plm regression fixed effects on two variables, - r

Question

I have the following simplified df:
problem <- data.frame(
stringsAsFactors = FALSE,
fkeycompany = c("0000001961",
"0000003570","0000003570","0000003570",
"0000003570","0000003570","0000003570",
"0000003570","0000004187","0000004187","0000004187",
"0000004187","0000016058","0000022872",
"0000022872","0000022872","0000022872","0000024071",
"0000050471","0000052971","0000052971",
"0000056679","0000058592","0000058592","0000058592",
"0000063330","0000099047","0000099047",
"0000099047","0000316206","0000316537",
"0000319697","0000351917","0000351917","0000351917",
"0000356037","0000356037","0000356037",
"0000700815","0000700815","0000700815","0000700815",
"0000704415","0000704415","0000704415",
"0000705003","0000720154","0000720154","0000720154",
"0000720154"),
fiscalyear = c(2018,2002,
2002,2004,2006,2007,2007,2014,2005,2005,
2009,2017,2003,2002,2004,2004,2010,2002,
2016,2008,2008,2002,2005,2005,2010,2014,
2000,2005,2005,2002,2002,2001,2005,2005,
2006,2007,2012,2015,2006,2006,2007,2008,
2003,2014,2014,2000,2004,2006,2008,2013),
zmijewskiscore = c(-0.295998372490631,-3.0604522838509,-3.0604522838509,
-9.70437199970406,-0.836774487816746,
0.500903351523752,0.500903351523752,-1.29210741224579,
-1.96529713996165,-1.96529713996165,
-1.60831783946871,-2.12343231229296,-3.99767761748961,
0.561261861396196,4.13793269655047,4.13793269655047,
5.61803398400963,-0.000195582736436772,
-3.93766039340527,-0.540037039625719,
-0.540037039625719,-1.93767533120689,-4.54446419505987,
-4.54446419505987,1.94389244672183,
0.941272649148121,-3.88427264672157,-0.342812414189714,
-0.342812414189714,-1.35074505582686,
-4.52746658422071,-0.130671284507204,-0.223517713694019,
-0.223517713694019,0.0149617517859735,
-2.95100357094774,-2.55146691134187,-1.86846592111008,
2.92283100206773,2.92283100206773,
4.65325023636937,6.1585365469118,-4.54449586848866,
-1.49969162335521,-1.49969162335521,-3.34071706450412,
-1.72382101559976,-1.53076052307727,
-1.77582320023177,-1.57280701642882),
lloss = c(0,1,1,1,1,
1,1,1,0,0,0,1,0,0,1,1,1,1,0,1,1,
1,0,0,1,0,0,1,1,0,0,1,1,1,1,0,0,
1,1,1,1,1,0,1,1,0,1,1,1,0),
GCO_prev = c(1,1,1,0,0,
0,0,0,0,0,0,0,0,1,1,1,1,1,0,0,0,
0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,1,0,0,0,0,0,0,0,0),
GCO = c(0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,0,0,0,
0,0,0,1,1,0,0,0,0,0,0,0,0,1,0,0,
0,0,0,1,1,0,0,0,0,0,0,0,0),
industry = c(9,5,5,5,5,
5,5,5,6,6,6,6,9,9,9,9,9,6,9,6,6,
9,8,8,8,8,9,9,9,9,8,9,5,5,5,9,9,
9,6,6,6,6,9,9,9,9,9,9,9,9))
I would like to run a plm regression on this with fixed effects on year and industry.
library(plm)
summary(plm(GCO ~ GCO_prev + lloss + zmijewskiscore, index=c("fiscalyear", "industry"), data=problem, model="within" ))
However, I get this error while running:
Error in pdim.default(index[[1L]], index[[2L]]) :
duplicate couples (id-time)
In addition: Warning message:
In pdata.frame(data, index) :
duplicate couples (id-time) in resulting pdata.frame
to find out which, use, e.g., table(index(your_pdataframe), useNA = "ifany")
I do not quite know how to fix this. If I am assuming correctly, it has something to do with there being more companies (fkeycompany code) than 1 that have for example for industry = 9, fiscalyear = 2003 for example. So for some industries, lets say 9, there are more rows (fkeycompanies, in this example 0000016058 & 0000704415) which contain the year 2003 (or at least, thats what I think is the issue, or am I wrong?). This is with more industries and years the issue I believe in my main dataset. How do I fix this error message?
Also, besides this issue, is the code correctly that I am running? Am I indeed regressing with year and industry effects?

