Scribd
Upload
45 views

Propensity Score Matching Example

The document discusses using propensity score matching to estimate the effect of a training program on earnings. It finds that after matching treated and control individuals, the training program lowered earnings by $6,000-$13,000 when using one period of data but increased earnings by $2,000-$3,000 when using the difference between two periods of data.

Uploaded by

dinesh dhanasekaran
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
45 views

Propensity Score Matching Example

The document discusses using propensity score matching to estimate the effect of a training program on earnings. It finds that after matching treated and control individuals, the training program lowered earnings by $6,000-$13,000 when using one period of data but increased earnings by $2,000-$3,000 when using the difference between two periods of data.

Uploaded by

dinesh dhanasekaran
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Propensity Score Matching Example

Ani Katchova

© 2013 by Ani Katchova. All rights reserved.


Propensity Score Matching Example

 We want to study the effect of a training program on individuals’ earnings.


 Data are from the National Supported Work project and Dehejia and Wahba (1999)
 Treatment is if a person received training (treatment)
 Independent variables are age, education, and married
 Outcome is real earnings (RE78)
 For the difference-in-differences model, the outcome is the difference in earnings after and
before treatment (REDIFF)

Treatment Number of obs Percent frequency


0 2,490 93%
1 185 7%

 We need to find matches for the 185 treated observations and then compare outcomes
 Note from the output that not all of the control observations were used as matches for the 185
treated observations.

2
Propensity score model (probit model)
Dependent variable is whether or not the individual participated in the program/treatment
Probit
coefficients
Age -0.05*
Education -0.17*
Married -1.47*

 Interpretation: individuals who are older, more educated, or married are less likely to receive
training.
 We are saving the propensity scores (predicted probabilities) from the probit model and using
them to find matches for the treated observations.
 The balancing property (similar characteristics between treated and control observations) is
satisfied.

3
Average treatment effect on the treated
Estimation method Differences Difference-in-differences
using one period data (RE78) using two period data (REDIFF)
T-test -15,204* 2,327*
Regression with dummy -6,901* 2,276*
ATET nearest neighbor -6,715* 2,989*
ATET radius matching -13,252* 2,215*
ATET kernel matching -7,009* 2,803*
ATET stratification matching -6,497* 2,776*
ATET matching using R -6,681* 2,604*
Results from Stata, except last row is from R

 Interpretation: After matching treated and control individuals, the effects of the training
program are to lower the earnings by about $6,000 to $13,000 using one period of data. In
other words, people who received training earn less than those that did not received training.
 Interpretation: Using two period data (the outcome variable is the difference of after and before
periods), the results show that people who received training increases their earnings by $2,000
to $3,000.
 We can also express the ATET as a percentage of the average values if scale is an issue (say
comparing earnings in countries that have different average incomes). The average earnings is
$20,502, so these are economically significant changes.
4

You might also like