CODEWITHSUNDEEP
r
Manuel Weinkauf
Manuel Weinkauf

Reputation: 350

How to make lm interpret eval in formula

I was hoping to get lm() to accept a character string as formula input but it does not quite work. This works so far:

 Test<-as.data.frame(matrix(sample(1:100, 12), 4, 3))
 colnames(Test)<-c("Y", "X1", "X2")
 lm(Y~X1, data=Test)
 XT1<-"X1"
 lm(Y~eval(parse(text=XT1)), data=Test)

This works fine and produces the correct output. However, when I trie to have more than one parameter:

 lm(Y~X1+X2, data=Test)
 XT2<-"X1+X2"
 lm(Y~eval(parse(text=XT2)), data=Test)

This does not produce the same result. I wonder why since obviously the character strings are both interpreted correctly:

 parse(text=XT1)
 parse(text=XT2)

For the background: In the end this should work in a function, where a matrix of parameters is supplied and the XT2 string should be dynamically created from the colnames of the matrix, so I'd need a solution that works with XT2 of all possible lengths between 1 and (hypothetically) infinity.

Upvotes: 3

Views: 1926

Answers (2)

Sweetbabyjesus
Sweetbabyjesus

Reputation: 563

with eval(parse(x)) you execute the string x as code, so in this case you first add X1 and X2 creating 1 variable and then run the regression.

Upvotes: 2

Roland
Roland

Reputation: 132576

Do not use eval(parse()) until you are an advanced R user (and then you usually won't need it). Just use as.formula:

lm(as.formula(paste0("Y ~ ", XT2)), data=Test)

Note that a better strategy for your goal would be:

lm(Y ~ ., data=Test[, c("Y", "X1", "X2")])

Upvotes: 8

Related Questions