New statistical methods to evaluate reproducibility

Abstract

This article proposes two alternative statistical methods to evaluate reproducibility and method error present in scientific papers that deal with quantitative measurements. Real data was used to demonstrate theirs use. These methods were presented by Lin, Bland and Altman. One of the advantages of these analyses over the traditionally used (the Dahlberg error, paired t test, Pearson's correlation coefficient) is that it is not required a normal distribution of the data, and only ten paired measurements should be enough. The method proposed by Dahlberg requires a minimum of 25 repeated measurements. The researcher must be aware of the implications that misinterpreted results or inconsistent conclusions may lead.

Reproducibility; Cephalometry; Statistics