[Rd] scoping/non-standard evaluation issue

peter dalgaard pdalgd at gmail.com
Wed Jan 5 00:04:55 CET 2011


On Jan 4, 2011, at 22:35 , John Fox wrote:

> Dear r-devel list members,
> 
> On a couple of occasions I've encountered the issue illustrated by the
> following examples:
> 
> --------- snip -----------
> 
>> mod.1 <- lm(Employed ~ GNP.deflator + GNP + Unemployed + 
> +         Armed.Forces + Population + Year, data=longley)
> 
>> mod.2 <- update(mod.1, . ~ . - Year + Year)
> 
>> all.equal(mod.1, mod.2)
> [1] TRUE
>> 
>> f <- function(mod){
> +     subs <- 1:10
> +     update(mod, subset=subs)
> +     }
> 
>> f(mod.1)
> 
> Call:
> lm(formula = Employed ~ GNP.deflator + GNP + Unemployed + Armed.Forces + 
>    Population + Year, data = longley, subset = subs)
> 
> Coefficients:
> (Intercept)  GNP.deflator           GNP    Unemployed  Armed.Forces  
>   3.641e+03     8.394e-03     6.909e-02    -3.971e-03    -8.595e-03  
>  Population          Year  
>   1.164e+00    -1.911e+00  
> 
>> f(mod.2)
> Error in eval(expr, envir, enclos) : object 'subs' not found
> 
> --------- snip -----------
> 
> I *almost* understand what's going -- that is, clearly mod.1 and mod.2, or
> the formulas therein, are associated with different environments, but I
> don't quite see why.
> 
> Anyway, here are two "solutions" that work, but neither is in my view
> desirable:
> 
> --------- snip -----------
> 
>> f1 <- function(mod){
> +     assign(".subs", 1:10, envir=.GlobalEnv)
> +     on.exit(remove(".subs", envir=.GlobalEnv))
> +     update(mod, subset=.subs)
> +     }
> 
>> f1(mod.1)
> 
> Call:
> lm(formula = Employed ~ GNP.deflator + GNP + Unemployed + Armed.Forces + 
>    Population + Year, data = longley, subset = .subs)
> 
> Coefficients:
> (Intercept)  GNP.deflator           GNP    Unemployed  Armed.Forces  
>   3.641e+03     8.394e-03     6.909e-02    -3.971e-03    -8.595e-03  
>  Population          Year  
>   1.164e+00    -1.911e+00  
> 
>> f1(mod.2)
> 
> Call:
> lm(formula = Employed ~ GNP.deflator + GNP + Unemployed + Armed.Forces + 
>    Population + Year, data = longley, subset = .subs)
> 
> Coefficients:
> (Intercept)  GNP.deflator           GNP    Unemployed  Armed.Forces  
>   3.641e+03     8.394e-03     6.909e-02    -3.971e-03    -8.595e-03  
>  Population          Year  
>   1.164e+00    -1.911e+00  
> 
>> f2 <- function(mod){
> +     env <- new.env(parent=.GlobalEnv)
> +     attach(NULL)
> +     on.exit(detach())
> +     assign(".subs", 1:10, pos=2)
> +     update(mod, subset=.subs)
> +     }
> 
>> f2(mod.1)
> 
> Call:
> lm(formula = Employed ~ GNP.deflator + GNP + Unemployed + Armed.Forces + 
>    Population + Year, data = longley, subset = .subs)
> 
> Coefficients:
> (Intercept)  GNP.deflator           GNP    Unemployed  Armed.Forces  
>   3.641e+03     8.394e-03     6.909e-02    -3.971e-03    -8.595e-03  
>  Population          Year  
>   1.164e+00    -1.911e+00  
> 
>> f2(mod.2)
> 
> Call:
> lm(formula = Employed ~ GNP.deflator + GNP + Unemployed + Armed.Forces + 
>    Population + Year, data = longley, subset = .subs)
> 
> Coefficients:
> (Intercept)  GNP.deflator           GNP    Unemployed  Armed.Forces  
>   3.641e+03     8.394e-03     6.909e-02    -3.971e-03    -8.595e-03  
>  Population          Year  
>   1.164e+00    -1.911e+00  
> 
> --------- snip -----------
> 
> The problem with f1() is that it will clobber a variable named .subs in the
> global environment; the problem with f2() is that .subs can be masked by a
> variable in the global environment.
> 
> Is there a better approach?

I think the best way would be to modify the environment of the formula. Something like the below, except that it doesn't actually work...

f3 <- function(mod) {
  f <- formula(mod)
  environment(f) <- e <-  new.env(parent=environment(f))
  mod <- update(mod, formula=f)
  evalq(.subs <- 1:10, e)
  update(mod, subset=.subs)
}

The catch is that it is not quite so easy to update the formula of a model. 

-- 
Peter Dalgaard
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com



More information about the R-devel mailing list