Convergence of test sizes for the Blackwelder’s non-inferiority test. The performance of two often used statistical procedures -- the classical asymptotic normal approximation and the same method with the Hauck-Anderson continuity correction -- to test non-inferiority between two proportions was studied. That study evaluates the performance of these two methods calculating the test sizes by enumerating all possible cases rather than through simulation, the hypothesis tests approach was used and was done for sample sizes until 300; the main conclusion in that work is that for these sample sizes, behavior of test sizes is erratic and uncontrolled, and its value is nearly always far above the nominal significance level. In that order of ideas, we consider important to know the performance of test sizes for big sample sizes $(nge 300)$. That is, we extended this analysis in this new research, having as our main goal to study numerically the performance of test sizes for these two statistical procedures, but now considering big sample sizes. Due to the fact the involved test sizes are big, some new additional computational difficulties were presented in this work, these difficulties were solved by implementing some theoretical properties in a program we wrote, in C++, to compute test sizes.

Keywords for this software

Anything in here will be replaced on browsers that support the canvas element