How to compare binary IV subgroups on various levels of a multinomial DV

[cjia27]

Greetings to everyone reading this post,

Consider a simple univariate analysis where we are looking at two variables: anemia and "treatment".

• Anemia, the DV, consists of four subgroups: Grade 1, Grade 2, Grade 3, and Grade 4.
• Treatment, IV, consists of two subgroups: Treatment A, Treatment B.

The table I would like to fill looks like such below:

Anemia Grades	Total	'Treatment A'	'Treatment B'	p-value
Grade 1
Grade 2
Grade 3
Grade 4

I am struggling to determine how I am able to compare the counts between 'Treatment A' and 'Treatment B' for each Grade category to generate a p-value in this simplistic analysis. I do not need specific code (I prefer struggling through it), but some finger-pointing towards the correct way to do this would be SUPER appreciated.

Thanks!

Charles Jia

[Ksharp]

---

I do not need specific code (I prefer struggling through it), but some finger-pointing towards the correct way to do this would be SUPER appreciated.

---

I like your spirit of discovery . If you just want do a univariate BINOMIAL distribution analysis. Could try this one, you need to make a hypothesis H0: p=?  fristly.

The key code of analysis is:

proc freq data=have;
table sex/binomial(level='Female');
run;

the output should look like this:

Here is the code I suggested, good luck !

data have;
set sashelp.heart(obs=200);
keep smoking_status sex;
run;
proc sort data=have;by smoking_status;run;
ods select none;
ods output  BinomialTest= BinomialTest;
proc freq data=have;
by smoking_status;
table sex/binomial(level='Female');
run;
proc freq data=have noprint;
table smoking_status*sex/out=freq list;
run;
ods select all;

data BinomialTest2;
 set BinomialTest;
 if Name1='P2_BIN';
 keep smoking_status cValue1;
run;
data report1;
 merge freq BinomialTest2;
 by smoking_status;
run;
proc sql;
create table report2 as
select *,sum(count) as ntotal from report1 group by smoking_status;
quit;

proc report data=report2 nowd;
columns smoking_status ntotal count,sex cValue1;
define smoking_status/group;
define ntotal/group 'Total';
define cValue1/group 'P-Value';
define sex/across 'Treatment';
define count/analysis '';
run;

[StatDave]

To compare levels of your treatment variable, you could use the approach shown in this note that uses multiple PROC FREQ steps and then adjusts the p-values for multiplicity. Or you could use the model-based approach with a generalized logit model followed by the NLMeans macro to do the treatment comparisons as illustrated in this note - to estimate differences instead of ratios (relative risks), drop options=ratio in the NLMeans macro call.