TOTAL: /20

STUDENT

Name (capitals):
First Name (capitals):
Student number:
Programme:
TEA CHER

HBE15e Econometrics : Trial exam 1

Teacher(s): Professor Jef Hendrickx

Duration of the exam: 3 hours

Nos. of pages exam copy, including cover sheet: 15 pages Nos. of attachments: 0

1. Write your name, first name and student number legibly on each sheet.
2. Switch off your mobile phone/smartphone and put it, together with other electronic devices
(smartwatch, headphones...), and backpack out of reach.
EXAM INSTRUCTIONS

Follow the instructions of the invigilator.

3. Pencil cases are not allowed, writing materials may be placed in a transparent plastic bag clearly
visible on the bench/table.

4. The following aids are allowed:

(Graphical) calculator, formula sheet (provided on computer or non-annotated own copy)

5. Write the answer only in the answer space provided on the front of the exam sheet. The back of
the exam sheet can be used as scrap paper.
6. You may not use your own paper. If necessary and if allowed, ask the invigilator for extra scrap
paper.
7. You may not detach the exam bundle. Submit all sheets after the exam, including the separate
scrap paper sheets.
DOCENT
Question 1 (with R) (2.5 points)
This question uses the Boston HDMA dataset with data about 2,380 mortgage applications
(Boston HDMA.xlsx). The dependent variable (deny) is 1 if a mortgage is denied and 0
otherwise. The independent variables are piratio (P/I ratio, which is the ratio of the applicant’s
anticipated total monthly payments to his or her monthly income), and black (race, which is 1
if applicant is black and 0 if applicant is white).
a) Estimate a linear probability model with deny as dependent variable and piratio and black
as independent variables. Use heteroskedasticity-robust standard errors if necessary.
Report and interpret the coefficients for piratio and black.
It is sufficient to provide the following answers.
R code

Is it necessary to use heteroskedasticity-robust standard errors? Why (not)?

Report and interpret the coefficients for piratio and black.

b) Estimate the probability that a mortgage is denied for a black applicant with a P/I ratio of
0.3.

2
c) Test whether the effect of the P/I ratio on accepting the loan is different for blacks than
for whites. You might have to adjust the model in a). Report the R-code, the test statistic,
the p-value and a conclusion.

R code

Test statictic

P-value

Conclusion

Question 2 (with R) (5 points)

We use panel data of 716 American women who were interviewed in 1982, 1983, 1985, 1987 and
1988. The sample consists of women who were employed, and whose schooling was completed,
when interviewed. The data file is named nls_panel.xlsx and contains 3,580 lines of data. It contains
the following variables
- id : Identifier for panel individual
- year : Year interviewed (1982, 1983, 1985, 1987, 1988)
- lwage : Logarithm of the wage (=ln(wage))
- educ : Number of years of education (completed)
- exper : Total work experience, in years
- tenure : Job tenure (in current job), in years
- black : Ethnicity (1 = black, 0 = white)
- south : Region (1 = south, 0 = north)
- union : Union membership (1= yes, 0 = no)

3
a) Estimate a fixed effects model with lwage as dependent variable and educ, exper, tenure, black,
south and union as independent variables. Report the coefficient and standard error for union.

R code

𝛽̂𝑢𝑛𝑖𝑜𝑛

SE

b) Describe the effect of union membership on the wages in this fixed effects model.

c) Test if the case-specific constants in the fixed effects model are all equal. Report the test statistic
and a conclusion. Why is this test useful?

Test statistic

Conclusion

Why useful?

4
d) Estimate a random effects model with the same variables as the fixed effects model. Again, you
don’t have to use robust standard errors. So estimate the model (see formula sheet)
𝑌𝑖𝑡 = 𝜶𝒊 + 𝛽1 𝑋1,𝑖𝑡 + ⋯ + 𝛽𝑝 𝑋𝑝,𝑖𝑡 + 𝛾1 𝑍1,𝑖 + ⋯ + 𝛾𝑘 𝑍𝑘,𝑖 + 𝐸𝑖𝑡 with 𝐸𝑖𝑡 ~ 𝑁(0, 𝜎𝑒2 )

𝜶𝒊 = 𝛼 + 𝑈𝑖 with 𝑈𝑖 ~ 𝑁(0, 𝜎𝑢2 )

with Y lwage and X1, X2,… the independent variables.

Test the hypothesis H0 : 𝜎𝑢2 = 0 . Report the R code, the test statistic, a conclusion and why this test
is useful.

R code

Test statistic

Conclusion

Why useful?

e) Test whether the coefficients in the fixed effects model differ significantly from the coefficients in
the random effects model (for the variables that occur in both models). Report the R code, the test
statistic and the conclusion again. What is the use of this test and what can you finally conclude?

R code

Test statistic

5
 Conclusion of
the test

Why useful
and final
conclusion?

f) Re-estimate the model from question a), but now use cluster-robust standard errors. Do you come
to different conclusions and, if so, where? What potential problems are addressed by working with
cluster-robust standard errors (instead of the default standard errors)?

