9. Does the model suffer from misspecification?
Conduct the test at 5% from the STATA
output (5 points)
A misspecification exists when a model does not adequately capture the relationships between
the variables (Wooldridge, Wadud & Mand Lye, 2017). Misspecification can be related to
including irrelevant variables, excluding relevant variables, or the incorrect functional form
(Wooldridge, Wadud & Mand Lye, 2017). The Ramsey RESET Test is used to detect
misspecification. If the null hypothesis is rejected using the Ramsey RESET Test, this may
indicate a misspecification.
Unstricted model:
^ 3 wage−¿^
Wage=β 0−β 1 games+ β 2 points+ β 3 age+ β 4 black+ δ 1 wage−hatδ 2 wage−+δ
Restricted model (Restriction; δ 1=δ 2=δ 3=0)
Wage=β 0−β 1 games+ β 2 points+ β 3 age+ β 4 black
Hypothesis:
H0: δ 1=δ 2=δ 3=0 => No ommited Variables, correct funtional model
H1: δ 1 ≠ 0 => Have ommited Variables, wrong functional model
Alpha = 0.05
df1: Number of linear restrictions specified in H0 (q) = 3
Number of observations in the unrestricted model = 269
Number of independent variables in the unrestricted model = 7
df2 = 269−7−1=261
Critical F-stat; F ( 0.05 , 3 , 261 )=2.63
Model suffers from misspecification if we reject H0. Reject H0 if the calculated F-stat is greater
than the critical F-stat.
Calculated F-stat; F ( 3,261 ) =0.31
Conclusion:
Since, calculated F-stat 0.31 < critical F-stat 2.60, we do not reject H0 at the 5% significant level.
Thus, the model does not suffer from misspecification.
10. Does the model suffer from heteroskedasticity? There are two tests to conduct. Perform
each one at 5%. (10 points)
Heteroscedasticity exists when the error terms' variance is unequal for the values of the
independent variables (Wooldridge, Wadud & Mand Lye, 2017). Heteroscedasticity can lead to
biased coefficient estimates and also make the t-tests, F-tests, and confidence intervals invalid
(Wooldridge, Wadud & Mand Lye, 2017). The Breusch-Pagan Test and the White Test are
commonly used tests to detect heteroscedasticity. In both tests, if the null hypothesis, which
states that the error variance is constant, is rejected, heteroscedasticity exists.
Breusch-Pagan/Cook-Weisberg Test
Hypothesis:
H0: Var e=δ 2=> Homoscedastic
H1: Var e ≠ δ 2=> Hetoscedastic
Residual model: u^ 2=δ 0+δ 1 black +δ 2 games+δ 3 points+ δ 4 age+u
Alpha = 0.05
d.f = 4
Critical Chi-squared = 9.487729
Calculated Chi-squared; chi 2 ( 4 )=29.03
Conclusion:
Because calculated Chi-squared 29.03 is greater than Critical Chi-squared 9.487729, so we reject
the H0. Due to rejecting H0, the variance of the residduals is not constant across the independent
variables. This model suffers from hetereokedasticity.
White’ Test
Ho: Var e=δ 2 => Homoskedastic
H1: Var e ≠ δ 2=> Heteroskedastic
∪=δ 0+ δ 1 black+ δ 2 games+ δ 3 points +δ 4 age
Alpha = 0.05
df = 13
Critical Chi-squared = 22.3621
Calculated Chi-squared; chi 2 (13 )=23.42
Conclusion:
Because Calculated Chi-squared 23.42 is greater than Critical Chi-squared 22.3621, thus we
reject the H0. Due to rejecting H0, the variance of the residuals is not across the independent
variables. Hence the model suffers from heteroskedasticity.
11. Based on the answers from questions five to ten, are the tests in questions two & three
reliable? Explain why or why not. (7 points)
From questions 5 to 10 it was found that at the 5% significance level, the model does not suffer
from multicollinearity or misspecification; however, in terms of the Breusch Pagan/Cook-
Weisberg Test and the White’s Test it is suffering from heteroskedasticity. So, although these
tests show that the model properly specifies (No Misspecification) and shows relationships
between the variables (No Multicollinearity), the heteroskedasticity that is present does call into
question the reliability of the F-test found in question 3. Specifically, heteroskedasticity makes
the variance formulas for the ordinary least squares (OLS) estimators invalid and changes the
standard error (Wooldridge, Wadud & Mand Lye 2017). Therefore, the test becomes invalid and
thus unreliable in showing the significance of games, points, age and black in showing wage.
With regards to question 2 interpretations, it is complicating by the issue of heteroscedasticity
causing the OLS model to not be the best linear unbiased estimator. Also, if the fact that
heteroscedasticity caused a variable to not be significant any longer if any of this occurred, it
would invalidate one the 4 model estimates; reducing some reliability in question 2. That said, as
long as the assumptions in 5, 6 and 7 are not violated, question 2 is somewhat reliable, as
heteroscedasticity does not influence the bias, consistency or R-squared of the OLS model.
