Introduction of Econometric

The data is found in the file wages2014 sample.dta and the exercise is based on work you should
have completed in lab 7. The questions in these labs give a good guide to the type of material
that should be included in your report.
a) Is there sex discrimination in earnings?
Explore the characteristics of the data. Estimate the regression of hourly pay on education and job
experience, then include a dummy for gender. Review the regressions to answer this question.
b) Does having a degree increase your earnings?
Include dummy variables to explore the salary benefit of different types of qualifications rather than
just education time. Comment on the earnings of different educational qualifications and gender.
c) Is the relationship linear?
Run the regression with hourly pay on experience, experience squared, education and the male
dummy. Review the revised model and comment on the non-linear relationship between wages and
experience.
Lab_7_dummies 14.docx 1/2
ECO 20042 – Introduction to Econometrics: Lab 7
Multiple Regression and Dummy Variables

This week we will look at some examples of how categorical variables affect multiple regression,
estimation; interpretation and hypothesis testing. We will use the data for your assessed exercise,
wages2014-sample which contains information from the Labour Force Survey (2000), among other things,
on the hourly pay of a sample of 1000 British employees. Load your data from KLE. Remember to start a
log file. You might find it useful to refer to your answers for lab 6 and lecture 7 to help with this exercise.
Write your answers in a word document as preparation for your report.
Check the definitions of the variables included in the data set, nore especially that educ is the number of
years of education above the minimum school leaving age (NOT the total years of education)

READ ALSO : contemporary Korean society

1. Is there sex discrimination in earnings?
Firstly, summarise hourly pay (hourpay) and hours for the whole sample and for men and women separately
to see if there is any indication of wage discrimination, hint add if gender==1 for men. This can
alternatively be done using the tab function that we used in Lab 1.
What are the average earnings and hours for all individuals?

What are the average earnings for men? And women? Comment

Plot the data – male and female wages with different markers on the same diagram. This is easily done by
creating a new variable gen wage_F=hourpay if gender ==2 and similarly for men. Then plot the two
variables on the same graph.
Run the same regression from Q2 last week i.e. regression of hourly pay on education and job experience
(exp) for all observations.

What is the estimated equation?

Interpret the estimated coefficients. (how much extra do you earn for an extra year of education, experience;
are their relative magnitudes as you would expect).

There is always much debate over the difference in earnings between men and women. One way way to look
at the difference is to include the dummy variable for men in the regression equation.
gen male=gender==1
Run the regression with hourly pay on experience, education and the male dummy. Comment on the
regression.
Estimated equation for men?
Estimated equation for women?
Interpretation?
Significance of coefficients?
Has the overall fit improved?
Re-estimate the model but allow there to be a difference in the reward to education and experience for men
and women (Hint: generate 2 interactive variables to allow slopes for education and experience to be
different between men and women, e.g. gen MLEXP=MALE*EXP etc)
Estimated equation for men?
Estimated equation for women?
Interpretation?
Significance of coefficients?
Has the overall fit improved?
Which model do you think is the best and why (think of all the tests you can do)?
Can you delete any of the variables to improve the model?

READ ALSO : Natural Selection

2. Does having a degree increase your earnings?
Create a table showing the earnings summarised by gender and edstatus (Hint: use the statistics menu,
……choose one/twoway table of summary statistics.
Comment on the earnings of different educational status and gender.

Create dummies to represent the eduation status in the following way:
gen Degree=edstatus==1
gen Alevel=edstatus==2
gen Olevel=edastatus=3
gen OtherQ=edstatus==4
Each individual will only have one of these variables equal to unity, or maybe none. You can check by
‘browsing’ the data.
Run the regression with hourly pay on experience, education, male dummy and the additional dummies.
Note that ‘no qualifications’ is the excluded caterory so the intercept value is for a female with no
qualifications.
Interpret the coefficients and check for significance and improvement in the model from Q1. Note
whether the magnitude of the coefficents o the edstatusdummies have the expected relative
size/order.
Explore to see if there are any significant interactive terms between education status and gender or
experience.
Is this model better than previous ones?

3) Is the relationship linear?
a) Generate a new variable experience squared (hint: gen exp2=exp^2). Run the regression with hourly pay
on experience, experience squared, education and the male dummy.
Review the revised model and comment on the non-linear relationship between wages and experience.
b) Often, the relationship may have another form of non-linearity. Generate a new variable – the natural log
of wages (i.e. gen lnwage=ln(hourpay) ). Regress lnwage on experience, experience squared, education
and the male dummy.
It would be a good idea to graph the observed and predicted wages from both models (remember to put lnw
prediction back to levels). This may require you to experiment with the symbols on the graph be much
smaller than the default) because otherwise you will not be able to see a series.
Review the revised model and compare with previous models.