Question 1

Quantile regression

Accepted Answer

A)  estimates marginal effects at the mean values of the independent variables.
B)  results in biased estimates for skewed distributions.
C)  can be estimated in Excel.
D)  results in estimates approximating either the median or other percentiles of the dependent variable. D

Question 2

Suppose you are interested in testing the claim that students who participate in band perform better on standardized tests but you are worried that the results might be biased because individuals are likely to self-select into joining the band.You only have test scores for those students that participate in band.Suppose that for a sample of 14,111 6^th-grade students you estimate the Heckman selection model (marginal effects listed,standard errors in parentheses) Second-Stage: $$\begin{aligned} Tes\widehat {t S}core _ { i } = & 58.932 +& 1.374 \lambda _ { i } +& 0.583 \text { Income } _ { i } + &2.156 \text { Male } _ { i } \ & ( 7.961 ) &( 0.358 )& ( 0.136 )&(1.679) \end{aligned}$$ First-Stage: $$\begin{array} { l } {\widehat{\text { Band }_{i}}}=&0.261+&0.045 \text { Siblings in Band }_{i}+&0.048 \text { Income }_{i}+&0.125 \text { Male }_{i}\ &( 0.043 ) &( 0.022 ) &( 0.012 )&(0.103) \ \end{array}$$ a)Explain why OLS is inappropriate in this circumstance and how this model improves on OLS. b)Which variable are you using to identify the model? Does this choice seem correct? Explain. c)Discuss the results above.

Accepted Answer

a)OLS is biased in the presence of sample-selection bias because the independent variables are correlated with the error term.It occurs when individuals non-randomly select themselves into a given outcome of the dependent variable.You can control for it using a Heckman two-stage correction in which you estimate the selection decision in the first-stage and use those results to calculate inverse Mill's ratios which are included in the second-stage to control for the potential bias.
b)Siblings in band is the variable that is use to identify the model.This variable should be correlated with band but uncorrelated with test score.
c)All variables are statistically significant.Most importantly,the lambda is significant implying that selection does impact this model.

Question 3

What is a difference-in-difference estimator? When is it appropriate to use one? How do you do so? Explain.

Accepted Answer

Difference-in-difference estimators attempt to replicate randomized clinical trials by comparing treatment and control groups before and after a treatment is imposed to estimate the impact of a given policy intervention.They are appropriate to use when you can identify a clear treatment and when you can observe a sufficient number of observations in the treatment and control group.You perform difference-in-difference estimation by:(1)identifying a policy intervention that affected a specific "treatment" group but not a specific "control" group, (2)gathering data before and after the policy intervention for the two groups, (3)estimating the population regression model

$$y _ { i } = \beta _ { 0 } + \beta _ { 1 } \text { After } _ { i } + \beta _ { 2 } \text { Treatment } _ { i } + \beta _ { 3 } ( \text { After } * \text { Treatment } ) _ { i } + \varepsilon _ { i }$$
by OLS,and (4)interpreting $\hat { \beta } _ { 3 }$ 
as the desired difference-in-differences estimator.

Question 4

When performing difference-in-difference estimation,the control group is the group

Accepted Answer

A)  for which the policy shock occurred.
B)  for which the policy shock did not occur.
C)  of observation in the "before" sample.
D)  of observation in the "after" sample.

Question 5

Suppose you wish to determine factors affecting the number of surfers observed surfing at a given surf spot,an appropriate model to estimate the model would be

Accepted Answer

A)  OLS.
B)  the logit.
C)  the ordered probit.
D)  the Poisson model.

Question 6

Suppose you wish to explain the number of nights per week that individuals eat dinner at a restaurant,an appropriate model to estimate would be

Accepted Answer

A)  Weighted Least Squares.
B)  the negative binomial model.
C)  OLS.
D)  the probit.

Question 7

Suppose you are interested in explaining the effect that family income (thousands)has on child birth weight and you are concerned that the true marginal effects differ for at different points in the birth weight distribution.In a sample of 22,365 live births,you estimate following $\begin{array}{llll}
10^{\text {th }}  \text { Percentile:} &  Burt \widehat {h W }etght _ { i } =&5.583+0.086 \text { Income }_{i}&+0.084 \text { Education }_{i} \
&& (1.473)(0.032)&( (0.029)\
25^{\text {th }}\text { Percentile:} &  Burt \widehat {h W }etght _ { i } =& 6.012+0.073 \text { Income }_{i}&+0.076 \text { Education }_{i}\
&& (2.533)   (0.027)& (0.038) \
50^{\text {th }} \text { Percentile:} &  Burt \widehat {h W }etght _ { i } =& 6.309+0.069 \text { Income }_{i}&+0.091 \text { Education }_{i}\
&& (1.925)(0.022)&(0.042) \
75^{\text {th }} \text { Percentile:} &  Burt \widehat {h W }etght _ { i } =& 6.733+0.057 \text { income }_{i}&+0.074 \text { Education }_{i}\
&& (2.437)(0.026)& (0.029) \
90^{\text {th }} \text { Percentile:} &  Burt \widehat {h W }etght _ { i } =& 7.024+0.033 \text { Income }_{i}&+0.065 \text { Education }_{i}\
&& (1.847)(0.013) & (0.031) \
\end{array}$
 
a)What is quantile regression? Which variable is the quantile of?
b)In what circumstances would quantile regression be preferable to OLS?
c)Discuss the results presented above.

Accepted Answer

The answer of Suppose you are interested in explaining the...

