Papers

Question 1

What question is Krueger trying to answer?

Answer 1

Studying the causal effect of class size on test scores. Does spending more money on education (e.g employing more teachers) improve learning outcomes?

Question 2

What is Krugers selection problem?

Answer 2

If he uses observational data on class sizes and student outcomes he’ll get bias results, as it won’t be randomised.
BB- Overestimation (if rich students go to school with smaller class sizes) Underestimation- Within schools, if performing students are sent to smaller classes

Question 3

What does Table 1 show Krueger and was the randomisation successful?

Answer 3

Table 1 shows the means of 3 different treatment groups. The ‘joint p-value’ was small, which usually means rejecting the null however, Kruger argues children were only randomly assigned within schools. Experiment was successful in randomisation.

Question 4

What does the estimate of Beta(1) tell Krueger?
Y[ics] = Beta[0] + Beta[1]SMALL[cs] + Beta[2]REG/A[cs] + Beta[3]X[ics] + alpha[s] + e[ics]

Answer 4

Beta(1) is the estimate of the difference in Y between children in ‘small’ classes and children in ‘regular’ classes (base group)

Question 5

In Kruegers regression what does each element mean?

Y(ics) = Beta(0) + Beta(1)SMALL + Beta(2)REG/A + Beta(3)X(ics) + alpha(s) + e(ics)

Answer 5

Y(ics) - Average percentile score on the SATs of student i in class c
SMALL - DV
REG/A - DV, reg with aide
X(ics) - vector of observed student and teacher covariates
alpha(s) - set of DV’s, one for each school. Included as random allocation was done within schools.
Beta(1)- estimate of difference in Y between children in SMALL and REG classes.
Beta(2)
Beta(3)

Question 6

What is the difference between the RF and the Actual class size in Kruegers regression?

Answer 6

As attrition was non-random, he’s worried that ‘Actual class size’ is non random.
Reduced form uses ‘allocated class size’ as treatment.

Question 7

From table 5 in Krueger, what is the outcome of the test scores?

Answer 7

Being in a small class increases test score percentile by; 5 points in Kindergarten, 7 points in First Grade, 6 points in Second Grade and 5 points in Third Grade. All statistically significant.

Question 8

What is the causal relationship DiNardo & Pischke trying to estimate?

Answer 8

The causal effect of computer use on wages (building on Kruger 1993). More-so trying to estimate the effect of computer skills on wages.

Question 9

What is the selection problem for DiNardo, and define what the BB and DTE bias means in this case?

Answer 9

The selection problem is that those who use computers (or those with computer skills) is not randomly allocated.

BB- Those who use computers at work would have earned more than those who don’t, in absence of the treatment.

DTE - Those who benefit most from using a computer at work are those who are allocated computers.

Question 10

What evidence is there from Table 1 that computers are not randomly allocated?

Answer 10

Table 1 shows that computer uses are more likely to be educated. White collar jobs have a much higher percentage, so E[x|D=1] does not equal E[x|D=0].

Question 11

Explain why the raw differentials and OLS differentials are different in DiNardo?

Answer 11

There are characteristics X which are positively correlated with computer use and with wages. The OVB formula can explain this.

Question 12

What do DiNardo & Pischke’s results tells us about the CIA?

Answer 12

The CIA is unlikely to be satisfied. They think it’s unlikely that using white-collar tools (computers) actually proxies for writing skills, as everyone has these. Instead they think that there’s a selection into the use of these tools.

Question 13

What is the causal relationship Dale and Krueger are trying to estimate?

Answer 13

The effect of attending a more selective college on earnings.

Question 14

What is the selection problem for Dale and Krueger?

Answer 14

Selective colleges might select students that have a higher potential to earn (e.g parents education, parents income - determines the quality of school attended).

Question 15

What would be the ideal experiment to solve this selective colleges problem?

Answer 15

To randomly allocate students to colleges

Question 16

How do Dale and Krueger solve their selection problem, and is the CIA justified?

Answer 16

They compare students who were accepted by similar colleges.

If the values of X(2) for these students is very similar, CIA could be justified..

Question 17

Write down the fully saturated model for 3 groups of students, who have identical acceptances and rejections, and explain how it would produce causal estimates. What’s the idea?

