Practical Data Analysis with JMP®

Collinearity

In a multiple regression model it is also important that the X variables vary independently of each other within our sample. If any of the X variables is highly correlated with any subset of X's—a condition called multicollinearity or sometimes just collinearity—then the estimates of the regression coefficients will likely be distorted and sometimes lead to nonsensical results. Think of it this way: if the experimenters increased the slag content of the mixture every time they increased the cement component, they would never be able to determine how much the strength is affected by each ingredient.

Note two important aspects of collinearity. First, it is a property of a sample rather than the parent population. Whether or not the ...