topic Re: formally test difference of two distributions in Statistical Procedures

formally test difference of two distributions

quiphan — Tue, 11 May 2021 03:04:04 GMT

hello,

I have two distributions A and B as per attached spreadsheet.

Using Sas stat, which procedure will test they are similar or different?

thanks in advance for your help.

Qui

Re: formally test difference of two distributions

Reeza — Tue, 11 May 2021 03:17:32 GMT

If you had the raw numbers I'd say PROC FREQ + Chi squared. If you don't I think you can try PROC NPAR1WAY.
You'll need to restructure your data so you have it in one column and another variable identifying your groups.

See this tutorial here to see if matches what you're trying to do:
https://stats.idre.ucla.edu/sas/faq/how-can-i-test-for-equality-of-distribution/

Re: formally test difference of two distributions

Ksharp — Tue, 11 May 2021 12:18:18 GMT

You could try Kolmogorov-Smirnov test is 0.0164.
This supports rejection of the null hypothesis
that the distributions are the same for the two samples.

Check its Example :
Example 85.2: EDF Statistics and EDF Plot

ods graphics on;
proc npar1way edf plots=edfplot data=Arthritis;
class Treatment;
var Response;
freq Freq;
run;
ods graphics off;

@Rick_SAS wrote a blog about it before .

Re: formally test difference of two distributions

FreelanceReinh — Tue, 11 May 2021 13:02:45 GMT

@Ksharp wrote:
You could try Kolmogorov-Smirnov test is 0.0164.
This supports rejection of the null hypothesis
that the distributions are the same for the two samples.

Check its Example :
Example 85.2: EDF Statistics and EDF Plot

Just to avoid a possible misunderstanding: The p-value 0.0164 refers to the example from the PROC NPAR1WAY documentation, not to @quiphan's sample data. Without knowing anything about the sample sizes, i.e., based on relative frequencies alone, no statistical test can determine whether the differences between the two empirical distributions are significant.

Re: formally test difference of two distributions

quiphan — Tue, 18 May 2021 06:01:58 GMT

Thanks everyone for the suggestions.

I ended up using proc ttest (paired) and npar1way to analyse the data.

Qui