topic Re: Comparing row proportions from an individual column of a Chi- sq test for significance? in Statistical Procedures

Comparing row proportions from an individual column of a Chi- sq test for significance?

supreets89 — Tue, 30 Jan 2018 17:41:36 GMT

Hi Everyone,

How can I test for significance using proportions in the following data:

Var X Y Z

A 50 60 10

B 100 200 300

If i want to show that the total proportion of Z in A (10/120) is different from Z in B (300 /600), which test of significance would I use and how would i code it? My data is structured the following way:
v1 v2 v3
A X 50
A Y 60
A Z 10
B X 100

B Y 200

B Z 300

I know Chi-sq would compare proportions in all three columns together, but how can i focus the significance test to just Z in A vs Z in B.

Re: Comparing row proportions from an individual column of a Chi- sq test for significance?

PaigeMiller — Tue, 30 Jan 2018 17:50:52 GMT

You could compare the two proportions using the formula here:

https://onlinecourses.science.psu.edu/stat500/node/55

just underneath the heading "Hypothesis Testing to Compare Two Population Proportions: p1,p2" (this assumes that groups A and B are independent, statistically)

You could use PROC FREQ to obtain the proportions, and then the formula itself isn't hard to program in a data step.

Re: Comparing row proportions from an individual column of a Chi- sq test for significance?

StatDave — Tue, 30 Jan 2018 18:54:47 GMT

All you care about is Z and not Z - the separate X and Y levels can be combined. PROC FREQ can do the chi-square test for you:

data a;
input r $ c $ count;
datalines;
A notZ 110
A Z     10
B notZ 300
B Z    300
;
proc freq;
table r*c / chisq;
weight count; 
run;

Re: Comparing row proportions from an individual column of a Chi- sq test for significance?

Ksharp — Wed, 31 Jan 2018 13:41:03 GMT

Or you want do Trend Analysis ?

proc freq;

table v1*v2/trend;

run;

Re: Comparing row proportions from an individual column of a Chi- sq test for significance?

sld — Wed, 31 Jan 2018 17:31:17 GMT

I think that the solution suggested by @SAS_Rob in this current Community thread Chi square test of association which links to this note http://support.sas.com/kb/22/565.html would provide you with the tools you need.