To handle multicollinearity among predictors.

To reduce bias in estimates when sample sizes are small or separation occurs.

To increase the variance of the estimators.

To allow for non-linear relationships.