tests/test_fitted.plm.R

In plm: Linear Models for Panel Data

# Test of fitted.plm
# 
# 1) Does fitted.plm run with various inputs?
# 2) Does fitted.plm run / issue a warning if a coefficients gets (silently) dropped in estimated model
# 3) Sundry
#
# NB: Correctness of calculation in fitted.plm is not checked currently in this file
#
# NB: there is also a non-exported function fitted_exp.plm and an associated test file
#     which calculates the fitted values by subtracting the "overall" residuals
#     (of the respective "overall"/"outer" model).

library(plm)
data("Grunfeld", package = "plm")
form <- inv ~ value + capital

mod_pool <- plm(form, data = Grunfeld, model = "pooling")
mod_fe   <- plm(form, data = Grunfeld, model = "within")
mod_re   <- plm(form, data = Grunfeld, model = "random")
mod_be   <- plm(form, data = Grunfeld, model = "between")


######## 1) General tests with various inputs ######## 
plm:::fitted.plm(mod_pool)
plm:::fitted.plm(mod_fe)
plm:::fitted.plm(mod_re)
plm:::fitted.plm(mod_be)

plm:::fitted.plm(mod_pool, model = "pooling")
plm:::fitted.plm(mod_fe,   model = "pooling")
plm:::fitted.plm(mod_re,   model = "pooling")
plm:::fitted.plm(mod_be,   model = "pooling")

plm:::fitted.plm(mod_pool, model = "within")
plm:::fitted.plm(mod_fe,   model = "within")
plm:::fitted.plm(mod_re,   model = "within")
plm:::fitted.plm(mod_be,   model = "within")

# some fail in v1.4-0 and v1.5-14:
# "Error in ercomp.plm(object) : ercomp only relevant for random models"
# plm:::fitted.plm(mod_pool, model = "random") 
# plm:::fitted.plm(mod_fe,   model = "random")
plm:::fitted.plm(mod_re,   model = "random")
# plm:::fitted.plm(mod_be,   model = "random")

plm:::fitted.plm(mod_pool, model = "between")
plm:::fitted.plm(mod_fe,   model = "between")
plm:::fitted.plm(mod_re,   model = "between")
plm:::fitted.plm(mod_be,   model = "between")


######## 2) Testcase with dropped coefficient in estimated model ######
# add linear dependent columns to dataset. Those columns get dropped in estimation
Grunfeld$lin_dep_col <- Grunfeld$capital
Grunfeld$lin_dep_col2 <- 2 * Grunfeld$lin_dep_col
form2 <- update.formula(form, . ~ . + lin_dep_col)
form3 <- update.formula(form, . ~ . + lin_dep_col + lin_dep_col2)

mod_pool_droppedCoef <- plm(form2, data = Grunfeld, model = "pooling")
mod_fe_droppedCoef   <- plm(form2, data = Grunfeld, model = "within")
mod_re_droppedCoef   <- plm(form2, data = Grunfeld, model = "random")
mod_be_droppedCoef   <- plm(form2, data = Grunfeld, model = "between")

mod_pool_dropped2Coef <- plm(form3, data = Grunfeld, model = "pooling")

mod_pool_droppedCoef$aliased
mod_fe_droppedCoef$aliased
mod_re_droppedCoef$aliased
mod_be_droppedCoef$aliased
mod_pool_dropped2Coef$aliased


# Below:
# Some of these failed due to dropped coefficients, was fixed in rev. 184
# [and some due to an additional error with ercomp.plm]

fitval_mod_pool_droppedCoef <- plm:::fitted.plm(mod_pool_droppedCoef)
fitval_mod_fe_droppedCoef   <- plm:::fitted.plm(mod_fe_droppedCoef)
fitval_mod_re_droppedCoef   <- plm:::fitted.plm(mod_re_droppedCoef)
fitval_mod_be_droppedCoef   <- plm:::fitted.plm(mod_be_droppedCoef)

# formal test of same results
if (!identical(fitval_mod_pool_droppedCoef, plm:::fitted.plm(mod_pool)))
  stop("not identical")
if (!identical(fitval_mod_fe_droppedCoef, plm:::fitted.plm(mod_fe)))
  stop("not identical")
if (!identical(fitval_mod_re_droppedCoef, plm:::fitted.plm(mod_re)))
   stop("not identical")
if (!identical(fitval_mod_be_droppedCoef, plm:::fitted.plm(mod_be)))
  stop("not identical")

plm:::fitted.plm(mod_pool_droppedCoef, model = "pooling")
plm:::fitted.plm(mod_fe_droppedCoef,   model = "pooling")
plm:::fitted.plm(mod_re_droppedCoef,   model = "pooling")
plm:::fitted.plm(mod_be_droppedCoef,   model = "pooling")

# some of these were all NA before rev. 605:
plm:::fitted.plm(mod_pool_droppedCoef, model = "within")
plm:::fitted.plm(mod_fe_droppedCoef,   model = "within")
plm:::fitted.plm(mod_re_droppedCoef,   model = "within")
plm:::fitted.plm(mod_be_droppedCoef,   model = "within")
# formal test
if (all(is.na(plm:::fitted.plm(mod_pool_droppedCoef, model = "within"))))
  stop("all values are NA")


# plm:::fitted.plm(mod_pool_droppedCoef, model = "random") # "Error in ercomp.plm(object) : ercomp only relevant for random models"
# plm:::fitted.plm(mod_fe_droppedCoef,   model = "random") # "Error in ercomp.plm(object) : ercomp only relevant for random models"
# plm:::fitted.plm(mod_re_droppedCoef,   model = "random")
# plm:::fitted.plm(mod_be_droppedCoef,   model = "random") # "Error in ercomp.plm(object) : ercomp only relevant for random models"

### test with data that becomes linear dependent due to within transformation
data("Cigar", package = "plm")
Cigar.p <- pdata.frame(Cigar)
Cigar.p[ , "fact1"] <- c(0,1)
Cigar.p[ , "fact2"] <- c(1,0)

# linear dependent columns are silently dropped in these functions, thus they work
mod_pool_cigar <- plm(price ~ cpi + fact1 + fact2, data = Cigar.p, model = "pooling")
mod_fe_cigar   <- plm(price ~ cpi + fact1 + fact2, data = Cigar.p, model = "within")
mod_pool_cigar$aliased
mod_fe_cigar$aliased
plm:::fitted.plm(mod_pool_cigar)
plm:::fitted.plm(mod_pool_cigar, model = "within")
plm:::fitted.plm(mod_fe_cigar)
plm:::fitted.plm(mod_fe_cigar, model = "within")


######## 3) Sundry 
# this gave an error pre rev. 522 (drop = FALSE was missing):
wi <- plm(inv ~ value, data = Grunfeld, model = "within")
plm:::fitted.plm(wi, model = "between")