Looping regression commands - r

Question

I'm going to simplify my problem as much as to prove that I'm not just throwing my assignment at you guys. I really want to learn how to get loop to work with regressions.
Let's suppose I want to run two OLS but I don't want to type the same ols command twice OR add another series of commands into my script. This is because a) I actually have way more than two regressions and b) I want to try to code this as efficient as I can (I have tried copying and pasting same ols commands). Also, I'm not just running a simple OLS as I'm also running HAC estimator depending on the serial correlation and heteroskedasticity test.
The code that i have came up with so far is,
Packages
if (!require("lmtest")) install.packages("lmtest")
library("lmtest")
if (!require("sandwich")) install.packages("sandwich")
library("sandwich")
Data
data<-read.csv(file.choose())
x1<-data$x1
x2<-data$x2
x3<-data$x4
x5<-data$x5
x6<-data$x6
x7<-data$x7
y1<-data$y1
Regressions
reg1<-(y1 ~ x1 + x2 + x3 + x4)
reg2<-(y1 ~ x2 + x4 + x6 + x7)
p<-0.05
Loop
for (i in 1:2) {
#OLS#
ols[i]<-lm(reg[i])
#Breusch-Pagan Test#
bptest(ols[i],varformula = NULL, studentize = TRUE)
bpp<-bptest(ols[i])$p.value
if(bpp>p) hs<-F else hs<-T
#Breusch-Godfrey Serial Correlation Test#
bgtest(ols[i],order=2,order.by=NULL,type=c("Chisq"))
bgp<-bgtest(ols[i])$p.value
if(bgp>p) sc<-F else sc<-T
#HAC Estimator#
HAC<-vcovHAC(ols[i],order.by=NULL,prewhite=FALSE,adjust=TRUE,diagnostics=FALSE,sandwich = TRUE,ar.method = "ols")
if (sc==T|hs==T) coeftest(ols[i],vcov.=HAC) else ols[i]
if (sc==T|hs==T) write.csv(coeftest(ols[i],vcov.=HAC),file="ols[i]HAC.csv") else write.csv(summary(ols[i])$coefficient,file="ols1.csv")
}
When I run this I get
Error in stats::model.frame(formula = reg[i], drop.unused.levels = TRUE) : object 'reg' not found
I have also tried the above code with
for (i in reg[1]:reg[2]) {
}
but it only returned
Error: object 'reg' not found.
Where did I go wrong?

Answer 1

This is too long for a comment, so I post it as a partial answer.
The difference seems to be the formula, and you are asking for a way to make your code more efficient. One way is to use a list of formulae and then combine the list with lapply. For instance,
reg <- list(
reg1 = as.formula(y1 ~ x1 + x2 + x3 + x4),
reg2 = as.formula(y1 ~ x2 + x4 + x6 + x7)
)
ols <- lapply(reg, function(x) lm(x, data=data))
Here, ols is a list of two elements, each of which is a regression corresponding to the formula-list. You can use the same principle for other functions, for instance:
bgtests <- lapply(ols, function(x)
bgtest(x,order=2,order.by=NULL,type=c("Chisq")))
This executes your bgtest function for each regression stored in ols. In a similar fashion, you can write it up so that it executes your heteroskedasticity corrections etc. The important point is this: you supply a list to lapply, and each element of that list is what is passed onto the function that you provide. The output of lapply is then a list with the output of that function.
In case you don't want to use lapply and to adress your actual question: the problem in your code is that there is no object called reg. Subsetting a non-existing object such as reg[1] hence does not work. If you execute the first lines of my code above, reg[1] and reg[2] become defined so that your loop should work.

Answer 2

The 'get' function is what you want, in conjunction with 'paste'. Below I fit two regressions using the cars data in R. Then I want to write a loop that extracts its coefficients. The 'get' function goes and find the object that matches the object name you specified.
> (reg1 <- lm(dist ~ speed, data = cars))
Call:
lm(formula = dist ~ speed, data = cars)
Coefficients:
(Intercept) speed
-17.579 3.932
> (reg2 <- lm(dist ~ 1 + I(speed^2), data = cars))
Call:
lm(formula = dist ~ 1 + I(speed^2), data = cars)
Coefficients:
(Intercept) I(speed^2)
8.860 0.129
> coeff <- matrix(0, nrow = 2, ncol = 2)
> for (i in 1:2)
+ {
+
+ # Main step
+ model <- get(paste("reg", i, sep = ""))
+ coeff[i,] <- coefficients(model)
+ }
> coeff
[,1] [,2]
[1,] -17.579095 3.9324088
[2,] 8.860049 0.1289687
>