Econometrics: Linear Regression
with one Regressor 2
Keisuke Kawata
Hiroshima University
䠄 Review 䠅 Least Squares Assumptions
The least squires assumption 1. Your data is
2. The mean of is zero:
3. The conditional mean of u does not depend on : For any t,t’,
• If the following least squares assumptions hold, OLS estimators � , � are –
– have
un iased a d onsistent esti ators.
the or al distri utio s u der the large sa ple size.
pure ra do sa pli g data.
� � = .
� � � = � � �′ .
Random sampling and � = assumptions
• The pure random sampling assumption is totally same requirement as the estimation for population mean.
• � � = is not
If � � =a, we should modify the population model as
�� = �′ + � � + �′ where �′ =� + �, ′ = − �.
Under the above specification, � �′ = must hold. The constant term can be interpreted as
Same conditional mean assumption
• In many case, � � � = � � �′ is e.g.)
• The estimation about the effect of education year on income.
⇒Cognitive/non- og itive a ility a d pare t’s hara teristi s ay e differe e.
• The estimation about the effe t of pare t’s income on hild’s education outcomes.
⇒The parents location may be different.
Potential outcome and Same conditional mean
• � � � = � � �′ is equivalent to � �� |�� = = � �� |�� = Proof
• Because � = �� − � �� ⇒� � � = � �� �� = � − � ��
• � � � = = � � � = ⇒� �� �� = = � �� �� =
Graphical Intuition
• � � = � � case ⇒
�
��
� + �
�Graphical Intuition
• � � � > � � ′� if � > �′ ⇒
�
�� + �
�Graphical Intuition
• � � � < � � ′� if � > �′ ⇒
�
��
� + �
�Variance of OLS estimator
• Under Least Squares Assumptions, the estimators � , � have
• The distribution of � is � � , � �1 where
� �1 = � �� ��� − �� �
�
where �� is mean of �, and �� � is the sample variance.
• The variance of estimator is small if –
Intuition
• Roughly speaking, we get the OLS estimator using the variation of explanation variable.
⇒ If there are no variation,
⇒ If the variation is not so large, the accuracy of OLS estimator is
Conclusion
• To get the estimator of linear population model, we often use the OLS estimation.
• The estimator of OLS is unbiased and consistent estimator if least squires assumptions hold.
• We can get accurate estimators if the sample size and the sample variance of explanation variable are large.