A method of cross validation in which models are assessed by calculating the sum of squares of differences between the observed values of a subset of the data and the relevant predicted values calculated from fitting a model to the remainder of the data.