The McGraw-Hill Series Economics essentials of economics brue, McConnell, and Flynn Essentials of Economics

Anderson–Darling normality test,
known as the 
A
2
statistic.
The underlying null hypothesis is that the variable under consideration is
normally distributed. As Figure 5.8 shows, for our example, the computed 
A
2
statistic is
0.289. The 
p value
of obtaining such a value of
A
2
is 0.558, which is reasonably high.
Therefore, we do not reject the hypothesis that the residuals from our illustrative example
are normally distributed. Incidentally, Figure 5.8 shows the parameters of the (normal)
distribution, the mean is approximately 0, and the standard deviation is about 0.8987.
Jarque–Bera (JB) Test of Normality
20
The JB test of normality is an 
asymptotic,
or large-sample, test. It is also based on the OLS
residuals. This test first computes the 
skewness
and 
kurtosis
(discussed in 
Appendix A
)
measures of the OLS residuals and uses the following test statistic:
JB
=
n
S
2
6
+
(
K
−
3)
2
24
(5.12.1)
4
3
Frequency
Residual
Histogram
(Response is mean hourly wage)
2
1
0
–
1.5
–
1.0
–
0.5
0
1.0
0.5
1.5
FIGURE 5.7
Histogram of residuals
for wages—education
data.
20
See C. M. Jarque and A. K. Bera, “A Test for Normality of Observations and Regression Residuals,”
International Statistical Review,
vol. 55, 1987, pp. 163–172.
guj75772_ch05.qxd 07/08/2008 12:46 PM Page 131

---

132
Part One
Single-Equation Regression Models
where 
n
=
sample size, 
S
=
skewness coefficient, and 
K
=
kurtosis coefficient. For a nor-
mally distributed variable, 
S
=
0 and 
K
=
3. Therefore, the JB test of normality is a test of
the joint hypothesis that 
S
and 
K
are 0 and 3, respectively. In that case the value of the JB
statistic is expected to be 0.
Under the null hypothesis that the residuals are normally distributed, Jarque and
Bera showed that 
asymptotically (i.e., in large samples) the JB statistic given in Equa-
tion (5.12.1) follows the chi-square distribution with 2 df.
If the computed 
p
value of the
JB statistic in an application is sufficiently low, which will happen if the value of the statis-
tic is very different from 0, one can reject the hypothesis that the residuals are normally
distributed. But if the 
p 
value is reasonably high, which will happen if the value of the
statistic is close to zero, we do not reject the normality assumption.
For our example, the estimated JB statistic for our wages-education example is 0.8286.
The null hypothesis that the residuals in the present example are normally distributed can-
not be rejected, for the 
p
value of obtaining a JB statistic as much as 0.8286 or greater is
about 0.66 or 66 percent. This probability is quite high. Note that although our regression
has 13 observations, these observations were obtained from a sample of 528 observations,
which seems reasonably high.