topic Re: Test for p-value in SAS Procedures

Test for p-value

cpulala — Tue, 12 Sep 2017 13:09:08 GMT

Hi I wanted to report p-value for this table

The column variables are ordinal variables (4 levels)

The row variables are categorical variables (4 levels)

Table cells are reported as proportions for each categorical variable by column. The total number of observations for each column is different.

group1 group2 group3 group4 p-value

cat_var P

level 1 % % % % P1

level 2 % % % % P2

level 3 % % % % P3

level 4 % % % % P4

My questions are

(1) which test to use for the P-value (row variable by column variable) in orange color

(2) which test for the p-value in green color, to have the p-value for each level by column variable (cell value reported as proportion)

Thanks!

Re: Test for p-value

PaigeMiller — Tue, 12 Sep 2017 13:25:31 GMT

Please state the hyptheses you are trying to test and get p-values for.

Re: Test for p-value

cpulala — Tue, 12 Sep 2017 13:31:58 GMT

Thanks for the note.

I wanted to test

(1) whether cat_var is statistically significant between each comparison group.

(2) whether the value (proportion) for each level of cat_var is different from each other.

For question(1), usually chi-square by using proc freq for 'categorical var by categorical var' is used, however, I am not sure which test to use for ordinal variable by categorical variable.

Thanks.

Re: Test for p-value

Rick_SAS — Tue, 12 Sep 2017 13:33:30 GMT

What are you testing for? Association? Trend? Are you trying to model cat_var by group_var (or vice versa)?

I suggest you use counts instead of proportions and analyze the data with PROC FREQ. The documentation has many examples that you can read and see if any of them are similar to what you want to analyze. You can also find many many SAS proceedings papers that describe the many statistical tests in PROC FREQ. Search for

sas "proc freq" test site:sas.com/proceedings

Re: Test for p-value

cpulala — Tue, 12 Sep 2017 13:46:31 GMT

Thanks. Please see the previous response on the hypothesis.

I understood the proc freq for categorical variables, but not sure which test to use when ordinal variables were involved. I can deem the ordinal variable as categorical, but only when there is no specific way for the oridinal variable by categorical variable (over 2 levels).