chapter  10
3 Pages

Regression model validation

WithMoh H. Malek, Jared W. Coburn, William D. Marelich

The validation of a regression model is important because it allows the investigator to compare the accuracy of the regression equation for the population used to develop the model. Traditionally, investigators have used data-splitting in which the sample is "split" into a derivation and validation group. In this approach, initially introduced by Mosier in 1951, the investigator randomly selects cases from the total data set and assigns each case into a derivation and validation group. The Predicted Residual Sum of Squares (PRESS) statistic is an attractive alternative to model validation, because it uses the entire data set and, therefore, avoids the need to split the data. When validating a new regression model or cross-validating an existing equation there are a number of indices which traditionally have been used in the exercise science literature. The constant error (CE) is an index which examines the difference between the predicted and observed value.