Economics Paper 10 Pages(Due In 24 Hours)
Paper 1: First Draft
Use the results from the first two homework assignments, what you have learned in class and from reading the papers and textbook for the course to answer the following questions. Did assigning a person to get an encouraging phone call increase their probability of voting? Can we get an estimate of the causal effect of getting a call encouraging you to vote on your probability of voting with non experimental data by using regression to adjust for differences between people who got a call and those that didn’t (why or why not)? Be clear on what is going wrong.
Structure of Paper
You can use the papers on the reading list as a guide for what a paper should look like. You can include your tables in the body of the paper or put them at the end of the paper. Your paper should be 10 pages long wihout counting tables and you may want to include the following sections:
Abstract: One paragraphs that sums the paper up. Write this first.
Intro: One page that sums the paper up with more detail than the abstract. Should tell the reader why the questions the paper answers are important and cover data, econometric methods, results and conclusion.
Data: Describe the data you use in the analysis and how it was generated. You may need to do some research online.
Methods: Describe the statistical methods used. Include the equations for the regressions you will run.
Results: Describe and interpret your statistical findings.
Conclusion: Interpret your findings.
*1
eststo control: estpost sum newreg contact busy age female vote98 vote00 vote02 if treat_real==0
eststo treatment: estpost sum newreg contact busy age female vote98 vote00 vote02 if treat_real==1
eststo difference: estpost ttest newreg contact busy age female vote98 vote00 vote02, by (treat_real)unequal
esttab control treatment difference using “HW1.csv”, cells(“mean(pattern(1 1 0) fmt(3))sd(pattern(1 1 0) fmt(3)) b(star pattern(0 0 1) fmt (3))p(pattern(0 0 1) fmt(4))”) title(“HW1.csv”)label replace plain
*2Yes, it is correctly implemented, the treated and untreated groups are nearly the same, shows they are randomly enough.
*3
ttest vote02, by(treat_real)
*t-stat (3.5212) is higher than the t-critical,and the p-value 0.0002 is small.it is statistically significant, yet not large in a practical sense.
*4
reg vote02 treat_real
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
tab treat_real, summarize(vote02)
reg vote02 treat_real newreg
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
reg vote02 treat_real newreg contact
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
reg vote02 treat_real newreg contact
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
reg vote02 treat_real newreg contact age
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
reg vote02 treat_real newreg contact age female
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
reg vote02 treat_real newreg contact age female vote98
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
reg vote02 treat_real newreg contact age female vote98 vote00
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
reg vote02 treat_real newreg contact age female vote98 vote00 state
outreg2 using “reg_1.txt”, replace nonote se label bdec(3)
*5
*more covariates leads to more precision and less bias.
*6
*we won’t get unbiased estimate of the causal effect because of the selection bais. not all the people will answer the phone due to all sorts of different reasons.
*1
eststo control: estpost sum newreg contact busy age female vote98 vote00 vote02 if contact==0
eststo treatment: estpost sum newreg contact busy age female vote98 vote00 vote02 if contact==1
eststo difference: estpost ttest newreg contact busy age female vote98 vote00 vote02, by (contact)unequal
esttab control treatment difference using “HW2.csv”, cells(“mean(pattern(1 1 0) fmt(3))sd(pattern(1 1 0) fmt(3)) b(star pattern(0 0 1) fmt (3))p(pattern(0 0 1) fmt(4))”) title(“HW2.csv”)label replace plain
*2
* no, Table 1 does not suggest that the control group will provide a good counter factual for the treatment group’s voting potential outcomes. almost every part are different except female.
*3
*the different group in homework1 are more similar, hm2 does not. as we know, old people are much more likely to answer the phone.
*4
reg vote02 contact
outreg2 using “reg_2.numbers”, replace nonote se label bdec(3)
tab contact, summarize(vote02)
reg vote02 contact newreg
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
reg vote02 contact newreg
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
reg vote02 contact newreg
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
reg vote02 contact newreg age
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
reg vote02 contact newreg age female
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
reg vote02 contact newreg age female vote98
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
reg vote02 contact newreg age female vote98 vote00
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
reg vote02 contact newreg age female vote98 vote00 state
outreg2 using “reg_2.numbers”, append nonote se label bdec(3)
*5
*adding covariates reduces bias and increase precision, the relationship between the covariates and the outcome are mostly positive.
*6
*hw1 has not very significant effect unlike the hm2. The tables are different for the treatment and control groups are different. in hm1, standard errors remain the same as adding the covariates. in hm2 when the covariates were added, the standard errors for contact decreased.
*7
*yes, adding covariates to the regression as in question 4 reduce the bias, the upward bias of contact was cancel when the variables were added. one of the most is the age for it has most different means.
