I have some questions on running a non-inferiority study outside of clinical trials.

First, I am having trouble determining a cut-score for the study I am running and cannot find direction in the literature other than the references to set the null proportion difference to -.10 or in a few cases -.05. I am struggling to determine the range of meaningful difference.

Second, I am measuring venue and whether collecting data in an alternate venues is "not appreciably worse" than the traditional standard for four different variables. For example, I measured the number of participant words per idea. Thus my data is continuous, and I am confused as to how to code this. Here is a truncated version of my data with the variables of interest being venue and wordidea:

data noninf; input subject $ gender $ state $ group $ venue $ ideas words wordidea; cards; 1 2 1 2 1 4 82 20.50 2 2 1 2 1 6 115 19.17 3 2 1 2 1 3 37 12.33 4 2 1 2 1 8 101 12.63 16 2 2 3 2 18 2217 123.17 30 1 3 1 3 6 485 80.83 ; run;

I tried reformatting my data:

data noninf; input v1_wi v2_wi; cards; 20.50 123.17 19.17 80.83; /*with more cards*/ run;

proc ttest data=noninf dist=normal tost(-10, 10); paired v1_wi*v2_wi; run;

with these results:

The TTEST Procedure

Difference: v1_wi - v2_wi

N Mean Std Dev Std Err Minimum Maximum

9 -26.2180 40.9299 13.6433 -113.4 26.7048

Mean 95% CL Mean Std Dev 95% CL Std Dev

-26.2180 -57.6795 5.2434 40.9299 27.6464 78.4122

TOST Level 0.05 Equivalence Analysis

Lower Upper Mean Bound 90% CL Mean Bound Assessment

-26.2180 -10 > -51.5884 -0.8477 < 10 Not equivalent

Test Null DF t Value P-Value

Upper -10 8 -1.19 0.8657 Lower 10 8 -2.65 0.0145 Overall 0.8657

Any help would be appreciated. Thank you, June