I'm performing a multiple linear regression about the price of alcohol. Two of the predictors are the categorical variable 'type of alcohol', the levels being 'beer', 'wine', 'spirits' and the numerical variable 'alcohol percentage', so there is an obvious relationship between the two predictors (for example, spirits will have higher alcohol percentage while beers will have a lower percentage).
By no surprise, the VIF values are 33 and 12, respectively, when using the 'car package', and the VIF value for 'spirits' is 18, while the other types are relatively acceptable, and the 'percentage of alcohol' remains at 12 with the 'DAAG package'.
Since the two predictors are obviously related, my first thought was to combine the two predictors by creating a factor variable that discretizes the alcohol percentage into intervals. However, I realized this is not plausible as some types of alcohols have overlapping data, blurring the line between them. Logically, it also makes sense that a wine of certain alcohol percentage will cost different from a spirit at the same percentage.
How should the predictors be modified to reduce the VIF?