Answer 17

Ln(wi) = B(0) + B(1)SAT(j) + B(2)G2 + B(3)G3 + e(i)
G2 - DV for student in group 2
Base group is group 1.
The idea is that students in the same group have the same unobserved ability.

Question 18

What is the causal relationship Angrist & Evans are trying to estimate?

Answer 18

The causal effect of family size (fertility) on labour supply.

Question 19

Why would a regression of hours worked on family size might not reveal this causal realtionship on labour supply?

Answer 19

Due to endogeneity: fertility and labour supply are jointly determined.
Higher female labour force participation -> lower fertility (family size)
Lower fertility ->higher labour force participation

Question 20

What instruments do Angrist & Evans use, are they plausible, do they meet the 3 assumptions?

Answer 20

2 instruments- dummy for having the first 2 children with the same sex and a dummy for twin birth.
Same sex is basically randomly assigned but twin birth isn’t (older mother more likely), but can be controlled.
Both meet relevance criteria (causal effect on the treatment).
Possible that same sex or twins have some other effect on labour supply (exclusion restriction) so they don’t meet the restrictions.

Question 21

How do Angrist & Evan’s accommodate for the difference in their estimates when using different instruments, what do the suggest and show as a result?

Answer 21

They suggest the difference can be attributed to the age of the third child (twinning the third child is older than that of third birth).

They show that if the age of the child is properly accounted for, the 2 IV estimates become very similar.

Question 22

What is the causal relationship that Levitt is trying to estimate?

Answer 22

The causal effect of police on crime, seeing if a larger police size reduces crime.

Question 23

Why is the variable of interest endogenous for Levitt?

Answer 23

As within cities when crime increases the size of the police force increases (positive relation). More police might also result in more reporting of crimes.

Question 24

How does Levitt deal with the endogeneity problem?

Answer 24

The time period for mayoral and gubernatorial elections are used as IV’s. They are valid if; police force increases in the election year (causal effect), indpendent of potential outcomes (election cycles are pre determined), however they don’t satisfy exclusion restriction (but can be controlled in a 2SLS setting.

Question 25

Why does one of the instruments in Levitt paper, might not satisfy the exclusion restriction?

Answer 25

If there’s is a relationship between electoral cycles and some sort of economic performance.

Question 26

What evidence in Figure 2 and Table 2 which is used to justify the identification strategy (observational data)? Levitt

Answer 26

F2 - shows changes in police force tend to be larger in election years.
T2 - Displays the first-stage estimates with and without various sets of covariates.

Question 27

Why does Levitt use a model looking at the relationship between within-city change in crime and the with-in change in police numbers?

Answer 27

Problem of ‘unobserved heterogeneity’ - idea that some cities might have high crime rate and high police numbers and vice versa.

Question 28

What is Ludwig & Millers research question?

Answer 28

The causal effect of pre school education and other service (‘Head start’) to poor children in their schooling and health?

Question 29

Explain the Regression L&M use and what is the treatment assignment and what does it compare?

Yc, Pc and cutoff?

Answer 29

The treatment assignment mechanism is a discontinuity - government provided additional funding to 300 poorest counties.

The comparison is between people in counties just below and just above the cut off.

Yc - Outcome for county c
Pc - County’s poverty rate (1 is poorest)
Cutoff occurs at c = 300

Question 30

Why do L&M estimate an ‘Intention To Treat’ model rather than a fuzzy RD model, ideally what would they do?

Answer 30

Because the treatment is the allocation of additional funding to set up Head Start, but they don’t know to what extent counties who recieved additional funding, spent on such services.

Ideally, they would scale the intention to treat by the effect of Gc (additional assistance) on Head Start itself to form a Fuzzy RD.

Question 31

What is the key indentification assumption for L&M, and what does it mean?

Answer 31

Either side of the discontinuity, Head Start treatment is as good as randomly assigned.

This means that, at the discontinuity, the potential outcomes are independent of treatment.

Question 32

What might violate the indentification assumption in L&M?

Answer 32

If counties could manipulate the poverty rate to obtain extra funding (unlikely as government used historical poverty rates to determine allocation).

Also if the cutoff was used for other funding.

