HYPOTHESIS TESTING

Results

These results were obtained by generating data from a normal distribution with a fixed mean and standard deviation, and then varying the null hypothesis being tested so the means moved apart. For each hypothesised mean, 1000 sets of data were generated and tests performed on each, with the acceptance frequency plotted.

We see the MML test criteria exhibit a wider acceptance range for the null hypothesis than do the classical tests. For a given hypothesised parameter value, the MML test is more likely to accept the null. Now consider a χ² test:

Again we see MML returning to the null hypothesis more often. This was true in general of the MML criteria. Further results and discussion including the T- and F-Tests are available in the full thesis.