topic Re: Chi-square test in a proc freq in Statistical Procedures

Chi-square test in a proc freq

juanvg1972 — Sun, 18 Nov 2018 19:12:45 GMT

Hi,

I am using 'proc tabulate' procedure to know whether are dependencies between two vars in a SAS dataset

This is my proc:

PROC FREQ DATA = tabsas.customer_churn_telco4;
 
TABLES rango_tenure * Churn /CHISQ;
 
RUN;

I attached the output of the freq. If I take a look to the output I see dependecies between 'churn' and 'rango_tenure' vars.

The distribution of 'churn' var change between groups of 'rango_tenure' values.

I see that p-value is under 0.05 then I can say that there is a dependency, is that true??

The value of chi-square test is high, Can I say that the bigger the value the stronger the dependency?, is that true??

If I use more vars in proc tabulate checking the dependecies with 'churn', I can see that the bigger value in chi-square test gives the most important var (the stronger dependency)?

Thanks in advance

Re: Chi-square test in a proc freq

Reeza — Sun, 18 Nov 2018 19:23:01 GMT

Churn is usually measured with a survival type analysis but it's possible you have restrictions that make this impossible. In that case, I think it would still be a more appropriate test to use the Cochran-Armitage test, which looks for a trend. A Chi Square just tells you that the data is different at some levels, not that there's necessarily any trend.

There's an example in the documentation that is similar to this.

However, I will come back to the survival aspect of churn. If you don't have event data for a portion of your population it makes sense to use survival type analysis especially if they can leave at some point later on, and basically just have not yet.

Re: Chi-square test in a proc freq

juanvg1972 — Sun, 18 Nov 2018 21:42:22 GMT

Thanks Reeza,

I will consider survival analysis in the future, but know I am interested in understandinf chi-square test in proc tabulate

Can you help me with my questions about chi-square results??

Thanks

Re: Chi-square test in a proc freq

juanvg1972 — Mon, 19 Nov 2018 08:09:18 GMT

I think 'survival analysis' is not my solution, because I am trying to get likelihood of churn (not years until churn) based on input vars, I am considering to use logistic regression or decision trees..., then for me is very importante dependencies and the results of chi-square test

Re: Chi-square test in a proc freq

Reeza — Mon, 19 Nov 2018 15:55:10 GMT

https://www.sas.com/content/dam/SAS/en_ca/User%20Group%20Presentations/Toronto-Data-Mining-Forum/Chen-SurvivalModel-March2012.pdf