Conceptual questions: Who is well?

1. HoosWell is a real voluntary program at UVA aimed at improving the wellness of their employees. This is a costly program, so understanding whether it works or not is important since resources could be allocated somewhere else. The Hoo’s Well team will measure success of the program based on two metrics: annual medical expenditures and self-reported health. Annual medical expenditure is the total dollar amount spent on health care. The team can observe this since all Hoos Well participant are part of UVA’s insurance plan. Self reported health is a scale from 0 to 10, where individual’s rank their health from 0 (poor health) to 10 (excellent health). The team has information from before, during and after the program was implemented. The program was implemented in 2017. A goal of the program would be for employees to have less medical expenditures and better self-reported health.

1. The first empirical exercise the team makes is a simple comparison of health outcomes before and after the program for the people enrolled in Hoo’s well, what would we conclude about the effectiveness of the program? Write down the non-technical conclusion, instead of interpreting estimates.
‣
Answer
2. Imagine the program does have beneficial effects (lower medical expenditures and better self-reported health), what is a hypothesis that can explain the findings in 1a?
‣
Answer
3. The second empirical exercise the team does is to run the following regressions:
$$AnnualMedicalExpenditures_i=\alpha_0+\beta_1HoosWell_i+\epsilon_i \\ SelfReportedHealth=\gamma_0+\delta_1HoosWell_i+\nu_i$$

Where Hoo’s Well is a binary variable indicating if an individual is in the program or not. This regression is only using information from after the program’s conclusion. What is the value of $\alpha_0,\ \beta_1,\ \gamma_0,\ \delta_1$? Show your work for full credit.

‣
Answer
4. Use your results in the previous question 1c and the information from table 1 (only using columns from after the program) to assess the magnitudes of the effect of Hoo’s well on both primary outcomes.
‣
Answer
2. Given your results from 1c, what is the conclusion on the effectiveness of the program? Write down the non-technical conclusion, instead of interpreting estimates.
‣
Answer
3. The team decide to further investigate the effectiveness of the program and runs 3 different models using annual medical expenditures, and one model to understand Hoo’s well participation. The results are in Table 2.

1. Using the results from model (2), What is the average annual expenditures for a 30 year old married white women who is not in the program? Show your work.
‣
Answer
2. Using the results from model (2), What is the marginal effect of being married on annual medical expenditures? Show your work, and provide a complete interpretation with numbers (Hint: think about all potential cases).
‣
Answer
3. Employee’s age at UVA range from 18 to 76. Using the results from model (3) provide a full interpretation of the effect of age on annual medical expenditures. Show your work.
‣
Answer
4. Comparing the model from column 1 vs column 2 vs column 3: Which estimate on the effect of the Hoo’s well program would you trust more? Provide a justification of why your preferred model is better is better than the other two, separately? (i.e. Say you prefer model 1 overall, then your answer should be structured as: Model 1 is better than 2 because ...., and Model 1 is better than 3 because...)
‣
Answer
5. Among smokers at baseline, about 31 percent enrolled in Hoo’s well, and among nonsmokers 35 percent enrolled in Hoo’s well. Since the models above do not include a covariate for smoking practices before entering the program, we would want to know if by including a covariate related to smoking the effect of the program would be more positive or more negative. Assuming that being a smoker increases annual medical expenditures, would the estimated effect of the Hoo’s well program be more positive or more negative once controlling for smoking status? Show your work.
‣
Answer
4. The team have already collected a number of important variables but they have money to collect one more variable. There are many contenders that your peers are considering. Out of the variables in consideration, which one would you pick? Explain why you pick that variable over each of the other ones.
• Variables already included in database: Variables in Table 2, and the following 2016 binary variables: academic staff or faculty (binary), heavy drinker, whether the person has a chronic condition, whether the person has high blood pressure.
• Variables in consideration: (1) school in which the individual works for (i.e. Batten, Arts and Science, etc), (2) years going to the gym before 2017, (3) political affiliation.
‣
Answer
5. The year following the first evaluation, they decide give a monetary reward of $100 for people to sign up, but this was only offered to people with even numbers in their UVA IDs (mine is dst2c, so I would get one). The results from this exercise was written in a report to President Ryan: “The study found that the people assigned to monetary reward were 12 percentage points more likely to have enrolled in the program within the analysis period (12 months) than the employees not assigned to the monetary reward. In terms of self reported health they found that the group of employees assigned to monetary reward reported 6.1 points in self reported health and the employees that were not assigned to the reward had 6.0 points in self reported health. The study reported that 2,208 employees join the Hoo’s well program out of the 3,239 eligible employees.”
1. Mention what would be your Z (instrument), D (main explanatory variable/endogenous variable), and Y (outcome). (Define them using the information from the context as oppose to saying general terms like “program”).
‣
Answer
2. Estimate the IV. Show your work, and make sure you are indicating which numbers represents which component.
‣
Answer
3. Finally, write down the interpretation of the IV estimate in this context using plain language.
‣
Answer

Multiple Choice

1. Using Table 2, What’s the average change in "Hoo’s well" participation corresponding to an increase in employee annual salary of $10,000?
1. 0.29 percent decrease in Hoo’s well participation
2. 2.9 percentage points increase in Hoo’s well participation
3. 0.0029 percent decrease in Hoo’s well participation
4. 2.9 percent decrease in Hoo’s well participation
5. none of the above
‣
Answer
2. If the exogeneity assumption is true $(Corr(D, ε) = 0)$, then we know that the OLS estimate will have statistically significant effects:
1. True
2. False
‣
Answer
3. Including more variables that are correlated with your outcome will always reduce bias on the main explanatory variable no matter what.
1. True
2. False
‣
Answer
4. If X is a random variable and c is a constant, then:
1. $Cov(X,C)\neq0$
2. $Corr(X,c)\neq0$
3. (a) and (b) are both true
4. $Var(X)\times Var(c)=0$
5. None of the above
‣
Answer
5. The sample sizes in Table 1 are different among treatment and control group. This is a concern in terms of the causal interpretation of the effect of the program
1. True
2. False
‣
Answer
6. In the main regression $Y_i = \alpha_0+\delta D_i+\epsilon_i$, a control variable we would want to pull out of the error term and should be correlated with the outcome, correlated with D and not a potential outcome. These rules imply that a covariate is also a good candidate for an instrument.
‣
Answer
7. The monotonicity assumption is not required for an unbiased estimate of IV
1. True
2. False
‣
Answer
8. Since people need internet access to sign up and to use Hoo’s well, a member of the team suggest using access to broadband internet at home as an instrument for participating in Hoo’s well. First, pick which positions you will take (defending or arguing against the IV). Then pick the statements that would support your argument (at most, pick 3 statements). If you pick arguing for the IV, you would want it to interpreted it as LATE. If you are arguing against it, you will have to pick all statements that show arguments against it. In your answer sheet, circle your position and write the letters corresponding to each statement you choose.
1. Arguing in favor of IV
2. Arguing against the IV

Statements:

3. Internet affect the likelihood of enrolling in Hoo’s well.
4. The Hoo’s well program is for UVA employees, and all employees would have access to the internet through their work, therefore it may not matter if they have internet access at home.
5. Having good internet at home is great for streaming movies.
6. When comparing people who do have access to people who don’t, we may just be comparing people in rural areas vs. people in non-rural areas. We also know that urbanicity could be correlated with health outcomes through food deserts, availability of primary care, educational outcomes, etc.
7. Who has access to faster vs. slower access to internet is random
8. We don’t think having access to internet makes people less likely to join Hoo’s well.
9. UVA employee’s have an average tenure at UVA of about 15 years.
‣
Answer