Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 7

Applied Econometrics: Compulsory Assignment

Lars Tonn

1 Multiple Choice

1.B

2.D

3.A

4.B

5.B

6.D

7.C

2 Cross-Sectional Analysis

8.

9. R Code:

# scale vector
ability_scaled <- c(scale(ability))

res <- cbind(labor,ability_scaled)


summary(ability_scaled)
res <- subset(res, select = -c(5))

ability−E [ ability ]
abilit y scaled =
sd ¿¿

10. Construct the variable age in your dataset. Considering an OLS regression, study the determinants
of the logarithm of hourly wage. Consider the same analysis for individuals older than 20. Provide
economic explanations of your results (maximum 3 sentences). Can you run a regression containing
age, potexper and educ? If not, explain why.

ln(Wage) = β0 + β1educ + β2potexper + β3potexper^2

The quadratic regression of log wage as a function of education and potential experience explains 11.7%
of the variation in log wages and reinforces the notion that education is the cornerstone of wage
growth, with an additional year of education predicting a 7.6% increase in wages. Potential experience,
on the other hand, has a concave relationship with logwage which suggests a diminishing
return/investment. Lastly, the variables ability, fathered, brknhome and siblings are not statistically
significant in explaining variation in wage (Appendix II).

b) Multicollinearity

If we were to include the variable age and combine it with education and potexper we would run into an
issue of multicollinearity due to the higher correlation with the variables potexper and age. Potexper on
the other hand is not highly correlated with education and can therefore be included in our model.

Do you think that the OLS regression is the best econometric strategy? Explain why (maximum 3
sentences).

Under the proposed model all the assumptions for OLS are met therefore the estimate is BLUE which
implies the best linear estimator. However, given the hypothesis that wages tend to disperse with
potential experience (implying heteroskedasticity) one of the main tenets of OLS is violated; namely the
assumption that the residual errors are homoscedastic. The Breusch-Pagan test rejected the null
hypothesis that the data is homoskedastic, that said, a weighted least squares approach could be
implemented which better accounts for heteroskedasticity and therefore delivers a more efficient result.

Appendix
I. Quadratic Model Fit
II. Linear model Regression Summaries
III. Quadratic Regression Summaries
IV. Residual Plot for Linear Regression

You might also like