It does this by comparing the prediction glitches of the two types more than a specific period. The take a look at checks the null hypothesis the two products hold the same general performance on normal, towards the choice that they don't. When the take a look at statistic exceeds a critical wor