Artificial Regression Test
Estimates a regression-based Hausman test for fixed effects individual slope models.
feistest(
model = NA,
robust = FALSE,
type = c("all", "art1", "art2", "art3"),
terms = NULL,
...
)model |
an object of class " |
robust |
logical. If |
type |
one of " |
terms |
An optional character vector specifying which coefficients should be jointly tested.
By default, all covariates are included in the Wchi-squared test. For " |
... |
further arguments. |
The Hausman test can be computed by estimating a correlated random effects model
(see Wooldridge 2010, pp. 328-334, Ruettenauer and Ludwig 2020). This is achieved by
estimating a Mundlak (Mundlak 1978) specification using random effects models
with plm.
Subsequently, feistest tests whether the time-constant variables / slope variables are correlated with
the unobserved heterogeneity by using a Wald chi-squared test with wald.test.
type="art1" estimates an extended regression-based Hausman test comparing fixed effects
individual slope models and conventional fixed effects models. For art1 the
Mundlak-specification (Mundlak 1978) includes the person-specific averages,
but additionally the person-specific slope estimates used for "detrending" in feis.
This allows to test whether we can omit the estimated values based on the slopes and reduce the model
to a conventional FE model. The Wald test of type="art1" is applied to the slope variables only.
type="art2" estimates the conventional regression-based Hausman test
(as described in Wooldridge 2010, pp. 328-334) comparing conventional
fixed effects models against random effects models.
type="art3" estimates a regression-based Hausman test comparing FEIS directly against RE,
thereby testing for inconsistency of the RE model due to either heterogeneous slopes or time-constant
omitted heterogeneity. For art3 the Mundlak-specification includes only the person-specific
slopes, and no averages. This allows to test whether we can omit the estimated values based on
the slopes and reduce the model to a conventional RE model.
(for a formal description please see Ruettenauer and Ludwig 2020).
Currently, the tol option in feis() is only forwarded in bsfeistest,
but not in feistest.
If specified (robust=TRUE), feistest uses panel-robust standard errors.
An object of class "feistest", containing the following elements:
wald_feis |
an object of class " |
wald_fe |
an object of class " |
wald_re |
an object of class " |
vcov1 |
the variance-covariance matrix of CREIS ( |
vcov2 |
the variance-covariance matrix of CRE ( |
vcov3 |
the variance-covariance matrix of CREIS without the means ( |
CREIS |
an object of class " |
CRE |
an object of class " |
CREIS2 |
an object of class " |
call |
the matched call. |
robust |
logical. If |
formula |
an object of class " |
type |
the type of performed test(s). |
terms |
character vector of covariates are included in the Wchi-squared test. |
Mundlak Y (1978).
“On the Pooling of Time Series and Cross Section Data.”
Econometrica, 46(1), 69.
ISSN 00129682.
Ruettenauer T, Ludwig V (2020).
“Fixed Effects Individual Slopes: Accounting and Testing for Heterogeneous Effects in Panel Data or Other Multilevel Models.”
Sociological Methods and Research, OnlineFirst.
ISSN 0049-1241, doi: 10.1177/0049124120926211, https://doi.org/10.1177/0049124120926211.
Wooldridge JM (2010).
Econometric Analysis of Cross Section and Panel Data.
MIT Press, Cambridge, Mass.
ISBN 0262294354.
data("mwp", package = "feisr")
feis.mod <- feis(lnw ~ marry + enrol | year,
data = mwp, id = "id", robust = TRUE)
ht <- feistest(feis.mod, robust = TRUE, type = "all")
summary(ht)
# Only for marry coefficient
ht2 <- feistest(feis.mod, robust = TRUE, type = "all", terms = c("marry"))
summary(ht2)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.