|
Abstract:
|
Alphabetic Criteria, such as the D criterion, fail to take into account two key elements of screening experiments. First, as noted in Gilmour and Trinca, while alphabetic criteria minimize the variance of the effect estimate, they don’t take into account the variability of the estimate of sigma. Second, they fail to consider that in practice practitioners use individual confidence intervals to test for effects, first screening main effects, then later screening second order effects. Devoting more degrees of freedom to estimating the error both lowers the variability of sigma and the critical value of the t-statistic used to calculate the margin of error. A new criterion is proposed which corrects for these deficiencies.
|
