This exam contains 4 questions on 9 pages (including this title page). You have to
answer 3 out of these 4 questions during 120 minutes. If you answer all questions,
only the first 3 questions will be taken into account.
The maximum number of points obtainable is 60. Thus each question is worth 20
points.
You may use a calculator, a dictionary. Use or possession of unauthorized materials
will automatically result in the award of 0 points for this examination.
Do your best to write legibly. Solutions which cannot be read with reasonable effort
will not be considered.
Good luck!
i N(0, 1),
Yi =
i = 1, . . . , n,
1 if Yi 0,
0 otherwise.
Description
= 1 if had at least one affair, 0 otherwise
= 1 if male, 0 otherwise
age in years
years married
= 1 if have kids, 0 otherwise
5 = very religious, 4 = somewhat, 3 = slightly, 2 = not at all, 1= anti
5 = very happy, 4 = happier than average, 3 = average,
2 = somewhat unhappy, 1 = very unhappy
3P
b) Derive the likelihood function, the log-likelihood function and the score function.
4P
Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2
=
=
=
=
601
64.53
0.0000
0.0955
-----------------------------------------------------------------------------AFFAIR |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------MALE |
.2168516
.1208316
1.79
0.073
-.019974
.4536773
AGE | -.0239581
.0103599
-2.31
0.021
-.0442631
-.0036531
YRSMARR |
.0547532
.0187859
2.91
0.004
.0179336
.0915728
KIDS |
.2060654
.1630705
1.26
0.206
-.1135468
.5256777
RELIG | -.1880535
.0514747
-3.65
0.000
-.2889421
-.0871649
RATEMARR | -.2678015
.0526933
-5.08
0.000
-.3710786
-.1645245
_cons |
.9744369
.3624455
2.69
0.007
.2640567
1.684817
------------------------------------------------------------------------------
To what do you have to pay attention when you interpret the slope parameters?
What can you say about marginal effects just looking at the coefficients? What
distinguishes the marginal effects in the Probit model from the marginal effects
in the Linear Probability Model?
2P
d) Find the probability that a 30 year old male with no kids who is slightly
religious and very happily married for ten years has an affair.
3P
e) Find the probability of having an affair for the same man if he was very unhappily married. How did the probability change with respect to the previous
part?
2P
f) You want to test whether the variables MALE, AGE and KIDS are jointly
significant and estimate a model without these variables. The results are displayed below.
Probit regression
Number of obs
LR chi2(3)
Prob > chi2
Pseudo R2
=
=
=
=
601
55.48
0.0000
0.0821
-----------------------------------------------------------------------------AFFAIR |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------YRSMARR |
.0324023
.0111055
2.92
0.004
.010636
.0541686
RELIG | -.1887376
.0513024
-3.68
0.000
-.2892884
-.0881868
RATEMARR | -.2677542
.0520324
-5.15
0.000
-.3697358
-.1657727
_cons |
.6436336
.2708041
2.38
0.017
.1128673
1.1744
------------------------------------------------------------------------------
3P
g) What are the problems if you simply estimate this model using OLS?
3P
Pr [Yi = yi ] =
exp(i )yi i
, yi = 0, 1, 2, . . . ,
yi !
where E [ Yi | Di ] = V [ Yi | Di ] = i = exp(0 + 1 Di ).
(
1, if individual i live in a city ,
Di =
0, otherwise,
You have dataset with 100 women and 40 percent are living in a city. On average,
women living in a city have 2.5 children, whereas women not living in city have 3
children.
a) Write-down the log-likelihood function of the model and derive the score and
hessian matrix.
3P
5P
c) Write down the restricted log-likelihood function under the null hypothesis
that living in a city does not affect the number of children. Calculate the ML
estimates under this hypothesis.
4P
d) Calculate the score of the unconstrained model at the values of the constrained
estimates.
4P
4P
Description
nonwife income
years of education
years of experience
(years of experience)2
age
=1 if woman has no small kids, 0 otherwise (base category)
=1 if woman has 1 small kid, 0 otherwise
=1 if woman has 2 or more small kids, 0 otherwise
Number of obs
LR chi2(7)
Prob > chi2
Pseudo R2
=
=
=
=
753
271.11
0.0000
0.0343
-----------------------------------------------------------------------------hours |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------nwifeinc | -9.168673
4.444128
-2.06
0.039
-17.89316
-.4441876
educ |
82.17018
21.49694
3.82
0.000
39.96848
124.3719
exper |
133.2854
17.31943
7.70
0.000
99.28476
167.286
expersq | -1.898172
.5402329
-3.51
0.000
-2.95873
-.8376143
age | -53.12515
7.07357
-7.51
0.000
-67.01162
-39.23868
kids1 |
-852.89
144.4529
-5.90
0.000
-1136.473
-569.3074
kids2 | -1882.149
301.9438
-6.23
0.000
-2474.909
-1289.388
_cons |
859.62
405.4201
2.12
0.034
63.71987
1655.52
-------------+---------------------------------------------------------------/sigma |
1122.785
41.58597
1041.146
1204.425
-----------------------------------------------------------------------------Obs. summary:
325 left-censored observations at hours<=0
428
uncensored observations
a) Write down the formal model on which these estimates are based. Use matrix
notation an in the lecture, i.e. define Xi and the coefficient vector explicitly,
state the dimensions, for what they stand etc. Which distribution is assumed
for the error term in this model?
4P
b) Derive the mean of annual hours of work for women who work, i.e. E [ Yi | Yi > 0, Xi ].
4P
c) What would be the effect of a having one small kid (with respect to not having
any small kids) on the expected annual hours of work of a working woman
with nwifeinc = 20, educ=12, exper=10, age=30?
4P
d) What would be the effect of a one small kid (with respect to not having any
small kids) on the expected annual hours of work of a woman with nwifeinc =
20, educ=12, exper=10, age=30?
4P
e) Why do you think running an ordinary least squares regression is not appropriate?
4P
h=0 exp(Zh )
j = 0, . . . , m.
4P
b) Show that the cross effect, the effect of a change in attribute of alternative
m 6= j on the probability of choosing j, has the opposite sign of .
4P
c) Using the previous results show that the following equality hold:
m
X
Pr [Y = j |Z ]
=0
Z
l
l=1
4P
4P
e) Compute the direct effect of a marginal change in general costs on the choice
probabilities for train.
2P
f) Compute the cross effect of a marginal change in the general costs for cars on
the probability of taking the train at sample mean.
2P
Formulary
pdf of a normal distribution
If X N[, 2 ], then the pdf of X is given by:
2 !
1
1 a
f (a) = exp
2
if truncation is X > a,
if truncation is X < a
and
() = ()[() ].
An important result is
0 < () < 1
= /(1 ) and = 2 .
ML testing
Let (R ) and (U ) denote the restricted ML estimates (under H0 ) and unrestricted
ML estimates, respectively.
LR test
LM test
LM =
a
ln L(R )
R )]1 ln L(R )
2(m) ,
[H(