REGRESSION FUNCTIONS
WEEK 5 (2nd Lecture)
1
Choice of Functional Form
o Slope and Elasticity
o Log-Lin Model
o Lin-Log Model
o Log Linear Model (Double Log Model)
o Linear Model
o Interpretation of Empirical Results
o Guidelines for choosing an appropriate
econometric model
2
3
Ln Wage = β0+β1Edu+β2Exp+β3Age+µ (Log Lin Model)
Coefficient Estimate Se t-value P-value
•
Intercept 1.50 0.50 3.00 0.0001
Education 0.33 0.05 6.60 0.0001
Experience 0.12 0.02 6.00 0.0003
Age 0.04 0.01 4.00 0.0001
4
Interpretation
• ln Wage = 1.5+ 0.33Edu+ 0.12Exp+ 0.04Age
• As the dependent variable is in logarithm form therefore it must
be greater than zero.
• Gauss Markov assumptions hold. Coefficients are statistically
and practically (economically) significant.
• A one unit increase in independent variable leads to 100*β
percentage change in dependent variable.
• If education is increased by one year ,we may expect monthly
wages to increase by 33%.
• If experience is increased by one year, we may expect monthly
wages to increase by 12%.
• If age is increased by one year, we may expect monthly wages to
increase by 4%. 5
Lin log model
Wage = β0+β1lnEdu+β2lnExp+β3lnAge+µ
•
Coefficient Estimate Se t-value P-value
Intercept 5000.0 987.65 5.06 0.0001
Education 4025 933.28 4.39 0.0001
Experience 5600 1552.20 3.61 0.0003
Age 2350 232.20 10.62 0.0001
6
Interpretation
• Wage = 5000+4025lnEdu+ 5600 lnExp+2350 ln Age
• As the independent variables are in logarithm form
therefore these must be greater than zero.
• Gauss Markov assumptions hold. Coefficients are statistically
and practically (economically) significant.
• One percent increase in independent variable leads to β
/100 unit change in dependent variable.
• If education is increased by 1%,we may expect that
monthly wages to increase by $ 40.25 (4025/100).
• If experience is increased by 1 % we may expect that
monthly wages to increase by $ 56 (5600/100).
• If age is increased by 1% ,we may expect that monthly
wages to increase by $23.25 (2325/100). 7
log linear Or double log model
ln PROD=β1+β2 ln AREA+β3ln LABOR+β4lnFERT+µi
8
Linear Model (Price in$, S & A in 000$)
Sales=β1+β2 PRICE+β3 ADVERT+µi
9
Results Interpretation
10
11
How Sales Changes?
12
Guidelines for Choosing Model
o Theory may suggest a particular functional form.
o Good practice to find out the rate of
change/elasticity of the regressand.
o Coefficients of the model should satisfy certain
prior expectations(Price negative…..).
o Sometimes more than one model may fit a given
set of data reasonably well. We may choose a
model with better R2 value but make sure that in
comparing R2 values ,dependent variables of the
two models should be the same, regressors can
take any form.
13
Guidelines…..
o One should not overemphasize the R2 measure in
the sense that the higher the R2 better the model.
Signs of the estimated coefficients, and their
statistical significance is more important than the
value of R2 . In this case model with a lower R2 may
be quite acceptable.
14
