A regression method for the power–duration relationship when both variables are subject to error