*8
*no, the regression does not eliminate all the bias. he treatment and control groups have a very different means.
1
Brandon Lang
Econ 104
Prof. Dobkin
1/31/18
Paper One
Abstract
This paper atte
m
pts to analyze the effect of sending eligible voters in Iowa a pre-recorded phone
call message telling them to vote in an upcoming election in 2002 to find if their likelihood of
voting is changed. To study the effects of the call, two approaches are used: one in which the
treatment group is assigned by subjects simply being sent the message with the control group
not be sent it, and the second in which the treatment group is assigned by subjects listening to
the message to completion and the control group not listening to the message. All kinds of data
are pooled together to help isolate the effect of the treatment when producing regression
estimates. Although the listening to completion treatment shows more of an effect on subjects’
likelihood of voting, both methods produce murky results as they are both subject to bias.
Introduction
Data
Methods
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
Th
is
stu
dy
re
so
ur
ce
w
as
sh
ar
ed
v
ia
C
ou
rs
eH
er
o.
co
m
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
2
vote02 = B0+B1treat_real+u
Results
Using the treatment variable:
Table 1
untreated treated difference
mean sd mean sd b p
vote02 0.59 0.49 0.61 0.49 -0.01** (0.01)
contact 0.00 0.00 0.46 0.50 -0.46*** (0.00)
newreg 0.05 0.21 0.05 0.22 -0.00 (0.57)
busy 0.00 0.00 0.03 0.17 -0.03*** (0.00)
age 55.80 18.95 55.78 18.82 0.01 (0.94)
female 0.56 0.50 0.56 0.50 0.00 (0.35)
main voter:
turnout in 01
0.73 0.44 0.73 0.44 0.00 (0.67)
vote98 0.57 0.49 0.57 0.49 -0.00 (0.67)
county 59.70 30.64 59.55 30.70 0.16 (0.56)
Observations 85931 15000 100931
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
This table shows the comparison of the means, standard deviations, and p-values of the
differences between the treatment and control groups. Results were successfully randomized, as
is indicated by the means for almost all of the sample characteristics being nearly or completely
identical between the treated and untreated groups, with the only exceptions being “contact”
and “busy” due to the untreated group not having received calls.
The difference in voting rates for the treatment and control groups is .0120234, or an increase
1.2 percentage points in the likelihood of voting. It is statistically significant as the t-stat (2.77) is
higher than the t-critical (1.960) and the p-value is very small (0.006). Even though it is statistically
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
Th
is
stu
dy
re
so
ur
ce
w
as
sh
ar
ed
v
ia
C
ou
rs
eH
er
o.
co
m
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
3
significant, the actual practical effect is very small.
Table 2
(1) (2) (3) (4) (5) (6) (7)
VARIABLES vote02 vote02 vote02 vote02 vote02 vote02 vote02
treat_real 0.0120**
*
0.012**
*
0.013**
*
0.012**
*
0.012**
*
0.012**
*
0.012**
*
(0.00434
)
(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)
newreg
–
0.315**
*
0.105**
*
0.160**
*
0.173**
*
0.160**
*
0.160**
*
(0.007) (0.007) (0.006) (0.006) (0.007) (0.007)
main voter: turnout in
01
0.582**
*
0.443**
*
0.442**
*
0.400**
*
0.400**
*
(0.003) (0.003) (0.003) (0.004) (0.004)
vote98
0.274**
*
0.257**
*
0.274**
*
0.275**
*
(0.003) (0.003) (0.003) (0.003)
age
0.001**
*
0.002**
*
0.002**
*
(0.000) (0.000) (0.000)
female
–
0.023**
*
–
0.023**
*
(0.003) (0.003)
busy
-0.016
(0.020)
Constant 0.594*** 0.609**
*
0.162**
*
0.104**
*
0.039**
*
0.065**
*
0.065**
*
(0.00167
)
(0.002) (0.003) (0.003) (0.005) (0.005) (0.005)
Observations 100,931 100,931 100,931 100,931 100,931 98,327 98,327
R-squared 0.000 0.019 0.260 0.318 0.321 0.304 0.304
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
This table shows the effects of regressing the effect of the treatment on the likelihood of a subject
voting in 2002. Covariates are added progressively to reduce bias and make more accurate
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
Th
is
stu
dy
re
so
ur
ce
w
as
sh
ar
ed
v
ia
C
ou
rs
eH
er
o.
co
m
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
4
estimates as to the effect of the treatment variable. Added covariates include whether or not a
subject is a newly registered voter (newreg), whether or not they voted in the 2000 election
(vote00), whether or not they voted in the 1998 election (vote98), the average age of the subjects
(age), whether or not the subject is female (female), and if the subject receiving the message was
busy when called (busy).
