Regression Loop by category - r

Question

I have a data set that has multiple engines and I want to create a for loop function to run a linear regression for each engine and to extract the coefficients of each regression. So I want the first regression to run across the six weeks with engine= Google and then the second one to run across the six weeks with engine =Bing, etc. A sample of the data set looks like this:
Engine Wk Imp Clicks lnSpend Actions CPA
google 1 100302 15791 10998 31 354.79
google 2 23893 4734 2866 16 179.18
google 3 318 16 37.83 11 3.44
google 4 7992 1980 1704.81 27 63.14
google 5 13206 3292 2732.13 26 105.08
google 6 10888 2966 2293.86 22 104.27
bing 1 23536 1808 1028.95 3 342.98
bing 2 86873 7196 2740.28 14 195.73
bing 3 54654 4398 1786.96 13 137.46
bing 4 45553 3353 1860.47 13 143.11
bing 5 41254 3322 1811.80 13 139.37
bing 6 38305 3117 1501.01 19 79.00
The regression equation is actions~ spend and this would remain constant across all the engines.
This is the code that I have so far:
for(i in unique(mydata$engine))
{
reg<- append(reg, lm(mydata$Actions~ mydata$lnspend, data=mydata[mydata$engine== i,]))
}
summary(reg)
However, when I do that, the regression runs on the full data set combining all of the engines together.
I also tried using a by function. The code that I have for that is
reg<- by(mydata$engine, function(mydata) lm(Actions~ lnspend, data=mydata))
sapply(reg, coef)
When I run that I get the following error:
"Error in unique.default(x, nmax = nmax)"
Any idea how to fix it?

Answer 1

Using data.table:
library(data.table)
mydata=data.table(mydata)
mydata[,as.list(lm(Actions~lnSpend)$coeff),by=Engine]
Engine (Intercept) lnSpend
1: google 17.632263 0.001318611
2: bing 4.735979 0.004341699

Answer 2

You can also make it work by adjusting your first for loop a bit:
Engine <- c('g','g','g','g','g','g','b','b','b','b','b','b')
Actions <- c(31,16,11,27,26,22,3,14,13,13,13,19)
lnSpend <- c(10998,2866,37.83,1704.81,2732,2293,1028,2740,1786,1860,1811,1501)
df <- data.frame(Engine,Actions,lnSpend)
reg <- c()
for (eng in unique(Engine)){
m <- lm(Actions~ lnSpend, data = df[which(df$Engine == eng),])
reg <- append(reg, m$coeff)
}
reg
# > reg
# (Intercept) lnSpend (Intercept) lnSpend
# 17.632629162 0.001318568 4.734059476 0.004344177

Answer 3

Your for/loop should work. You just did not filter your dependent and independent variables in for/loop by engine type, so R takes the full dataset:
Consider either explicitly referencing the subset filter in each variable:
for(i in unique(mydata$engine))
{
reg<- append(reg, lm(mydata$Actions[mydata$engine== i] ~ mydata$lnspend[mydata$engine== i],
data=mydata[mydata$engine== i,]))
}
Or leave anonymously so data argument dictates the structure:
for(i in unique(mydata$engine))
{
reg<- append(reg, lm(Actions ~ lnspend,
data=mydata[mydata$engine== i,]))
}
summary(reg)