Comparing Partitions


"The history of statistics is replete with measures of association and correlation. Just because a measure has plausibility at face value, ranges from 1 down to 0, and has a known sampling distribution under an extreme null hypothesis of total randomness is no guarantee of its usefulness or interpretability. What does .50 means? And does .70 in one set of data really indicate greater association than .60 in another?"

David L. Wallace
Comment on Journal of the American Statistical Association
September 1983, volume 78, Number 382

The framework used in this website is presented in the article Carriço et al. (2006) for comparing several typing methods applied to a collection of Streptococcus pyogenes. The article proposes the use of Simpson's Index of Diversity, Adjusted Rand and Wallace coefficients as a quantitative way to evaluate the congruence of results between different typing methods.

This framework was further extended by Pinto et al. (2008) defining Wallace coefficient confidence intervals.

The use of resampling methods for confidence intervals estimation for pairwise agreement measures was evaluated in Severiano et al. (2011). In this study, simulations showed that the jackknife pseudo-values method performs better than the bootstrap at accurately estimating confidence intervals for pairwise agreement measures.

In this tutorial we provide a brief definition of the coefficients proposed and their use. You can also find here the ECCMID presentation of the first version of this website.

Share |

Design adapted from Peel Out Labels - Original Design downloaded from Free Templates - your source for free web templates