Statistical analysis suggested that, for certain mixes, the drying method had an effect on the DI results, with the chloride conductivity test being the most sensitive.
The interaction effect between age and direction of reach or lean proved to be insignificant, with the frequency of CoP direction changes in anterior posterior direction during star excursion balance test being the only exception.
Both models were subject to rigorous validation procedures and the ultimate test being the use of test sets to determine the predictive power of the models.
All comparisons were examined for statistical significance using Pearson's chisquare test, being the threshold for significance p < 0.05.
We recreated all the statistical functions used in MEXPRESS in Javascript, with the Pearson correlation and the non-parametric Wilcoxon's rank-sum test being the two main functions.
There was considerable variation between tests in their ability to predict carcinogenicity, with the cell-transformation test and the bacterial-mutation test being the most accurate (94% and 93% accurate respectively).