Question 8

In which of the following cases would you want to use quantile regression?

Accepted Answer

A)  When individuals non-randomly select different outcomes of the dependent variable.
B)  When you are attempting to replicate a randomized clinical trial.
C)  When you are dealing with non-negative count data.
D)  When you suspect that the marginal effects are different for different values of the dependent variable.

Question 9

Quasi-experimental methods attempt to

Accepted Answer

A)  control for sample-selection bias.
B)  estimate marginal effects at different points in the distribution of the dependent variable.
C)  account for endogeneity of an independent variable.
D)  replicate randomized clinical trials.

Question 10

What is sample-selection bias? Why does it present a problem for OLS? How can you control for its presence? Explain.

Accepted Answer

The answer of What is sample-selection bias? Why does it...

Question 11

The first-stage in the Heckman selection correction is estimating

Accepted Answer

A)  the individual's self-selection decision and using those estimates to calculate predicted values of the self-selection decision.
B)  estimating the regression model and calculating the residuals.
C)  the individual's self-selection decision and using those estimates to calculate inverse Mill's ratios.
D)  estimating the regression model and calculating the residuals and using those estimates to calculate predicted values of the dependent variable.

Question 12

In which of the following cases would you want to use a Heckman selection correction model?

Accepted Answer

A)  When individuals non-randomly select different outcomes of the dependent variable.
B)  When you are attempting to replicate a randomized clinical trial.
C)  When you are dealing with non-negative count data.
D)  When you suspect that the marginal effects are different for different values of the dependent variable.

Question 13

You can control for sample-selection bias by performing

Accepted Answer

A)  two-stage least squares.
B)  weighted least squares.
C)  a Heckman selection correction.
D)  difference-in-difference estimation.

Question 14

Sample-selection bias occurs when

Accepted Answer

A)  the researcher selects a bad sample.
B)  the sample contains an independent variable that is correlated with the error term.
C)  individuals randomly select the sample to which they belong.
D)  individuals non-randomly select themselves into a given outcome of the dependent variable.

Question 15

What is quantile regression? When might it be preferred to OLS? Explain.

Accepted Answer

The answer of What is quantile regression? When might it...

Question 16

When performing difference-in-difference estimation,the treatment group is the group

Accepted Answer

A)  for which the policy shock occurred.
B)  for which the policy shock did not occur.
C)  of observation in the "before" sample.
D)  of observation in the "after" sample.

Question 17

Suppose you are interested in explaining the number of surfers surfing at your favorite spot on a given day.After collecting data on the number of surfers,the height of the waves (in feet),the water temperature,and whether the day was a weekend on a sample of 92 days,you estimate the following marginal effects for the Poisson model $$\begin{array} { lll l} 
\text { Surfers } _ { i } = & 13.653 + 6.029 \text { Height } _ { i } +& 0.596 \text { Temperature } _ { i } +& 4.076 \text { Weekend } _ { i } \
&( 3.824 ) ( 2.137 ) & ( 0.202 ) & ( 1.382 )
\end{array}$$ 
a)Why is OLS not appropriate in this circumstance? How does a Poisson model improve on OLS?
b)Discuss the results.
c)What assumption is necessary for Poisson to be the appropriate model? How would you test this assumption? If this assumption fails,what alternative model could you estimate?

Accepted Answer

The answer of Suppose you are interested in explaining the...

Question 18

Sample-selection bias presents a problem because it

Accepted Answer

A)  results in OLS coefficient estimates that are biased and inconsistent.
B)  does not account for the correlation between the independent variable and the error term.
C)  does not account for the time-invariant component of the error term.
D)  does not account for the autoregressive structure of the error term.

Question 19

In which of the following cases would you want to estimate a Poisson model?

Accepted Answer

A)  When individuals non-randomly select different outcomes of the dependent variable.
B)  When you are attempting to replicate a randomized clinical trial.
C)  When you are dealing with non-negative count data.
D)  When you suspect that the marginal effects are different for different values of the dependent variable.

Question 20

In which of the following cases would you want to estimate a Negative Binomial model?

Accepted Answer

A)  When individuals non-randomly select different outcomes of the dependent variable.
B)  When you are attempting to replicate a randomized clinical trial.
C)  When you are dealing with non-negative count data.
D)  When you suspect that the marginal effects are different for different values of the dependent variable.

Exam 15: Quantile Regression, Count Data, Sample Selection Bias, and Quasi-Experimental Methods

Quantile regression

What is a difference-in-difference estimator? When is it appropriate to use one? How do you do so? Explain.

When performing difference-in-difference estimation,the control group is the group

Suppose you wish to determine factors affecting the number of surfers observed surfing at a given surf spot,an appropriate model to estimate the model would be

Suppose you wish to explain the number of nights per week that individuals eat dinner at a restaurant,an appropriate model to estimate would be

In which of the following cases would you want to use quantile regression?

Quasi-experimental methods attempt to

What is sample-selection bias? Why does it present a problem for OLS? How can you control for its presence? Explain.

The first-stage in the Heckman selection correction is estimating

In which of the following cases would you want to use a Heckman selection correction model?

You can control for sample-selection bias by performing

Sample-selection bias occurs when

What is quantile regression? When might it be preferred to OLS? Explain.

When performing difference-in-difference estimation,the treatment group is the group

Sample-selection bias presents a problem because it

In which of the following cases would you want to estimate a Poisson model?

In which of the following cases would you want to estimate a Negative Binomial model?