Is there a short cut to test all variables besides one?-CodePudding

I'm trying to run a regression model for all two-way interactions and have about ~70 variables for each observation.

I have a certain variable, say z, that I want to model the main effects for but exclude from all two way interactions.

So right now I have something like this:

lm(y ~ .^2, data = d)

I'd like an easy way to do this:

lm(y ~ .   (not z)^2, data = d)

I know it's a lot of variables, I'm an academic researcher and just need to see what's significant when I run the model with everything in it. For my purposes, z makes sense as a main effect but not as an interaction from an intuitive perspective.

Thank you!

CodePudding user response：

As Ritchie Sacramento and I discussed in the comments, this should work:

lm(mpg ~ (. - carb)^2   carb, data=mtcars)

The . - carb would create the formula with all but the carb variable, which is then added afterwards.

For your data, it would be something like this:

lm(y ~ (.-z)^2   z, data = d)

CodePudding user response：

I think you can put this together with some string manipulations and reformulate.

Sample data (we're not even going to try to fit a model, just figure out how to construct the formula, so this should be OK).

dd <- data.frame(a = 1:3, b=1:3, c=1:3, d=1:3, e=1:3, f=1:3)

Let's suppose a is the response and f is the focal variable that you want to include only as a main effect.

v1 <- paste(setdiff(names(dd), c("a","f")), collapse = " ")
v2 <- sprintf("(%s)^2", v1)
form <- reformulate(c(v2, "f"), response = "a")
## a ~ (b   c   d   e)^2   f
colnames(model.matrix(form, data = dd))

results:

 [1] "(Intercept)" "b"           "c"           "d"           "e"          
 [6] "f"           "b:c"         "b:d"         "b:e"         "c:d"        
[11] "c:e"         "d:e"

Confirming that @FlapJack's answer also works:

 colnames(model.matrix(a ~ (. - f)^2   f, data = dd))
 [1] "(Intercept)" "b"           "c"           "d"           "e"          
 [6] "f"           "b:c"         "b:d"         "b:e"         "c:d"        
[11] "c:e"         "d:e"

(On the other hand, you could use my framework to do more complicated things like include/exclude variables on the basis of regular expressions ...)