ECON4150 - Introductory Econometrics Lecture 16: Instrumental ...
Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1...
Transcript of Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1...
![Page 1: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/1.jpg)
Seminar 9
ECON4150
University of Oslo
April 25, 2014
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 1 / 16
![Page 2: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/2.jpg)
Names dataset
Randomly assigned ”white-sounding” names and ”African americansounding” names to resumes and sent out as job application
Recored the number of ”call backs” from employers
Data used in article published in AER, 2004, Vol. 94, no. 4: ”AreEmily and Greg More Employable that Lakisha and Jamal? A FieldExperiment on Labor Market Discrimination”
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 2 / 16
![Page 3: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/3.jpg)
Key variables
Variable name Descriptionfirst name applicant’s first namefemale 1 = femaleblack 1 = blackhigh 1 =high quality resumecall back 1 =applicant was called backchicago 1 = data from Chigaco
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 3 / 16
![Page 4: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/4.jpg)
Call back rate by race
The call back rate for whites is 0.097 while for blacks is 0.064.
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 4 / 16
![Page 5: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/5.jpg)
Confidence interval
YW − Y B + /− 1.96 ∗ se(YW − Y B)
0.097 − 0.064 + /− 1.96 ∗√
(s2Wnw
+s2BnB
)
0.033 + /− 1.96 ∗√
(0.2952
2435+
0.2462
2435)
[0.018, 0.048]
Using equation 3.19 and s2w is sample variance for white and is 0.2952 andnw number of observation white = 2435. The difference is statisticallysignificant at a 5% level. The difference implies that 9,7% of resumes ofwhites receive call back while only 6,5% of resumes of black receive callback which is a large difference.
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 5 / 16
![Page 6: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/6.jpg)
Call back rate by race for men
White men have a call back rate of 8,8% while black men have a call backrate of 5,8% giving a difference of: 3%
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 6 / 16
![Page 7: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/7.jpg)
Call back rate by race for women
White women have a call-back rate of 9,9% while black women have a callback rate of 6,6% giving a difference of 3,3%.
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 7 / 16
![Page 8: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/8.jpg)
Significance of gender difference
The coefficient femaleblack = female*black and measure the differentialeffect of being black for female relative to male. The effect is insignificantindicating that the racial gap is equal for men and women.
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 8 / 16
![Page 9: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/9.jpg)
High versus low quality CV
High quality resumes have a call back rate of 8,7% while low qualityresumes have a call back rate of 7,3%. (difference 1,4%)
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 9 / 16
![Page 10: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/10.jpg)
High versus low quality CV - whites
High quality resumes have a call back rate of 10,8% while low qualityresumes have a call back rate of 8,5%. (difference (2,3%)
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 10 / 16
![Page 11: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/11.jpg)
High versus low quality CV - blacks
High quality resumes have a call back rate of 6,7% while low qualityresumes have a call back rate of 6,2%. (difference (0,5%)
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 11 / 16
![Page 12: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/12.jpg)
High vs low quality
(1) (2)call back call back
high 0.0141 0.0229(0.00779) (0.0120)
black -0.0231∗
(0.0106)
high black -0.0178(0.0156)
cons 0.0734∗∗∗ 0.0850∗∗∗
(0.00530) (0.00801)
N 4870 4870
Standard errors in parentheses∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 12 / 16
![Page 13: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/13.jpg)
High vs low quality
Black is significant, thus there is a significant difference betweenblack and whites
High is insignificant, thus there is no significant difference betweenhigh quality resumes and low quality resumes
Interaction high black is insignificant, thus there is no significantdifference in call back between blacks and whites with high qualityresumes.
Thus racial difference stems from a difference in call back ratesbetween whites with low quality resumes and blacks with low qualityresumes. Blacks with low quality resumes receive fewer call backsthan whites with low quality resumes (8,5%-6,2% = 2,3%)
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 13 / 16
![Page 14: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/14.jpg)
Failure to randomize
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 14 / 16
![Page 15: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/15.jpg)
Failure to randomize - variables we use
None of the coefficients are individually significant
Neither is the coefficients jointly significant as the p-value of theF-statistic of the full regression is 0.8385. (Joint test that all includedvariables are zero)
We cannot reject the null that treatment was randomly assigned.
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 15 / 16
![Page 16: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/16.jpg)
Failure to randomize - all variables
If you run the regression with all the control variables the p-value of theF-statistic of the full regression is still insignificant. However, if you lookindividually at computer skills it is statistically significant and indicatesthat the blacks have slightly higher computer skills. However, statistically,5% of the control variables will be significant at the 5% level even if thetrue effect is zero as the probability of falsely rejecting the null is 5%.