Question 33

What is Bleemer & Mehta’s research question?

Answer 33

Studying the causal effect of choosing Economics as a major on wages.

Question 34

Give 2 reasons why comparison of average wages between Econ grads and other grads from social sciences might not reflect the causal impact of studying economics on wages

Answer 34

Non-random selection into majors. Students self select (determined by student background and characteristics) and colleges might have requirements for particular subjects.

Question 35

What econometric method do they use to solve B&M’s selection problem?

Answer 35

Fuzzy RD method - they compare students who just achieved the required GPA with students who just missed the required GPA.

Question 36

What evidence do B&M provide that the indentification assumptions were met?

Answer 36

The potential outcomes, Y(econ) Y(Not econ), are smooth at the discontinuity e.g (wages are smooth across discontinuity before policy was implemented)

No evidence of bunching around the threshold.

Question 37

Interpret Figure 1&2 from B&M and use them to estimate the causal effect

Answer 37

F1 - shows that the probability of majoring in Econ increases by 36.1% per person(2.7 SE), so statistically significant- first stage.

F2 - shows earnings increased by almost $8000 with a standard deviation of almost $2. Reduced form.
The ratio of the two gives an IV estimate of 8000/0.361

Question 38

What Regression framework do Goldin & Rouse use to estimate the impact of blind auditions on female musicians, and explain the elements?

Answer 38

They use a DiD regression: P[i jtr] = alpha + BetaF[i] + gammaB[jtr] + Delta(F[i] x B[jtr] + …

F[i] - female DV
B[jtr] - Screen DV
(F[i] x B[jtr]) -Interaction of Female and Screen dummies
Delta - DiD estimate

Question 39

Why are the results in Table 4 surprising, and how do they explain the results in G&R?

Answer 39

It shows that the female success rate is actually lower in blind auditions. Some estimates are statistically significant.

They suggest that’s this is because the basic DiD model does not hold ‘ability’ constant.

Question 40

How do G&R modify their basic regression model to deal with the results of table 4?

Answer 40

They incorporate individual fixed effects - therefore controls for ‘ability’. The table shows linear probability estimates

Question 41

Explain the results of table 6 in G&R

Answer 41

They show that Delta (DiD estimate ) is positive in some, but not all audition rounds. For example, the effect of a blind audition in column 2 shows an increase in success probability, but it’s not very precisely estimated.

Question 42

What is the research question in Currie & Walker?

Answer 42

What is the health effect of traffic congestion (emissions). In particular, what was the effect of a policy which reduced congestion on birthweight and prematurity.

Question 43

Give 2 reasons why a comparison of births in more and less polluted areas might not reflect the causal impact of pollution on child health

Answer 43

‘Correlation study cannot demonstrate a causal effect’.

This is because wealthier people tend to move away from motorways, so women living near motorways are more likely to have other characteristics linked too poor pregnancy (lower income & education, higher prob of being a teen mother).

OR

Emission sources tend to be located in urban areas, where individuals maybe more educated, better access to healthcare etc

Question 44

What is the indentification assumption C&W use? What evidence do they provide to support it?

Answer 44

Parallel Trends assumption

The assumption is that the characteristics of mothers near a toll plaza, change over time, in a way that’s comparable, to those of mothers who live further away.

Important to consider that people may move in repsense to the introduction of the EZPass.

Question 45

Why do C&W include ‘mother fixed-effects’ in some of their analyses?

Answer 45

So they can directly control for possible changes in the composition of mothers before and after the introduction of EZPass.

Question 46

Simplify C&W’s regression model and explain the terms

Answer 46

Basic DiD model: Outcome[it] = a + b[1]EZPass[it] + b[2]Close[it] + b[4]EZPass[it] x Close[it]

b[1] - Difference in outcome (birthweight) between births before and after intro if EZPass
b[2] - Difference in outcome between births of mothers who live close to a toll and those who live far away
b[4] - DiD estimate: Difference between outcome before and after intro on EZPass between mothers who love close and further away

Question 47

Describe the first stage and reduced form in B&M’s Method?

Answer 47

First Stage - Effect of discontinuity on the probability of being an economics major.

Reduced Form - The effect of the discontinuity on earnings.