[R] dynlm question: How to predefine formula for call to dynlm(formula) call

[Achim Zeileis]

On Sat, 18 Apr 2009, Ron Burns wrote:

> I want to set up a model with a formula and then run dynlm(formula) because I 
> ultimately want to loop over a set of formulas (see end of post)
>
> R> form <- gas~price
> R> dynlm(form)
>
> Time series regression with "ts" data:
> Start = 1959(1), End = 1990(4)
> <snip>
>
> Works OK without a Lag term
>
> R> dynlm(gas ~ L(gas,1))
>
> Time series regression with "ts" data:
> Start = 1959(2), End = 1990(4)
> <snip>
>
> Works OK with a Lag with this type of call
>
> R> form <- gas~L(gas,1)
> R> dynlm(form)
> Error in merge.zoo(gas, L(gas, 1), retclass = "list", all = FALSE) :
> could not find function "L"
>
> Does not work using a predefined formula with a Lag  (This type of call works 
> using dyn$lm from library(dyn))

The problem with "dynlm" is that it defines it's lag/diff functionality 
locally (unlike "dyn" which re-uses the usual lag/diff functions) and in 
the setting above this conflicts with the non-standard evaluation, 
unfortunately. I don't know a good solution to this...

> How do I make the call (or how do I setup form) so that this works in dynlm?

In your specific problem, I think it is worth to take the extra step and 
do the processing yourself because...

> To be specific the following is an example of what I was attempting to do:
>  m1 <- gas ~ L(gas,1)
> m2 <- gas ~ L(gas,1) + price
> m3 <- gas ~ L(gas,1) + price + d(gas)
> m4 <- gas ~ L(gas,1) + price + d(gas) + L(d(gas),1)

...these models correspond to different samples. m4 will lose one more 
observation at the beginning by lag+diff. Of course, it is possible to 
address this in dynlm as well but I (personally) find it simpler to do 
the data processing first and then the modeling and model selection. I 
would do something like:

## data processing
dat <- ts.intersect(gas, price,
    gas1 = lag(gas, k = -1),
    dgas = diff(gas),
    dgas1 = lag(diff(gas), k = -1))

## models
form <- list(
    gas ~ gas1,
    gas ~ gas1 + price,
    gas ~ gas1 + price + dgas,
    gas ~ gas1 + price + dgas + dgas1)

## fitting
mod <- lapply(form, lm, data = dat)

## evaluation
sapply(mod, AIC)
sapply(mod, AIC, k = log(nrow(dat)))

hth,
Z

> M <- c(m1,m2,m3,m4)
> A <- array(0,c(4,2))
>
> for(i in 1:4){
>  g <- dynlm(M[[i]])   ## works if use dyn$lm from library(dyn) and use 
> appropriate m's
>  A[i,1] <- AIC(g,k=2)
>  A[i,2] <- AIC(g,k=log(length(fitted(g))))
> }
> colnames(A) <- c("AIC","BIC")
> rownames(A) <- c("m1","m2","m3","m4")
> A
>
> -- 
>
> R. R. Burns
> Retired in Oceanside, CA
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
>