topic Re: Comparing means for statistical significance in Statistical Procedures

Comparing means for statistical significance

einstein — Wed, 13 Sep 2017 19:05:09 GMT

I have the average of two rating catgories among 3 age groups.

Is there a way to test if the means in Rating 1 are statistically different among the three age groups (is 9.03 statistically different from 9.45 and 8.75)?

And if there is also a way to test if the mean for rating 1 is statistically different from the mean for rating 2?

Thanks!

	Rating 1 (1-10)	Rating 2(1-10)
Age
18 – 34	9.03	8.66
35 – 54	9.45	7.56
55+	8.75	6.65

Re: Comparing means for statistical significance

PeterClemmensen — Wed, 13 Sep 2017 19:19:26 GMT

Do you have the original data generating these means?

Re: Comparing means for statistical significance

einstein — Wed, 13 Sep 2017 19:21:59 GMT

Yes, I have the original data that is computing these means.

Re: Comparing means for statistical significance

Reeza — Wed, 13 Sep 2017 21:12:08 GMT

Then consider looking at an ANOVA

Re: Comparing means for statistical significance

PGStats — Thu, 14 Sep 2017 04:29:06 GMT

Since your ratings have a finite range [1-10], you should consider using non parametric tests such as the Kruskal-Wallis test available with option Wilcoxon in proc npar1way.

If rating 1 and rating 2 have been measured on the same individual, you should consider testing rating 1 - rating 2 against zero with one of the location tests available in proc univariate such as the Wilcoxon signed rank test.

Re: Comparing means for statistical significance

einstein — Thu, 14 Sep 2017 15:02:52 GMT

Then consider looking at an ANOVA

I don't want to be dense, but if I use ANOVA, is this the correct way to get I what I need?

Be able to compare rating_1 vs. rating_2 for each age grouping?

proc anova;
class rating_1 rating_2;
model age=rating_1 rating_2 rating_1*rating_2;
means rating_1 rating_2 /t tukey sidak;
means rating_1*rating_2 /t tukey sidak;
format age age. ;
run;

Re: Comparing means for statistical significance

einstein — Thu, 14 Sep 2017 15:14:40 GMT

Since your ratings have a finite range [1-10], you should consider using non parametric tests such as the Kruskal-Wallis test available with option Wilcoxon in proc npar1way.

I just tried this and I think it worked! Assuming I am understanding this correct, this is letting me compare if the values within the age groupings for rating_1 and rating_2 are statistically significant from each other. Thanks!

  proc npar1way wilcoxon correct=no data=tables;
      class age;
      var rating_1 rating_2;
      exact wilcoxon;
      format age age. ;
   run;

This one below I'm not quite sure how to implement - would you happen to have some sample code to do this?

If rating 1 and rating 2 have been measured on the same individual, you should consider testing rating 1 - rating 2 against zero with one of the location tests available in proc univariate such as the Wilcoxon signed rank test.

Thanks!

Re: Comparing means for statistical significance

PGStats — Thu, 14 Sep 2017 17:25:56 GMT

Suppose you create a variable rating_diff = rating_1 - rating_2 then

proc univariate data=myData;
by age;
var rating_diff;
run;

and look for the location test results in the report.

Re: Comparing means for statistical significance

einstein — Thu, 14 Sep 2017 19:49:17 GMT

Got it! It worked!

Thank you so much!