The results of data-driven evaluation of ground-motion models could be ambiguous if the test data are not independent of all the evaluated models. In such a case, the results describe both the explanatory and the predictive powers of the models. As an example, we demonstrate how a superseded ground-motion model appears to perform better than its successor, an antiintuitive result. We hope to raise the awareness of the seismichazard community on the importance of data independence when conducting and interpreting a data-driven evaluation of ground-motion models. The evaluation can still be useful even if test data cannot be made entirely independent. but the result should be interpreted with care.