Can the lognormal distribution be rehabilitated?