Multiple Linear regression - Transformation on Percentage regressors



I am performing a MLR (Multiple Linear Regression) on data set that has percentage parameters along with some numerical and categorical variables.

However while checking the MLR model assumptions, i found that i need to do an Squared transformation on my percentage variable. I could have blindly used the squared transformation to improve my model. But squaring a parameter with percentage values does not make any sense to me, because they might not have same implications as and when we transform a numeric variable.

Any Suggestion’s on how to proceed further is highly appreciated.


I’m not aware of a requirement to square a percentage variable. If I assume that the percentage is independent (i.e. it isn’t a percentage of one of the other variables or of the dependent variable), then using a percent (0 to 100) is a simple linear transformation of the values of the orignal parameter, therefore there is no reason to square it.

However, consider that your goal is to best model the dependent variable. If you find a transformation of one of the independent variables improves the results, then you should consider using it. Just be careful when using the model to ensure you properly transform the independent data.