Page 263 - Contributed Paper Session (CPS) - Volume 2
P. 263
CPS1853 M. Irsyad Ilham
In addition to informal testing, there are three formal testing procedures
that can be used, namely Chow Test to choose between common effects
models and fixed effects models; Hausman test used to select between fixed
effects model and random effects model; as well as the Breusch-Pagan
Lagrange Multiplier (BP-LM) test to choose between common effects models
and random effects models.
The test used to check whether FEM better that CEM. This test obtaining
residual sum square to calculate. The null hypothesis is common effect model
is better than fixed effect model. The pattern of chow test is
( − )/( − 1)
=
()/( − − )
RRSS is residual sum of squares from common effect model. URSS is residual
sum of squares from fixed effect model. Whether n is number of observation,
k is number of independent variable, and T is period of time. If F bigger than
(;−1,−−) , it conclude to reject null hypothesis, which means that fixed
effect model better than common effect model.
In order to test random effect model better than fixed effect model,
Hausman test should be used. The null hypothesis of the test state that no
correlation between individual error and independent variable. In other words,
the null hypothesis says that random effect model better than fixed effect
model. The formula is
2
̂
̂ ′
̂
= [ − ] −1 [ − ]~
Where is covariance matrix of − estimation, shows random effect
̂
̂
̂
model regression coefficient vector. The vector indicate the array of fixed
effect model regression coefficient. The letter n and k mean number of
observation and number of independent variable. If the value of W bigger than
2 , it conclude to reject null hypothesis so fixed effect model better than
(;)
random effect model.
To know whether random effect model better than common effect model,
it uses Breuch-Pagan Lagrange Multiplier (LM) test, which developed by
Breuch-Pagan. The test base on the value of residual from common effect
model. The null hypothesis is intercept is not random variable or common
effect model is better than random effect model. The formula of the test is
∑ (∑ ̂) 2
2
= [ =1 =1 − 1] ~
1
2( − 1) ∑ ∑ ̂ 2
=1
=1
Where n and T are number of observation and total period of time. The symbol
̂ is residual from common effect model. If the value of LM bigger than
2 , it conclude to reject null hypothesis which means that random effect
(; 1)
model better than common effect model.
Building the appropriate panel data regression model from econometric
criteria, are needed a test and deal with several problems according to the
model assumption. If the selected model is fixed effect model or common
effect model, the assumption which should be full-filled are normality,
252 | I S I W S C 2 0 1 9