pbsytest {plm} | R Documentation |
Bera, Sosa-Escudero and Yoon Locally–Robust Lagrange Multiplier Tests for Panel Models and Joint Test by Baltagi and Li
Description
Test for residual serial correlation (or individual random effects) locally robust vs. individual random effects (serial correlation) for panel models and joint test of serial correlation and the random effect specification by Baltagi and Li.
Usage
pbsytest(x, ...)
## S3 method for class 'formula'
pbsytest(
x,
data,
...,
test = c("ar", "re", "j"),
re.normal = if (test == "re") TRUE else NULL
)
## S3 method for class 'panelmodel'
pbsytest(
x,
test = c("ar", "re", "j"),
re.normal = if (test == "re") TRUE else NULL,
...
)
Arguments
x |
an object of class |
... |
further arguments. |
data |
a |
test |
a character string indicating which test to perform:
first–order serial correlation ( |
re.normal |
logical, only relevant for |
Details
These Lagrange multiplier tests are robust vs. local misspecification of the alternative hypothesis, i.e., they test the null of serially uncorrelated residuals against AR(1) residuals in a pooling model, allowing for local departures from the assumption of no random effects; or they test the null of no random effects allowing for local departures from the assumption of no serial correlation in residuals. They use only the residuals of the pooled OLS model and correct for local misspecification as outlined in Bera et al. (2001).
For test = "re"
, the default (re.normal = TRUE
) is to compute
a one-sided test which is expected to lead to a more powerful test
(asymptotically N(0,1) distributed). Setting re.normal = FALSE
gives
the two-sided test (asymptotically chi-squared(2) distributed). Argument
re.normal
is irrelevant for all other values of test
.
The joint test of serial correlation and the random effect
specification (test = "j"
) is due to
Baltagi and Li (1991) (also mentioned in
Baltagi and Li (1995), pp. 135–136) and is added
for convenience under this same function.
The unbalanced version of all tests are derived in Sosa-Escudero and Bera (2008). The functions implemented are suitable for balanced as well as unbalanced panel data sets.
A concise treatment of the statistics for only balanced panels is given in Baltagi (2013), p. 108.
Here is an overview of how the various values of the test
argument relate to the literature:
-
test = "ar"
:-
RS*_{\rho}
in Bera et al. (2001), p. 9 (balanced) -
LM*_{\rho}
in Baltagi (2013), p. 108 (balanced) -
RS*_{\lambda}
in Sosa-Escudero/Bera (2008), p. 73 (unbalanced)
-
-
test = "re", re.normal = TRUE
(default) (one-sided test, asymptotically N(0,1) distributed):-
RSO*_{\mu}
in Bera et al. (2001), p. 11 (balanced) -
RSO*_{\mu}
in Sosa-Escudero/Bera (2008), p. 75 (unbalanced)
-
-
test = "re", re.normal = FALSE
(two-sided test, asymptotically chi-squared(2) distributed):-
RS*_{\mu}
in Bera et al. (2001), p. 7 (balanced) -
LM*_{\mu}
in Baltagi (2013), p. 108 (balanced) -
RS*_{\mu}
in Sosa-Escudero/Bera (2008), p. 73 (unbalanced)
-
-
test = "j"
:-
RS_{\mu\rho}
in Bera et al. (2001), p. 10 (balanced) -
LM
in Baltagi/Li (2001), p. 279 (balanced) -
LM_{1}
in Baltagi and Li (1995), pp. 135–136 (balanced) -
LM1
in Baltagi (2013), p. 108 (balanced) -
RS_{\lambda\rho}
in Sosa-Escudero/Bera (2008), p. 74 (unbalanced)
-
Value
An object of class "htest"
.
Author(s)
Giovanni Millo (initial implementation) & Kevin Tappe (extension to unbalanced panels)
References
Bera AK, Sosa–Escudero W, Yoon M (2001). “Tests for the Error Component Model in the Presence of Local Misspecification.” Journal of Econometrics, 101, 1–23.
Baltagi BH (2013). Econometric Analysis of Panel Data, 5th edition. John Wiley and Sons ltd.
Baltagi B, Li Q (1991). “A Joint Test for Serial Correlation and Random Individual Effects.” Statistics and Probability Letters, 11, 277–280.
Baltagi B, Li Q (1995). “Testing AR(1) Against MA(1) Disturbances in an Error Component Model.” Journal of Econometrics, 68, 133–151.
Sosa-Escudero W, Bera AK (2008). “Tests for Unbalanced Error-Components Models under Local Misspecification.” The Stata Journal, 8(1), 68-78. doi:10.1177/1536867X0800800105, https://doi.org/10.1177/1536867X0800800105.
See Also
plmtest()
for individual and/or time random effects
tests based on a correctly specified model; pbltest()
,
pbgtest()
and pdwtest()
for serial correlation tests
in random effects models.
Examples
## Bera et. al (2001), p. 13, table 1 use
## a subset of the original Grunfeld
## data which contains three errors -> construct this subset:
data("Grunfeld", package = "plm")
Grunsubset <- rbind(Grunfeld[1:80, ], Grunfeld[141:160, ])
Grunsubset[Grunsubset$firm == 2 & Grunsubset$year %in% c(1940, 1952), ][["inv"]] <- c(261.6, 645.2)
Grunsubset[Grunsubset$firm == 2 & Grunsubset$year == 1946, ][["capital"]] <- 232.6
## default is AR testing (formula interface)
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"))
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"), test = "re")
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"),
test = "re", re.normal = FALSE)
pbsytest(inv ~ value + capital, data = Grunsubset, index = c("firm", "year"), test = "j")
## plm interface
mod <- plm(inv ~ value + capital, data = Grunsubset, model = "pooling")
pbsytest(mod)