I want to remove from a linear regression formula all interaction that does not exist between two categorical variables.
The dataset has the following categories:
store, tv_brand, tv_size, tv_resolution, and tv_price.
The thing is that there are stores that only sell certain brands and brands that do not manufacture tv with certain resolutions. For that reason when I fit the values considering the interaction:
lm(log1p(tv_price) ~ tv_brand + tv_size + tv_resolution +
store * tv_brand + brand * resolution,
data = data)
The linear regression contains a lot of NA values for the missing interaction.
Even though these values NA does not affect the final linear regression, I would like to remove them because it might be confused with actual aliases (100% correlated Interaction) like a store that only sells a brand or something similar.
How can I create a linear regression where store * tv_brand interactions omit the nonexisting ones?
