Evaluation of confidence limit estimates of cluster analysis on molecular marker data.

BACKGROUND Diversity studies employ cluster analysis as a statistical tool, whereby relationships between individuals are shown in a dendrogram, mostly accompanied by bootstrap support for merging branches to indicate confidence limits. The objective of this study was to evaluate the reliability of the currently applied method of obtaining confidence limit estimates in cluster analysis and to propose an improved alternative bootstrap method. RESULTS It was illustrated via a simulation study that conventional bootstrap support for cluster analysis was affected by the sample size. The reliability of merging branches decreased with increasing number of individuals in the sample. Unlike the current bootstrap support for cluster analysis, the proposed method provides confidence intervals for the similarity coefficients between individuals. To facilitate the interpretation of similarity coefficients and confidence intervals, alternative graphical presentations are proposed for both 'similarity coefficients' and 'confidence interval range'. CONCLUSION The proposed bootstrap method is not affected by the number of individuals in the sample.