Using distance covariance for improved variable selection