Question

General Requirements: 1. Choose a Y (outcome measure) that someone cares about (at your firm using real data, preferably). It can be a good thing like profit, time spent on project (efficiency measure), or a worthwhile project. If you can find a Y of interest, your whole project will be of interest. 2. Once you choose a CONTINUOUS Y of interest, choose a policy oriented X. Your project's implications revolve around b, (the relationship between X and Y). So if Y=sales and X=advertising the beta gives interesting information (i.e. is advertising effective?). If X=months then that is not as interesting ... (i.e. each month sales increase by 10%...that's nice but does not tell the manager how to improve sales). So choose an X that implies ACTION. Not all of your Xs have to imply that ACTION should or shouldn't be taken, but at least one. 3. Don't use a Macroeconomic Y like Dow Jones or Unemployment in U.S. or Per Capita income. Someone else has already done studies much more in depth than you will here. Choose a Y that YOU know about... like something from your work, hobbies, or home. 4. The most common problem that students encounter designing their project is the Y variable they want to use is not continuous. You must have a continuous Y, not a series of 1 through 4, or yes/no in order to use OLS regression. Y must be able to take on a multitude of values. 5. The next most common problem is that students don't match the Y and X values. For example, Y is per store, or per day. YOUR X's MUST ALSO BE per store or per day. 6. You must include at least 4 separate X variables in the regression and more if you have them (note: this cannot be one categorical variable with 5 dummy variable categories). 7. You must have at least one categorical X variable. A variable such as marital status that has 2 categories still just counts as one X variable for the calculation of #6 above. 8. The number of observations must be greater than or equal to 50. 9. Working in groups is required: you may have at most 5 group members. 10. Use a significance level of 10%. 11. The paper should be written as ONE continuous paper with NO SECTIONS. 12. Appendix should be clearly labeled and begin on a new page.

Name: general requirements 1 choose a y outcome measure that someone cares about at your firm using real data preferably it can be a good thing like profit time spent on project efficiency measure 77214
Uploaded: 2023-03-31T16:19:19-08:00
Duration: 3 min 13 s
Channel: Sri K
Description: general requirements 1 choose a y outcome measure that someone cares about at your firm using real data preferably it can be a good thing like profit time spent on project efficiency measure 77214

General Requirements:
1. Choose a Y (outcome measure) that someone cares about (at your firm using real data, preferably). It can be a good thing like profit, time spent on project (efficiency measure), or a worthwhile project. If you can find a Y of interest, your whole project will be of interest.
2. Once you choose a CONTINUOUS Y of interest, choose a policy oriented X. Your project's implications revolve around b, (the relationship between X and Y). So if Y=sales and X=advertising the beta gives interesting information (i.e. is advertising effective?). If X=months then that is not as interesting ... (i.e. each month sales increase by 10%...that's nice but does not tell the manager how to improve sales). So choose an X that implies ACTION. Not all of your Xs have to imply that ACTION should or shouldn't be taken, but at least one.
3. Don't use a Macroeconomic Y like Dow Jones or Unemployment in U.S. or Per Capita income. Someone else has already done studies much more in depth than you will here. Choose a Y that YOU know about... like something from your work, hobbies, or home.
4. The most common problem that students encounter designing their project is the Y variable they want to use is not continuous. You must have a continuous Y, not a series of 1 through 4, or yes/no in order to use OLS regression. Y must be able to take on a multitude of values.
5. The next most common problem is that students don't match the Y and X values. For example, Y is per store, or per day. YOUR X's MUST ALSO BE per store or per day.
6. You must include at least 4 separate X variables in the regression and more if you have them (note: this cannot be one categorical variable with 5 dummy variable categories).
7. You must have at least one categorical X variable. A variable such as marital status that has 2 categories still just counts as one X variable for the calculation of #6 above.
8. The number of observations must be greater than or equal to 50.
9. Working in groups is required: you may have at most 5 group members.
10. Use a significance level of 10%.
11. The paper should be written as ONE continuous paper with NO SECTIONS.
12. Appendix should be clearly labeled and begin on a new page.

Added by Ruben J.

