Från Tentor Flashcards

1
Q

Consider the regression model above and say that the outcome is binary (either zero or one). Your
regression estimate, 𝛽1̂ = 0.05. How do you interpret this estimate?

A

The treatment increased Y by 5 percentage points.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Which of the following statements is correct regarding measurement errors?

A. Random measurement error in the outcome variable leads to inconsistent and biased estimates.

B. Random measurement error in the treatment variable leads to attenuation bias.

C. Random measurement error in the treatment variable only increase standard errors.

D. Random measurement error is not a problem in any variable as long as it is strictly random.

A

B. Random measurement error in the treatment variable leads to attenuation bias.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Including outcome variables in a regression model as control variables is:

A

C. Problematic as it can lead to bias in the estimated coefficient of interest.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

𝐶𝑟𝑖𝑚𝑒𝑖𝑡 = 𝛽0 + 𝛽1 𝑃𝑜𝑙𝑖𝑐𝑒𝑖𝑡 + 𝜖𝑖𝑡

Write down the fixed effects regression model based on the regression model above.

Based on this equation, describe what kind of problem you can potentially solve using fixed effects compared to the equation above?

What kind of problems can a fixed effects model not solve? (5 points)

A

Här inkluderar vi αi i stället för b0.

Som är fixerade effekter saker som är konstant över tid men som varierar mellan stater.

Vi kan ocksåp addera λt som är saker som är konstant över stater men som varierar över tid.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝑖𝑡 = 𝛽0 +𝛽1𝑇𝑖 +𝛾1(𝑇𝑖 ∗𝑌𝑟1990)+𝛾2(𝑇𝑖 ∗𝑌𝑟1993)+𝛾3(𝑇𝑖 ∗𝑌𝑟1999)+⁡𝜆𝑡 +⁡𝛼+ui

Vilken koefficient visar den kausala effekten av behandlingen?

1999 var behandlingen

A

γ3

Det är skillnaden mellan de som fick behandlingen och de som inte fick det.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

In an IV setting, I can show that my instrument is unrelated to pre-determined variables using regression
models. Which argument can these results strengthen?

A

Att den är exogen

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

The average marginal effect in the model ln Y = β0 + β1X + β2X^2 can be formulated as…..

A

β1 + 2β2Xbar^2

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Consider the following equation: 𝑌𝑖 = 𝛽0 + 𝛽1𝑋1𝑖 + 𝛽2𝑋2𝑖 + 𝑢𝑖, whmostly interested in 𝛽1. Leaving 𝑋2 out of the equation is a problem only if…….

A

𝛽2 ≠ 0 and 𝑋2𝑖 is correlated with 𝑋1t.. b2 ≠ betyder att x2 har en effect på y.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

A proxy variable is used…..

A

to decrease or remove the omitted variables bias from the estimated
coefficient of the variable of interest.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Regression discontinuity requires that….

A

There is no discontinuity in the density of the running variable at the threshold.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

A measurement error in 𝑋𝑖 is similar to the omitted variables problem imaking our point estimates inconsistent. ?

A

Yes

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

If the true β1 is positive, random measurement error in the independent variable X will

A

Underestimate β1

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Varför skiljer sig IV estimat från OLS estimat?

A

The IV estimate is going to be larger than OLS because the numerator>denominator in the IV model, which is Cov(Y,Z)/Cov(X1,Z). It’s either that the numerator is blowing up, because there is a connection between Z and X1, or the first stage is weak.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Om en förändring i vår X variabel på 10% ger en minskning i Y på -2. Vad är då β1?

Givet Y = β0 +β1X1 + e

A

Y = -2
β0 är okänd, så vi skippar den.
X1 = 10

-2 = β1(10)

Lös ut β1

β1 = -2/10

β1 = 0,2

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Vad händer om vi har mätfel i X?

A

Vårt b1 estimat går mot noll! ”Bias towards zero”.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Vad händer om vi får mätfel i Y?

A

Givet vanliga mätfel så kommer det att öka variansen i regressionen och β1^. Men det ger inget bias i β1.

17
Q

We cannot use a regression discontinuity design if….

A

There is no discontinuity in the probability of being treated at the
threshold.

18
Q

Paneldata:

Regression gällande pris före och efter skatt och hur de påverkar försäljningarna av varor. Vi har data från två tillfällen mellan två länder.

Förklara i detalj vilken regression du skulle köra.

A

Det står paneldata i frågan osv. Men eleven har beskrivit en diff in diff.

Yi + β0 + β1Τi + β2afteri + β3Ti * afteri + ui

Där Ti är en dummy som indikerar huruvida det är ett land som får behandlingen eller ej. Ti = 1 om behandling och = 0 annars.

Efter är en dummy för vilken period man är i

Sen har vi interaktionstermen Ti*after som är = 1 om vi tittar på ett land i efterperioden som har fått behandling.

β3 är den koefficienten vi är intresserade av och vilken visar den kausala behandlingseffekten.

19
Q

Hur testar man parallell trend antagandet vid diff-in-diff?

A

Det finns inget statistiskt test eftersom att CF inte går att observera.

Vi kan därför istället grafiskt kolla på de parallella gränderna mellan de olika länderna och se så att de är lilla varandra innan. Vi kan då anta att behandlingslandet skulle följa samma utveckling som kontroll landet om ingen behandling hade satts in.

Tror också man kan göra placebo test?

20
Q

Many countries obligate sellers to show after tax prices on price tags. The reasoning behind such legislation is that people don’t fully take into account taxes if before tax prices are shown. We would like to test whether consumers are really prone to this
mistake and making after tax prices more salient helps them to overcome their bias. Assume there are two countries next to each other (Country 1 and Country 2). Before 2010, both countries allow retailers to show before tax prices on price tags but from 2011 Country 1 obligates them to show after tax prices (Country 2 sticks to its old
policy). You have access to the monthly sales data of 100 retail stores from both
countries for 2005-2015.

You do a diff n diff. Then Assume you get access to the sales data of 100 stores in Country 3 which
introduced the after tax prices law in 2013. How would you modify your
empirical strategy to exploit this data too? (4 points)

A

I den nya modellen skulle vi inkludera dummys (D) för varje år (utöver 2010, det relevanta året) och skapa en ny interaktionsterm för den andra behandlingseffekten 2013.

Ti = behandling

Yi + βo + β1Ti + β2d2009 + β3d2011 + β4d2012 + β5d2013 + βd62014 + δ1Ti2011 + δtid2013 + ui