Uma alternativa para avaliar discordância entre duas medidas via modelo de regressão linear simples sem intercepto

Abstract

This paper proposes a statistical method to assess the disagreement between two normally distributed measures. This method consists to test the coefficient of a simple linear regression model without intercept. Artificial dataset from literature and simulated data were used to illustrate this method and compare it to other methods that are wrongly used (such as the paired t-test or the Pearson correlation coefficient). It is noteworthy that, whereas Kendall’s coefficient of concordance and the Intraclass correlation coefficient assess agreement between two quantitative measurements, this new method is an alternative to test the disagreement between these two measurements. This work also presents a computational implementation of the method in the free software R. Finally, this work showed that the proposed method fulfilled its role, even when other procedures (wrongly used) have failed.