6
Question 3 (with R) (2.5 points)

This exercise uses data from the STAR experiment. In kindergarten all pupils start in a ‘regular’ class
with one teacher per class. The next year, in first grade, these pupils are redistributed. Some pupils go
to a class with a teacher and an additional teacher aide, while the other pupils remain in a ‘regular’
class with just one teacher. At the end of kindergarten and first grade, all pupils take a reading and
math test. Our outcome variable is the total score on this test. So for each pupil we have a result on
this test in kindergarten and in first grade. We want to find out whether the use of an additional teacher
aide leads to better results for the pupils on average.
The data for this excercise are available both in wide format (star (wide).xlsx) as in long format (star
(long).xlsx).
For the data in broad format, totalscore0 (resp. totalscore1) gives the result on the math and reading
test in kindergarten (resp. first grade) for each pupil.
For the data in long format, the variable totalscoreit gives the score of pupil i on the math and reading
test in year (or grade) t (0 = kindergarten, 1 = first grade).
In both data sets, the variabele aide1 equals 1 if the first grade pupil is in a class with an additional
teacher aide, and 0 otherwise. Remember that in kindergarten, all pupils are in a ‘regular’ class with
just one teacher.

Use an appropriate model to determine whether the use of an additional teacher aide leads to better
results on average.

Your answer must contain the following elements:

• Clearly indicate which dataset you used.
• Give the (theoretical) model equation.
• Give the R commands you used to estimate your model. Use (cluster) robust standard errors.
• Make a table with the estimated parameters and standard errors.
• Interpret the estimated parameters in your model.
• Finally, give a well-founded answer to the research question.
(1.5 page answer space)

7
8
Question 4 (3 points)

For this question we use a dataset from 2002 with the wages of 1377 Belgian employees. This contains
the following variables:

• W (wage) : wage (in euro)

• A (age) : age in years (for your information : median age is 41)
• A41 (age41) = A – 41 (centered age)
• diploma : diploma in 3 categories (1 = low-skilled, 2 = medium-skilled, 3 = high-skilled)
• M (medium) : dummy for medium-skilled
• H (high) : dummy for high-skilled
We estimate the following regression models

Model 1 : Wi =  + 1 A41i + 2 Mi + 3 Hi + Ei

Model 2 : Wi =  + 1 A41i + 2 Mi + 3 Hi + 4 A41i × Mi + 5 A41i × Hi + Ei

Both models are estimated in R, using the default standard errors. This gives the following output

Model 1

Model 2

9
a) Give an interpretation of all 6 estimated coefficients in model 2.
b) What’s the average difference in wage between a high-skilled and a low-skilled 30-year
old employee (using model 2)? If you would like to test if this difference is significant,
what would be the null hypothesis (in symbols)?
c) Does the effect of age on wages depends on the diploma? Perform an appropriate test.
Give the null- and alternative hypothesis (in symbols), calculate the test statistic and the
p-value and give a conclusion. You may assume homoscedasticity for this part.
d) Explain in detail how the White test works for model 2.
• What is it used for?
• Which auxiliary regression has to be estimated (write down the full theoretical
equation applied to this case)?
• Suppose this auxiliary regression has an R² of 0.08737. Calculate the p-value for
the test and give a conclusion.
• What are the consequences of this conclusion regarding the above output of
model 2?
(1.5 page answer space)

10
11
Question 5 (3 points)

Assume the following random effects model

𝑌𝑖𝑡 = 𝜶𝒊 + 𝛽1 𝑋𝑖𝑡 + 𝛽2 𝑍𝑖 + 𝐸𝑖𝑡 with 𝐸𝑖𝑡 ~ 𝑁(0, 𝜎𝑒2 )

𝜶𝒊 = 𝛼 + 𝑈𝑖 with 𝑈𝑖 ~ 𝑁(0, 𝜎𝑢2 )

and with Eit mutually uncorrelated, and Uj mutually uncorrelated and uncorrelated with Eit.

So the random effects model can also be written as

𝑌𝑖𝑡 = 𝛼 + 𝛽1 𝑋𝑖𝑡 + 𝛽2 𝑍𝑖 + 𝐸𝑖𝑡 + 𝑈𝑖 = 𝛼 + 𝛽1 𝑋𝑖𝑡 + 𝛽2 𝑍𝑖 + 𝑉𝑖𝑡

with Vit = Eit + Ui .

Show that the intra-class correlation  is given by

𝜎𝑢2
𝜌 = 𝐶𝑜𝑟𝑟(𝑉𝑖𝑡 , 𝑉𝑖𝑠 ) =
𝜎𝑢2 + 𝜎𝑒2

(1.5 page answer space)

12
13
Question 6 (2 points)

Research has shown that offering free nuts with a drink has an influence on the consumption
behavior of pub visitors. An owner of two pubs (one in Antwerp (A) and one in Brussels (B)) is
trying to see whether this can indeed influence consumption behaviour. This is done as follows
in 2 periods of 4 weeks.

In the first 4-week period, he doesn’t offer free nuts. He records the average weekly drink bill
over that period for a random sample of regular customers, both in Brussels and in Antwerp.

In the second period of 4 weeks, he will serve free nuts to the regulars of the Antwerp pub. The
regulars in the Brussels pub do not receive free nuts. He again records the average weekly drink
bill over this period for a random sample of regulars in Antwerp and Brussels.

He managed to collect the results in the table below.

Average weekly
Obs Group Period bill (€)
1 A 1 20.40
2 A 1 20.50
3 A 1 19.30
4 A 1 19.20
5 A 1 20.80
6 A 1 22.00
7 A 2 21.50
8 A 2 23.50
9 A 2 24.10
10 A 2 23.60
11 A 2 22.50
12 A 2 24.00
13 B 1 20.00
14 B 1 20.80
15 B 1 20.50
16 B 2 21.50
17 B 2 21.20
18 B 2 19.90
19 B 2 20.00
20 B 2 21.50

Make an estimate as accurately as possible of the effect of handing out free nuts on the
average weekly drink bill. Describe and motivate your approach. What is an important
assumption in the method used? (answer space on next page)
14
15