Partial least squares (PLS) is a data reduction method that can be useful when dealing with multicollinearity in multiple linear regression. PLS is often used in predictive analyses and discriminant analyses. The PLS algorithm creates new uncorrelated variables from old variables, but in a different way than PCA. PLS regression is considered one of the least restrictive extensions to the multiple linear regression model since it can be used in situations in which there are more predictor variables than observations.