Dear All,
I am trying to implement a regression with state-specific trends in R. I
can implement this in Stata with ease, but I am hoping to preserve my R
workflow. I suspect there is an "R formula trick" that I'm just
not
understanding and I would be grateful to anyone who could help me
understand "the R way."
I posted a similar question to Stackoverflow, but the solution from there
is not playing nice with plm() and I am hoping for a better answer,
http://stackoverflow.com/questions/34232834/fixed-effects-regression-with-state-specific-trends
My panel data looks like the following,
df <- data.frame(state = rep(c("a", "b", "c"),
each = 10), time = seq(1,
10, 1), stringsAsFactors = FALSE)
I would like to create state-specific trends that look like these three new
columns
df2 <- cbind(df, model.matrix( ~ state - 1, data = df) * df$time)
To achieve this in Stata, I would simply write
xi: i.state i.state*time
reg y x _I*
The problem I have with the SO model.matrix solution is that I have greater
than 200 "states." Currently, I am generating more than 200 trend
columns
with model.matrix and then I am building a very long formula string with
each column by name, I then turn that string into a formula with
as.formula(). Passing the resulting formula to lm() works but it is
cumbersome, and it is also choking the fixed effects version of plm().
Is there an R formula trick to this that I am missing?
Thank you,
Jake
[[alternative HTML version deleted]]