Statistically Significant and Insignificant Adverse Effects

Abstract

The current chapter reviews many statistical hypothesis tests, adequate for assessing prevalence (proportion of a population) and rate data (frequency of events per time unit) versus control or versus zero.

Only independent adverse effects are assessed here, and they are tested for statistically significant presence.

Both attention is given to paired and unpaired data, and particular attention is given to explicit time dependent Poisson methods, as well as log likelihood ratio tests, that provide better power than traditional tests for the purpose.

Bayesian crosstabs may be prone to overdispersion, but in the example given no adjustment was needed.

Methods for analyzing contingency tables larger than 2×2 are given, and longitudinal data like survival data and Cox regressions are used for addressing times to event and computing hazard ratios of adverse effects.