TTEST on raw data. 

https://stats.oarc.ucla.edu/sas/modules/an-overview-of-statistical-tests-in-sas/

If you have repeated measures as well, is there a reason you're aren't using a repeated measures type of analysis instead of just PROC TTEST?

https://documentation.sas.com/doc/en/statug/15.2/statug_glm_examples07.htm

[Madhu6]

Hello Reeza and everyone,

Thank you so much for your input. 

Please see the attachments for control and test datasets. Here is the code to import them.

proc import out=Control datafile='C:\Users\SDSP\Desktop\Control.txt';
run;
proc import out=Test datafile='C:\Users\SDSP\Desktop\Test.txt';
run;

Treatment : '1' is control and '2' is test.

Then I merged the two tables by treatment.

data final_table;
merge control Test;
by Treatment;
run;

Now I want to compare the two groups (control and test) and see if they are different:

I used the below code (I am sure it is not correct). Can someone please help me with this

proc glm;
class Treatment;
model Day0--Day720 =
Treatment */ nouni;
repeated 8 (0 3 7 15 30 90 180 360 720) polynomial / summary printe;
run;

Thank you,

Madhu