Statistical inference is overemphasized in cluster investigations: the case of the cluster of breast cancers at the Australian Broadcasting Corporation studios in Brisbane, Australia
The aim of statistical analyses in cluster investigations is to estimate the probability that the aggregation of cases could be due to chance. As a result of several statistical problems – including the post-hoc nature of the analysis and the subjective nature of implied multiple comparisons – this cannot be carried out with any certainty. In cluster investigations, expert opinion should carry much more weight than P-values, which are exceedingly difficult to interpret.
Document Type: Research Article
Publication date: April 1, 2008