The Bland-Altman diagram shows four types of data errors. These types are systematic errors (1) (average shift), (2) proportional error (trend), (3) inconsistent variability and (4) excessive or irregular variability. A Bland-Altman plot is a useful representation of the relationship between two variables coupled on the same scale. It allows you to perceive a phenomenon, but not to test it, that is, gives you no probability of error when making a decision on variables, as a test would. All standard methods of calculating the typical error are based on the assumption that the typical error of each subject has the same average size. While the typical error varies between subjects, statisticians say the data indicate heterossticity or uneven errors. In this situation, the analysis provides you with some kind of typical average error, which is too high for some subjects and too low for others. To get rid of heterosis, you must either perform separate analyses for subgroups of subjects with similar defects (males. B females), or find a way to transform the variable to make uniform the typical error of the transformed variable. The transformation of the protocol often makes the error uniform when larger values of the original variables present more errors. You should look for uneven errors in calculating insurance statistics. I explain, as on the calculation page. The two most important aspects of accuracy are reliability and validity.
Reliability refers to the reproducibility of a measurement. You simply quantify reliability by taking several measurements on the same themes. Poor reliability impairs the accuracy of a single measurement and reduces your ability to track changes in measurements taken at the clinic or in experimental studies. Validity refers to the agreement between the value of a measure and its actual value. You quantify validity by comparing your measurements with values as close as possible to actual values. Poor validity also affects the accuracy of a single measurement, and it reduces your ability to characterize the relationships between variables in descripti studies. Bland JM, DG Altman. The measurement agreement in method comparison studies.
Stat Methods Med Res 1999; 8: 135-60 OK, do we need the correlation coefficient? Why can`t we just use the typical error? Hmmm. Well, these two are certainly related, because a typical small error usually means a high correlation. But they also measure different things. The typical error is a simple measure of the variation within each subject, while the correlation coefficient tells us something about the reproducibility of the hierarchy of subjects in case of repetition. A high correlation means that subjects generally retain their same places between tests, while a low correlation means that they are all combined.