Negative values of the intraclass correlation coefficient are not theoretically possible.

In their methodological review of the indices of reproducibility, Tammemagi et al. [l] mentioned the intraclass correlation coefficient, which, they say, varies “from 1 for perfect disagreement, to 0 for random agreement, and to +l for perfect agreement.” However, such an interpretation of negative values of this statistic, although previously asserted hy Deyo et al. [12], is misleading. We argue that negative values are not theoretically possible although possibly ohservahle as explained below. In the framework of a two-way layout model, we define