Obsidian | Level 7

## proc freq - agree option for matched pairs

I have a health status scale measured at 2 different timepoints (baseline and 1 month) with 5 different response levels (0-4). I want to see if there is a statistically significant difference in within group change from baseline to the 1 month time point for each of these response levels. My questions are as follows:

1. Am I using the right code to accomplish this task?

2. If so, which statistics do I report to highlight if the difference is significant?

3. Finally, I think the procedure below gives an overall p-value and not a level-by-level p-value and statistics. How do I test a level         by level difference between baseline and 1 month response? Like the difference between response level 0 between baseline and     1 month.

I am running the following proc freq code (analogous to paired t-test for continous variables):

proc freq data=health;
tables hs_BASELINE*hs_1_MONTH / agree norow nocol nopercent;
run;

The result is as follows:

SAS Output

Frequency
Table of hs_BASELINE by hs_1_MONTH
hs_BASELINE hs_1_MONTH
0 1 2 3 4 Total
0
 327
 55
 25
 12
 17
 436
1
 236
 100
 37
 17
 14
 404
2
 140
 66
 42
 21
 22
 291
3
 106
 39
 36
 14
 21
 216
4
 62
 38
 33
 30
 67
 230
Total
 871
 298
 173
 94
 141
 1577
Frequency Missing = 325

 Statistics for Table of hs_BASELINE by hs_1_MONTH

Test of Symmetry
Statistic (S) 328.8670
DF 10
Pr > S <.0001

Kappa Statistics
Statistic Value ASE 95% Confidence Limits
Simple Kappa 0.1402 0.0139 0.1130 0.1674
Weighted Kappa 0.2379 0.0171 0.2044 0.2714

 Effective Sample Size = 1577Frequency Missing = 325

 WARNING: 17% of the data are missing.
Super User

## Re: proc freq - agree option for matched pairs

Was it repeated measure problem ?

Check PROC CATMOD for contingency table.

Especially the example in its documentation.

Discussion stats