Does data rounding-off influence reproducibility index estimates?

Abstract

The intraclass correlation coefficient (ICC) is the reproducibility index commonly used to assess the reproducibility level of continuous data. Such data are collected with finite precision (i.e., a finite number of decimal digits corresponding to the level of data rounding-off error). We therefore investigated the consequences of having finite precision for data on ICC estimates, thus comparing ICC estimates in presence of rounding-off to those associated with data with infinite precision. As long as the rounding-off level is less than 0.3sigmaX (where sigmaX is the standard deviation of the outcome measured), there is no practical consequence on the ICC estimate. Therefore, in most practical situations the consequences of rounding-off data on the estimation of reproducibility levels can be ignored.