ECON4150 (Dep. of Economics) Seminar 9 April 25, 2014 16 / 16
![Page 17: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/17.jpg)
Seminar 9 - regular exercises
ECON4150
University of Oslo
April 25, 2014
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 1 / 9
![Page 18: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/18.jpg)
Exercise 13.3 a - Average treatment effect
Average treatment effect is given by the average SAT score in thetreatment group less the average SAT score in the control group:
X̄treatment − X̄control − 1241− 1201 = 40
The estimated effect of being assigned to a preparatory course is 40 pointson the SAT test.
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 2 / 9
![Page 19: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/19.jpg)
Exercise 13.3 b - Non-random assignment
Non-random assignment occurs if the treatment is not assigned randomlybut instead based (fully or partly) on characteristics or preferences of thesubjects. Example:
All participants were offered the preparatory course but only thosethat wanted took it
The information about the preparatory course was given over theschool e-mail so only those that check their e-mail could attend
Only those that struggled on the mid-term was offered the course
etc, etc.
If there is non-random assignment the outcome will reflect both the effectof the treatment and the effect of the non-random assignment and wecannot separate out the effect of the treatment as E (ui |Xi ) 6= 0
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 3 / 9
![Page 20: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/20.jpg)
Exercise 13.3 b - Non-random assignment
To estimate the pure treatment effect the treatment and control must beidentical on observables and non-observables.Test for non random assignment:
Are treatment and control identical on observables (W). Here:Gender distribution
Run a regression of treatment variable (Xi ) on the observables andcompute the F-statistics of whether the coefficients of the W’s arezero.
Randomization has failed of we can reject the joint hypothesis that allcoefficients are zero.
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 4 / 9
![Page 21: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/21.jpg)
Exercise 13.3 b - Non-random assignment
Manually when we have only one control variable:
Let p̂ be the probability that a male is assigned to treatment group.Which if there is complete randomization should be 0.5 thus:
H0 : p̂men = 0.5
H1 : p̂men 6= 0.5
tmen =p̂men − H0,1
se(p̂men)=
p̂men − H0,1√1n p̂men(1− p̂men)
=0.55− 0.50√1
1000.55(1− 0.55)= 1
From e.q. 2.7. σX =√p(1− p) when X is binary and
eq 2.48 st.dev(X̄ ) = σX√n
t90,10%two−sided = 1.66. Cannot reject the null. No evidence ofnon-random assignment
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 5 / 9
![Page 22: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/22.jpg)
Exercise 13.5a
Question: Is drop out of male ahtletes a threat to internal validity?
Potential problem Attrition - subjects drop out after they have beenrandomly assigned to treatment and control.Cause a bias in the OLS estimator if the attrition isrelated to the treatment itself because treatment andcontrol will differ on observables or unobservables. Thusthe results may be due to differences in who remained ineach group rather than the effect of the treatment.If attrition is related to the treatment we get a sampleselection bias and the treatment and the error term willbe correlated
Solution The attrition is here unrelated to the treatment and thus wedo not get a bias. However, the results are the averagetreatment effect on the population excluding male athleteswhich may be different from the average treatment effect ofthe full population.
No internal validity problem as the attrition is unrelated to the treatment.ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 6 / 9
![Page 23: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/23.jpg)
Exercise 13.5b
Question: Engineers in the control set up a LAN to share?
Problem: Failure to follow treatment protocol: When theparticipants in the experiment do not, or only partiallycomply with the treatment protocol.Subjects assigned to the control group receive treatmentor subjects assigned to treatment do not receivetreatment which bias the estimate of the treatmenteffect
Solution: If the experimenter knows who actually received thetreatment, the treatment effect can be estimated using theinitial random assignment as instrument for treatment.
Thus the LAN is a threat to internal validity as it creates a bias due tofailure to follow treatment protocol, however this bias can be corrected forby using IV.
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 7 / 9
![Page 24: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/24.jpg)
Exercise 13.5c
Question: The art majors never learn how to access their internetaccounts
Potential problem Failure to follow treatment protocol
Solution The treatment is defined as ”availability of internet in dormrooms”
The art majors have received the treatment and thus there is no threat tointernal validity. (Treatment is not defined as ”use of internet in dormrooms”.
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 8 / 9
![Page 25: Seminar 9 - Forsiden - Universitetet i Oslo€¦ · high 1 =high quality resume call back 1 =applicant was called back chicago 1 = data from ... ECON4150 (Dep. of Economics) Seminar](https://reader034.fdocuments.net/reader034/viewer/2022042301/5eccaad0a0af283cb576d7ae/html5/thumbnails/25.jpg)
Exercise 13.5d
Question: Economics students sell access to their internet
Potential problem: Failure to follow treatment protocol, students thatwere not supposed to receive treatment do recieve treatment
Solution: Use the initial random assignment as instrument fortreatment if you have information about who actuallyreceived treatment
Answer: Again failure to follow treatment protocol as in b. and the OLS isbiased but can be corrected for using IV if the experimenter knows whoreceived access and who was initially assigned access.
ECON4150 (Dep. of Economics) Seminar 9 - regular exercises April 25, 2014 9 / 9