EC902/907: Quantitative Methods: Econometrics A, Fall 2020 Midterm

1.   We are planning an RCT and, given the available information on sample size and the         standard deviation of the outcome variable, we expect the standard error of our               estimate to be around 10. If the true effect was around 20, how likely it is that we obtain an estimate which is statistically different from zero at the 5% level?

.   About 50%

b.   About 5%

c.   About 95%

d.   It is not possible to answer given the available information.

2.   Imagine that we are planning an RCT. If the true effect was around 0, how likely it is that we obtain an estimate which is statistically different from zero at the 5% level?

a.   About 50%

.   About 5%

c.   About 95%

d.   It depends on the sample size.

3.   In a randomized controlled trial, if the sample size becomes four times larger, standard errors tend to be:

a.   twice as large

b.   50% smaller

c.   four times as large

d.   four times smaller

4.   Randomization inference

We conduct an RCT with 4 participants, and we randomly assign 2 individuals to the     treatment group and 2 to the control group. We observe that the outcome for the two treated individuals is equal to 4 and 2, and the outcome for the two non-treated           individuals is equal to 2 and 0. Using Fisher’s Exact Test, the p-value under the null        hypothesis that the treatment has no effect is equal to:

a.   1/3

b.   1/6

c.   2/6

d.   2/3

5.   Warwick students’ grade point average (GPA) is on average 65, and the standard             deviation is around 5. If we take random samples of 100 students, the sample mean will be 95% of the time:

a.   between 64 and 66

b.   between 63 and 67

c.    between 60 and 70

d.   between 55 and 75

6.   In a recent paper, Bayer and Kuhn (2020) study how the share of working-age families   living with their parents in a given country (“proportion living with parents”) affects the case fatality rate (proportion of deaths from Covid-19 relative to the total number of     people diagnosed, “CFR”) in that country, using country-level information from 24         different countries. When they regress the CFR on the variable “proportion living with   parents”, the estimated beta is equal to 0.05, with st. error equal to 0.06.

If we give face value to this estimate, would you say that the impact of the “proportion living with parents” on CFR is:

a.   statistically significant at standard levels and economically large

b.   not statistically significant at standard levels but potentially economically large

c.    not statistically significant at standard levels nor economically relevant

d.   statistically significant at standard levels but economically small

7.   In a recent paper, Bayer and Kuhn (2020) study how the share of working-age families   living with their parents in a given country (“proportion living with parents”) affects the case fatality rate (proportion of deaths from Covid-19 relative to the total number of     people diagnosed, “CFR”) in that country, using country-level information from 24         different countries. When they regress the CFR on the variable “proportion living with   parents”, the estimated beta is equal to 0.05, with st. error equal to 0.06.

How should we interpret this estimate?

a.   It captures the causal impact of the “proportion living with parents” on the CFR

b.   It  provides  an  upper  bound  of  the  causal  impact  of  the  “proportion  living  with parents” on the CFR, if we assume that countries with a  higher “proportion  living with parents” tend to be better in some dimensions that tend to decrease the CFR

c.    It  provides  a  lower  bound  of  the  causal  impact  of  the  “proportion  living  with parents” on the CFR, if we assume that countries with a  higher “proportion  living with parents” tend to be better in some dimensions that tend to decrease the CFR

d.   It  provides  a  lower  bound  of  the  causal  impact  of  the  “proportion  living  with parents” on the CFR, if we assume that countries with a  higher “proportion  living with parents” tend to be worse in some dimensions that tend to decrease the CFR

8.   In a recent paper, Bayer and Kuhn (2020) study how the share of working-age families   living with their parents in a given country (“proportion living with parents”) affects the case fatality rate (proportion of deaths from Covid-19 relative to the total number of     people diagnosed, “CFR”) in that country, using country-level information from 24 different countries. In this analysis, would it be a good idea to control for the number of doctors per capita in the country?

a.   Bad control, because it is correlated with the “proportion living with parents”

b.   Good control, because it is predetermined and it might affect the CFR

c.    Good control, because it is predetermined and it cannot affect the CFR

d.   Bad control, because it is not correlated with the CFR

9.   In a recent paper, Bayer and Kuhn (2020) study how the share of working-age families   living with their parents in a given country (“proportion living with parents”) affects the case fatality rate (proportion of deaths from Covid-19 relative to the total number of     people diagnosed, “CFR”) in that country, using country-level information from 24          different countries.

Imagine that they had access to an additional variable that measures the “frequency with which individuals meet with their parents” . If they want to estimate the causal  impact of “proportion living with parents”, would it be a good idea to include this      additional variable as a control.

a.   Bad control, because it might be affected by the “proportion living with parents”

b.   Good control, because it is correlated with the “proportion living with parents” and it might affect the CFR

c.    Good control, because it is not correlated with the “proportion living with parents” and it might affect the CFR

d.   Bad control, because it is not correlated with the “proportion living with parents”

10. In a recent paper, Bayer and Kuhn (2020) study how the share of working-age families   living with their parents in a given country (“proportion living with parents”) affects the case fatality rate (proportion of deaths from Covid-19 relative to the total number of     people diagnosed, “CFR”) in that country, using country-level information from 24         different countries.

The CFR might be subject to measurement error, given that many cases of Covid-19 are  not diagnosed. Let us assume that this measurement error was random. How would this affect their estimate?

a.   The estimate will suffer an attenuation bias

b.   The estimate will be still consistent but less precise

c.   The estimate will be subject to an upwards bias

d.   The estimate will be a lower bound of the true effect

11. In a recent paper, Bayer and Kuhn (2020) study how the share of working-age families   living with their parents in a given country (“proportion living with parents”) affects the case fatality rate (proportion of deaths from Covid-19 relative to the total number of     people diagnosed, “CFR”) in that country, using country-level information from 24         different countries.

The variable “proportion living with parents” might be subject to measurement error. Let us assume that this measurement error was random. How would this affect their  estimate?

a.   The estimate will suffer an attenuation bias

b.   The estimate will be still consistent but less precise

c.   The estimate will be subject to an upwards bias

d.   The estimate will be a lower bound of the true effect

12. Which of the following statements is false:

In a randomized control trial, including additional controls in the regression...

a.   tends to increase the R-square

b.   is not expected to affect statistically the point estimate

c.    may help to increase the precision of the estimation

d.   does not affect the standard errors of the main estimate

13. In a randomized random trial, how should we assign people to the treatment and the control group in order to maximize precision?

a.   same number of people in the treatment and in the control group

b.   more people in the treatment group than in the control group

c.    more people in the control group than in the treatment group

d.   the relevant thing is the total sample size, not how they are distributed across the treatment and control groups.

14. Which of the following statements is false:

a.   A potential weakness of RCTs is that they may fail to capture general equilibrium effects

b.   RCTs tend to have high internal validity

c.    In RCTs the consistency of the estimate requires that the SUTVA assumption is satisfied

d.   RCTs tend to have large external validity

15. Let us consider the following (unusual) estimator. alpha= Y_1 - Y_0  + λ/N

where λ is a random variable uniformly distributed between -1 and +1 and N is the sample size

Which of the following statements is true. This estimator is:

a.   unbiased but inconsistent

b.   biased but consistent