Error in eval(predvars, data, env): invalid 'envir' argument of type 'closure'

Question

I am trying to do forward selection and am having difficulties using string formulas:

> df <- data.frame(x0=c(1,2,3), x1=c(3,2,1), y=c(1,0,1))
> f0 <- lm("y ~ 1", data=df)
> f1 <- formula(lm("y ~ .", data=df))
> step(f0, direction="forward", scope=f1)
Start:  AIC=-2.51
y ~ 1

Error in eval(predvars, data, env) : 
  invalid 'envir' argument of type 'closure'

I know I could just omit the string-formula portion if I wanted:

> f0 <- lm(y ~ 1, data=df)
> f1 <- formula(lm(y ~ ., data=df))
> step(f0, direction="forward", scope=f1)
Start:  AIC=-2.51
y ~ 1

       Df Sum of Sq     RSS      AIC
<none>              0.66667 -2.51223
+ x0    1         0 0.66667 -0.51223
+ x1    1         0 0.66667 -0.51223

Call:
lm(formula = y ~ 1, data = df)

Coefficients:
(Intercept)  
     0.6667

But I'd like to be able to dynamically name my dependent variable without having to hardcode it.

There’s absolutely no need to use strings if you want to “dynamically name [your] dependent variable”. You can construct formulas at runtime from expressions, e.g. `as.formula(bquote(.(var) ~ .))`, where `var = as.name('y')`. — Konrad Rudolph, Aug 19 '19 at 16:57
Also have a look at [Formula with dynamic number of variables](https://stackoverflow.com/q/4951442/10488504) — GKi, Aug 20 '19 at 15:03

score 2 · Accepted Answer · answered Aug 19 '19 at 16:52

You're almost there. You just need to add an as.formula function around your string formulae e.g.

df <- data.frame(x0=c(1,2,3), x1=c(3,2,1), y=c(1,0,1))
f0 <- lm(as.formula("y ~ 1"), data=df)
f1 <- formula(lm(as.formula("y ~ ."), data=df))
step(f0, direction="forward", scope=f1)

# make some string formulae objects
step0 <- "y ~ 1"
step1 <- "y ~ ."

# use as.formula
s0 <- lm(as.formula(step0), data=df)
s1 <- formula(lm(as.formula(step1), data=df))
step(s0, direction="forward", scope=s1)

GKi · Answer 2 · 2019-08-20T06:52:27.363

As @konrad-rudolph already suggested in the comments you can use bquote to have a dynamically name of the dependent variable in a regression like:

dependentVariable  <- as.name("y")
f0 <- lm(as.formula(bquote(.(dependentVariable) ~ 1)), data=df)
f1 <- formula(lm(bquote(.(dependentVariable) ~ .), data=df))
step(f0, direction="forward", scope=f1)

or if you don't mind using strings a solution close to @meenaparam answer:

dependentVariable  <- "y"
f0 <- lm(as.formula(paste0(dependentVariable," ~ 1")), data=df)
#f1 <- formula(lm(as.formula(paste0(dependentVariable," ~ .")), data=df)) #Does call lm
#f1 <- as.formula(paste0(" ~ ", paste(names(df)[!grepl(dependentVariable, names(df))], collapse="+"))) #Does not call lm
f1 <- reformulate(names(df)[!grepl(dependentVariable, names(df))]) #Or using reformulate which creates a formula from a character vector
step(f0, direction="forward", scope=f1)

The error comes from:

a <- lm("y ~ 1", data=df)
b <- lm(as.formula("y ~ 1"), data=df)
environment(formula(a)) #<environment: 0x56252c8a5fe0>
environment(formula(b)) #<environment: R_GlobalEnv>

And so the following minimum change to your steps will work:

f0 <- lm(as.formula("y ~ 1"), data=df) #as.formula is added here
f1 <- formula(lm("y ~ .", data=df))
step(f0, direction="forward", scope=f1)

Error in eval(predvars, data, env): invalid 'envir' argument of type 'closure'

2 Answers2