1 Fixed Effects Suppose (once again) you are interested in estimating how health affects productivity. Specifically, you are interested in the following regression: Yi = β0 + β1Di + ϵi , (1) where Yi is a variable representing the test scores of a student i (a number that goes from 1 to 100), and Di is a dummy variable representing the health of student i. Specifically, suppose that Di=1 for individuals who have a specific disease and Di=0 for individuals who do not have this disease. For this question, let’s use anemia as the disease of interest. If you don’t know what anemia is, it might help to do some research before answering these questions. Suppose you have data from India and Sweden. You use all of your data to run regression 1, and you obtain the following estimates, both statistically significant: β0 = 85, β1 = −20. 1. What is the average test score of students with anemia? 2. What is the average test score of students without anemia? 3. Is there a significant difference between the test scores of anemic and non-anemic students? Another member of your research team points out that India and Sweden are very different in a number of ways. Specifically, India has much lower income per capita, higher anemia rates, and lower quality public schools. In other words, she worries that country-level income might be an omitted variable in the regression above. 4. Do you expect country-level income to be positively correlated, negatively correlated, or not correlated with anemia prevalence? Explain. 5. Do you expect country-level income to be positively correlated, negatively correlated, or not correlated with student test scores? Explain. 6. Using your answers to the previous 2 questions, do you expect the true value to be larger in magnitude (i.e. more negative) or smaller in magnitude (i.e. less negative) than the estimated β1 (which was -20)? Suppose you do not have data on country-level income, and you also worry that there are other country-level characteristics (not just income) that could be serving as omitted variables. You decide instead to run the following regression: Yic = β0 + β1Dic + β2Sc + ϵi , (2) where the c subscript indicates that there are two different countries (c =Sweden and c =India): so Yic represents the test score for student i from country c. The variable Sc is a dummy variable that equals 1 if the student is from Sweden, and equals 0 if the student is not from Sweden (i.e., is from India). 7. How does this regression deal with the endogeneity problem described above? When you estimate this regression, you obtain the following estimates (all statistically significant): β0=70, β1=-10, β2=20. 8. What is the average test score of students with anemia in India? 9. What is the average test score of students without anemia in India? 10. What is the average test score of students with anemia in Sweden? 11. What is the average test score of students without anemia in Sweden? 12. Is there a significant difference between the test scores of anemic and non-anemic students? 13. Is there a significant difference between the test scores of Indian and Swedish students? 14. Compare the estimates for β1 in regression 1 and 2 and provide an explanation for why they are different. 2 Experiments Now suppose that you find out about a randomized controlled trial conducted in India, where a randomly selected group of anemic students received iron supplements for 6 months. Because iron supplementation helps combat anemia, this resulted in lower rates of anemia for those in the treatment group. Specifically, when these students (who were all anemic at the beginning of the experiment) were surveyed after 6 months, 87% of the control group was still anemic, while only 56% of the treatment group was anemic. First, think about the relationship between anemia (as your dependent variable) and being assigned to the iron supplement group (as your independent variable). 1. Are there any reverse causality issues in this situation? (In other words, should we be worried about anemia driving the group assignment?) Why or why not? Hint: don’t overthink this question. 2. Are there any other endogeneity concerns in this situation? Why or why not? 3. Based on your answers to the previous 2 questions, is it correct to say that iron supplementation causally reduced anemia by 31 percentage points? 3 Instrumental Variables You decide to use the RCT described above to answer your original question about whether anemia causally affects school performance. In other words, you decide to use the experiment as an instrument for anemia. You are still interested in regression 1, but now you plan to estimate it in two steps. In the first stage, you will estimate Di = α0 + α1Ti + ui , (3) where Ti is a dummy variable equal to 1 if the student was randomly assigned to the iron supplementation treatment, and equal to 0 if the student was randomly assigned to the control group. In the second stage, you will estimate Yi = β0 + β1Dˆ i + ϵi , (4) where Dˆ i is the predicted Di estimated from the first stage. If your instrument is valid, β1 give you the causal effect of anemia on test scores. 1. For the instrument to be valid, it must satisfy the exogeneity condition. What does that mean in this context? Do you think your instrument (Ti) satisfies this condition? 2. For the instrument to be valid, it must be relevant. What does that mean in this context? Based on the previous description of the experiment, do you think your instrument satisfies this condition? 4 Difference-in-Difference Finally, you decide to explore one more method to answer your research question. You find out about another program in India which was not an experiment. In this program, a non-profit organization picked 10 of the most disadvantaged schools (out of a total of 40 schools in that district), to which they provided iron-fortified flour (for them to use in the school lunches). You collect data on test scores for all 40 schools in 2011, one year after the 10 schools first received the iron-fortified flour. Using this 2011 data, you run the following regression: Yis = β0 + β1Fs + ϵis, (5) where Yis are test scores for student i in school s, and Fs is a dummy variable equal to 1 if the school received the fortified flour. You estimate the following coefficients (both statistically significant): β0=83, β1 = −15. 1. Why do you think β1 is negative? 2. Does β1 represent the causal effect of iron-fortified flour on test scores? Why or why not? You later discover that there also exists test score data from the year 2009 – one year before the selected schools first started receiving the fortified flour. Because of this, you are now able to utilize a difference-in-difference strategy. The difference-in-difference regression looks like this: Yist = β0 + β1Fs + β2At + β3FsAt + ϵist, (6) where the t subscript refers to the year. That is, Yist is the test score for individual i in school s in year t. At is a dummy variable equal to 1 if the year is 2011, and equal to 0 if the year is 2009. 3. What coefficient (or combination of coefficients) gives you the average test score among schools with iron-fortified flour in 2011? 4. What coefficient (or combination of coefficients) gives you the average test score among schools without iron-fortified flour in 2009? 5. What coefficient (or combination of coefficients) captures the time trend (the change in test scores over time that took place for schools without the fortified flour)? 6. What coefficient (or combination o f coefficients) captures the difference between flourrecipient and non-recipient schools in 2009? 7. What coefficient is the difference-in-difference coefficient of interest? 8. What assumption needs to be true if you are going to be able to interpret your difference-in-difference coefficient as the effect of flour fortification on test scores? If your assumptions are valid, you have successfully estimated the causal effect of flour fortification on student test scores. 9. How does this estimate help shed light on your initial question of interest (how health affects productivity)?

Sri K.

Question

Please give Ace some feedback