(described in Section 5.1); zero is a "good" value. For us, values for z0 for our five confidence intervals ranged from — 0.043 to 0.059, with a median of — 0.012. The correlation between observed and fitted values is a statistic that summarizes the model's predictive power. But the number computed from the data that gave rise to the model (0.70) can be overly optimistic. There are many approaches to getting around that optimism, a simple one being 10-fold cross-validation, as in . To implement 10-fold cross-validation one divides the data set at random into 10 distinct parts. Nine of them would be used to fit the generalized linear model, and the correlation coefficient between actual and fitted values would be calculated for the 10-th part, that is, the part that did not figure in the fitting. This would be repeated 10 times in succession, and the resulting 10 values of the correlation averaged. Other sample reuse methods can be used to accomplish the same task.
Was this article helpful?