Hello everyone.
In addition to the equivalence test provided by JMP, is there a way to conduct non-inferiority tests through confidence intervals? By drawing the figure below, we are looking for a way to combine the equivalence test with the non-inferiority test, depending on whether the confidence intervals belong to a particular interval.
E.g Univariate A confidence interval plot like the one in Bivariate I'm wondering if it's applicable to
Thank you in advance.
In distribution there is Test Equivalence, and in fit y by x there is Equivalence Test.
This distribution platform has an updated graph that is really nice but the Fit Y by X hasn't been updated. (for years) The Fit Y by X graph for Equivalence test is very very confusing. Note that there are two tones of blue to indicate equivalent or not. There isn't an option to select which Level is the Reference Level so all pairs are reported.
However, there is an easy way to make the figure using Graph Builder and a simple report from Fit Y by X. In Fit Y by X, set alpha to 0.1, then get the means and standard deviations and make it into a data table.
From this table you have the means, standard deviations and 90% confidence intervals.
Use graph builder to plot Level by Mean, and then put the 90%CI into the interval role.
Then add reference lines to the graph. One level is the Reference (or Standard) sample. I add the difference to consider practically zero to the mean of the Reference level for the upper decision limit, and subtract for the lower decision limit.
If the confidence intervals of the Test Samples extend past the upper and lower decision
limit lines, then the Test Sample is not equivalent.
Both figures use the same data and same decision interval. On the right panel it is immediately obvious that TEST 1 is not equivalent to Reference. With a little more formatting It would be easy to make this graph look like the one in the Test Equivalence report from Distribution. Non-Inferiority essentially uses the same approach with a little different hypothesis.
Note the equivalence test in Fit Y by X does not show you what interval was used. I think it might be a two one sided T-test with a two tailed 95% CI (alpha 0.1) but you'd have to look that up somewhere.