Answer 1

Given your data, the observational unit for panel data is firms (fkeycompany). You might want to add the industry as another fixed effect, but it cetainly is not the time index (the time index goes into the 2nd slot of argument index and I will assume it is fiscalyear). There are plenty of questions with answers to the topic. Also, do read the packages first vignette where the data specification for the index argument is explained.
I suggest to convert to pdata.frame first.
However, there are double constellations of fkeycompany and fiscal year, see below code where the output of table with a value > 1 hints you to the combinations.
library(plm)
pdat.problem <- pdata.frame(problem, index = c("fkeycompany", "fiscalyear"))
#> Warning in pdata.frame(problem, index = c("fkeycompany", "fiscalyear")): duplicate couples (id-time) in resulting pdata.frame
#> to find out which, use, e.g., table(index(your_pdataframe), useNA = "ifany")
table(index(pdat.problem), useNA = "ifany")
#> fiscalyear
#> fkeycompany 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2012 2013
#> 0000001961 0 0 0 0 0 0 0 0 0 0 0 0 0
#> 0000003570 0 0 2 0 1 0 1 2 0 0 0 0 0
#> 0000004187 0 0 0 0 0 2 0 0 0 1 0 0 0
#> 0000016058 0 0 0 1 0 0 0 0 0 0 0 0 0
#> 0000022872 0 0 1 0 2 0 0 0 0 0 1 0 0
#> 0000024071 0 0 1 0 0 0 0 0 0 0 0 0 0
#> 0000050471 0 0 0 0 0 0 0 0 0 0 0 0 0
#> 0000052971 0 0 0 0 0 0 0 0 2 0 0 0 0
#> 0000056679 0 0 1 0 0 0 0 0 0 0 0 0 0
#> 0000058592 0 0 0 0 0 2 0 0 0 0 1 0 0
#> 0000063330 0 0 0 0 0 0 0 0 0 0 0 0 0
#> 0000099047 1 0 0 0 0 2 0 0 0 0 0 0 0
#> 0000316206 0 0 1 0 0 0 0 0 0 0 0 0 0
#> 0000316537 0 0 1 0 0 0 0 0 0 0 0 0 0
#> 0000319697 0 1 0 0 0 0 0 0 0 0 0 0 0
#> 0000351917 0 0 0 0 0 2 1 0 0 0 0 0 0
#> 0000356037 0 0 0 0 0 0 0 1 0 0 0 1 0
#> 0000700815 0 0 0 0 0 0 2 1 1 0 0 0 0
#> 0000704415 0 0 0 1 0 0 0 0 0 0 0 0 0
#> 0000705003 1 0 0 0 0 0 0 0 0 0 0 0 0
#> 0000720154 0 0 0 0 1 0 1 0 1 0 0 0 1
#> fiscalyear
#> fkeycompany 2014 2015 2016 2017 2018
#> 0000001961 0 0 0 0 1
#> 0000003570 1 0 0 0 0
#> 0000004187 0 0 0 1 0
#> 0000016058 0 0 0 0 0
#> 0000022872 0 0 0 0 0
#> 0000024071 0 0 0 0 0
#> 0000050471 0 0 1 0 0
#> 0000052971 0 0 0 0 0
#> 0000056679 0 0 0 0 0
#> 0000058592 0 0 0 0 0
#> 0000063330 1 0 0 0 0
#> 0000099047 0 0 0 0 0
#> 0000316206 0 0 0 0 0
#> 0000316537 0 0 0 0 0
#> 0000319697 0 0 0 0 0
#> 0000351917 0 0 0 0 0
#> 0000356037 0 1 0 0 0
#> 0000700815 0 0 0 0 0
#> 0000704415 2 0 0 0 0
#> 0000705003 0 0 0 0 0
#> 0000720154 0 0 0 0 0
Once fixed, you will be able to run a model along these lines.
For a time-fixed-effects model:
model <- plm(GCO ~ GCO_prev + lloss + zmijewskiscore, data = pdat.problem, model="within", effect = "time")
Or time-fixed-effects with industry as an additional fixed effect:
model2 <- plm(GCO ~ GCO_prev + lloss + zmijewskiscore + factor(industry), data = pdat.problem, model="within", effect = "time")