Numerical performance increased by finger training: A fallacy due to regression toward the mean?