- it is unpractical to split data in many circumstances
- withhold or set aside each observation one by one from the sample and use remaining (n - 1) observations to estimate the coefficients for a particular candidate model
- fit the candidate model n-times
- estimation of the deleted observations each time results in n-prediction errors or PRESS residuals
- PRESS statistic is given as
- used to generate another R2-like statistic that reflects prediction capabilities, given as