ECON7IE Topic 2 Class Exercises
ECON7IE Block 1 Topic 2: Binary dependent variables
Class exercises on Part 1 of lecture: Estimation
For linear models, we interpret both the coefficients’ signs and their sizes.
For nonlinear models, we focus more on the signs of the coefficients than on their sizes.
1. Determinants of married women’s labour force participation: coefficient interpretation
Consider a dataset containing information from a survey of married women in the USA in 1975:
lfp Dummy variable = 1 if woman participated in labour force, 0 otherwise
ku6 No. of kids under 6 years old
k618 No. of kids 6-18 years old
wa Age
we Years of education
ha Husband's age
he Husband's years of education
hw Husband's wage ($/hour)
cit Dummy variable = 1 if live in large city, 0 otherwise
A model for labour force participation (i.e. dependent var is lfp) is estimated using three methods: a
linear probability model (LPM), a logit model, and a probit model, with the following results:
Variable LPM Logit Probit
ku6 -0.307*** -1.506*** -0.908***
k618 -0.017 -0.092 -0.054
wa -0.009 -0.042 -0.026
we 0.057*** 0.275*** 0.167***
ha -0.005 -0.026 -0.015
he -0.012 -0.056 -0.035
hw -0.012** -0.059** -0.036**
cit 0.006 0.043 0.034
_cons 0.796*** 1.488 0.900
N 753 753 753
R-squared 0.140
Pseudo R-squared 0.112 0.112
Legend: * p<0.05; ** p<0.01; *** p<0.001
1.1 In the LPM:
a) Interpret the coefficient on the woman’s education (we) [2]
b) Interpret the coefficient on the number of kids younger than 6 (ku6) [2]
c) Do they have the expected signs? [3]
1.2 In the logit model:
a) What happens to the probability that a married woman participates in the labour force if she
lives in a large city (cit)? [2]
𝛽
b) Use the expression 100(𝑒 − 1) to interpret the size of the coefficient on we in terms of odds.
[3]
1.3 In the probit model:
a) Are the signs and significance of the coefficients the same as the logit model? i.e. would any
of your substantive conclusions differ depending on which model you chose to use? [3]
b) Are the magnitudes of the coefficients the same as the logit model? Why or why not? [3]
Page 1 of 2
ECON7IE Topic 2 Class Exercises
Class exercises on Part 2 of lecture: Marginal effects and testing
2. Determinants of married women’s labour force participation: marginal effects and testing
Using the same dataset, the following restricted logit model is estimated:
. logit lfp ku6 we hw
Iteration 0: log likelihood = -514.8732
Iteration 1: log likelihood = -472.23083
Iteration 2: log likelihood = -472.20068
Iteration 3: log likelihood = -472.20068
Logistic regression Number of obs = 753
LR chi2(3) = 85.35
Prob > chi2 = 0.0000
Log likelihood = -472.20068 Pseudo R2 = 0.0829
------------------------------------------------------------------------------
lfp | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ku6 | -1.030593 .1638489 -6.29 0.000 -1.35173 -.7094546
we | .25358 .0391446 6.48 0.000 .176858 .3303021
hw | -.0731387 .0203473 -3.59 0.000 -.1130187 -.0332588
_cons | -2.030206 .4492277 -4.52 0.000 -2.910676 -1.149736
------------------------------------------------------------------------------
2.1 Use the expression below to calculate the marginal effect of a woman obtaining an extra year of
education, by following the steps provided:
𝑒 (𝑏1+𝑏2𝑋2+⋯ )
𝑃𝑟𝑜𝑏(𝑙𝑓𝑝 = 1) =
1 + 𝑒 (𝑏1+𝑏2𝑋2+⋯ )
a) Calculate the probability of a woman participating if she has 1 child younger than 6 years of
age, 12 years of education, and her husband earns $20 per hour. [3]
b) Calculate the probability of a woman participating if she has 1 child younger than 6 years of
age, 13 years of education, and her husband earns $20 per hour. [3]
c) Calculate and interpret the marginal effect. [2]
2.2 The model shown above is a restricted model. The full logit model (on page 1) has a log likelihood
value of -457.44.
a) Write out the null and alternative hypotheses that will be used to conduct a likelihood ratio
test to compare these two models. [2]
b) Calculate the likelihood ratio test statistic. [2]
c) Compare the test statistic to a critical value from the chi-squared table below, and conclude.
[3]
Page 2 of 2
