Blackwell Reference Online

Extract

Cross‐validation is the process of estimating the expected criterion‐related validity of a battery of selection tests in a sample other than the one in which the original validities were estimated. When regression analysis is used to estimate test validity, regression weights are calculated to represent the optimal relationship between the selection tests and some measure of job performance . Due to measurement error, however, their accuracy will generally be less when they are applied to a new sample. This reduction in accuracy is called “shrinkage,” and it is important to know how much shrinkage to expect before using the tests in actual selection decisions ( Gatewood and Feild, 1994: 232 ). Two major approaches to cross‐validation exist. One method splits the validation sample into two smaller subsamples. The first of these subsamples (called a derivation subsample) is used to calculate the validity coefficients. The second sample (called a holdout subsample) is then used to estimate shrinkage ( Murphy, 1983 ). An alternative strategy to cross‐validation is to use the entire available sample for estimating predictors' validities, and then estimate the likely shrinkage by using formula estimation techniques. Such formulae provide mathematical estimates of expected validities if a prediction equation was cross‐validated in an infinite number of new samples ( Schneider and Schmitt, ... log in or subscribe to read full text