Final Remarks

Abstract

1. Rank tests with ties: tie is rank number with multiple values. 2. Continuity correction for χ² test: if discrete variables are approximated by the normal distribution (= continuous), then a better approximation is obtained by adding +1/2 (or subtracting) to the probabilities, for example: χ² = (0 − E)²/E replace with = (|0 − E| − 0.5)²/E. 3. Unrequested alternative tests (contrast tests) are often included gratuitously in the analysis for example with the chisquare test). 4. Logistic regression: a lot of tables prior to the actual testing is produced. This is predominantly meant for assessing the goodness of fit of the so-called loglinear model. Look for more information in the textbook “Statistics Applied to Clinical Trials, Springer Dordrecht, 2009, by the same authors. If the data do not adequately fit the model, then the final analysis is often not executed anymore. So, you do not have to worry too much about this issue. 5. The four non-parametric tests are probably the most valuable tests for the analysis of data from clinical trials (Wilcoxon, Mann–Whitney, Friedman, Kruskall–Wallis), and they are particularly safe for those who are not fond on statistics but rather on solid results.