On the choice of statistical model for one-dimensional P-algorithms