The addition of more covariates reduces bias and increases the precision of a model’s prediction,
as the inclusion of more variables that are strong predictors of the outcome reduces dilution in
the regression. This indicates that the treatment effect on the likelihood of voting is understated
without the inclusion of the other covariates.
Using the contact variable:
Table 3
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Table 4
NoContact Contact Difference
mean sd mean sd b p
vote02 0.59 0.49 0.67 0.47 -0.08*** (0.00)
age 55.59 18.93 58.57 18.74 -2.98*** (0.00)
female 0.56 0.50 0.58 0.49 -0.02* (0.01)
newreg 0.05 0.22 0.04 0.21 0.00 (0.10)
Vote98 0.57 0.50 0.61 0.49 -0.04*** (0.00)
main voter:
turno~01
0.73 0.44 0.77 0.42 -0.04*** (0.00)
Observations 94021 6910 100931
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
Th
is
stu
dy
re
so
ur
ce
w
as
sh
ar
ed
v
ia
C
ou
rs
eH
er
o.
co
m
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
5
(1) (2) (3) (4) (5) (6)
VARIABLES vote02 vote02 vote02 vote02 vote02 vote02
contact 0.0768*** 0.060*** 0.052*** 0.052*** 0.046*** 0.040***
(0.00611) (0.006) (0.006) (0.006) (0.005) (0.005)
age
0.006*** 0.006*** 0.006*** 0.002*** 0.002***
(0.000) (0.000) (0.000) (0.000) (0.000)
female
-0.029*** -0.029*** -0.023*** -0.023***
(0.003) (0.003) (0.003) (0.003)
newreg
-0.221*** -0.047*** 0.159***
(0.007) (0.007) (0.007)
vote98
0.421*** 0.274***
(0.003) (0.003)
main voter:
turnout in 01
0.399***
(0.004)
Constant 0.591*** 0.272*** 0.279*** 0.316*** 0.271*** 0.065***
(0.00160) (0.005) (0.005) (0.005) (0.005) (0.005)
Observations 100,931 100,931 98,327 98,327 98,327 98,327
R-squared 0.002 0.050 0.058 0.067 0.219 0.304
Standard errors in parentheses
***p<0.01, **p<0.05, *p<0.1
Conclusion
Although It is apparent that the control group in Table 3 will not provide a good counter factual
for the treatment group earnings outcomes as the results in every category are very different in
most categories except for female. This indicates that the observed groups aren’t similar enough
and that bias is likely present. The fact that this is no longer completely randomly assigned but
also has an opt-in element is not accounted for and likely contributes to the different outcomes.
When focusing on the effects of subjects simply receiving the treatment, the means of the
treated and control groups were nearly identical, but this table shows greater differences
between the groups which are unlikely to simply be the result of random chance. A potentially
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
Th
is
stu
dy
re
so
ur
ce
w
as
sh
ar
ed
v
ia
C
ou
rs
eH
er
o.
co
m
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
6
major contributing difference may be that the kinds of people who would pick up the phone to
listen to the message to completion are different from those who wouldn’t, with one potential
difference being that those who opt-in could be unemployed. This could mean that subjects who
listen to the message are likely to be either school-aged young people or elderly retired people.
The addition of covariates reduced the bias given that as more variables were added, the upward
bias of contact was diminished. The covariate which reduced the bias the most was age,
decreasing the effect of contact on the outcome by 1.8%. A probable reason for this could be
that, as speculated previously, a significant portion of the treatment group could be retirees. The
mean age of the treatment group being about 58 years old indicates that this is probable. As this
population is likely able to spend more time at home on account of being unemployed, it’s
reasonable to say that they have more free time to be at home to pick up the phone when called
and listen to the message to completion, potentially increasing their likelihood of voting.
Table 1 shows that the treatment and control groups have different means which indicates that
they aren’t similar enough to draw accurate conclusions about the treatment’s effects without
bias. Due to the contact variable being opt-in, the randomness of the experiment was diminished.
Due to the it being indicated that age has a large effect on contact, it is possible that older people
are more likely to vote generally, so even though they may listen to the message to completion,
they may have intended to go out to vote before even receiving it, making the treatment
pointless and giving it an upward bias in its effect on the outcome. The absence of other variables
which may effect subjects’ likelihood of voting in the election very likely causes further upward
bias for the treatment.
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
Th
is
stu
dy
re
so
ur
ce
w
as
sh
ar
ed
v
ia
C
ou
rs
eH
er
o.
co
m
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
7
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
Th
is
stu
dy
re
so
ur
ce
w
as
sh
ar
ed
v
ia
C
ou
rs
eH
er
o.
co
m
Powered by TCPDF (www.tcpdf.org)
https://www.coursehero.com/file/36871964/Paper-1-pdfpdf/
http://www.tcpdf.org