联系我们： 手动添加方式: 微信>添加朋友>企业微信联系人>13262280223 或者 QQ: 1483266981

PADM 5310 Individual Empirical Project – Fall 2022 In this project, you will come up with your own substantive research question and try to answer it with real data and the econometric methods you learn in this class. Your question needs to be feasible, and just as important, it needs to be a question you care about. Group Size You can work individually (recommended) or in teams of 2. Time frame This project has 3 parts. The due dates for each written submission are: Research Question & Annotated Bibliography write-up Friday, Oct 14th Data Description write-up Monday, Nov 11 Final report write-up deadline Monday, Dec 12, 4:30pm (The final report due date might be updated later, based on guidance from Cornell’s Registrar’s office) Overall Project Grade Your project grade will be computed using the following: 12.5% Research question write-up 12.5% Data description write-up 75% Final project write-up Types of Research Questions – Examples The goal of this project is to get you to use the tools we develop in class, in an application that you care about. The core tool to be used is multivariate regression analysis. This tool is used in many contexts; here are a few common categories (and there are many others!): 1. Estimating causal effects of a policy. If we increase the EITC, what impacts will that have on labor supply If the minimum wage goes up, what impact will that have on wages and employment in the restaurant sector When states adopted the Medicaid expansion as part of Obamacare, what impact did that have on insurance rates These are all questions of causal impacts of policy changes. There are several strategies to estimate these impacts, with one leading strategy that is well suited for our toolkit being: a. Difference-in-Difference analysis. This is one strategy that can be used to estimate causal effects of a policy. To do so, you need to identify a “before” and “after” period (corresponding to when the policy was implemented, or significantly changed), and a “treated” and “control” group. (These groups could be different geographic areas, or could be different demographic groups.) We will discuss this strategy more in class. 2. Documenting and understanding demographic disparities. There are a lot of unequal outcomes in economic and social life, and our tools can be used to better understand these. For example, in the US women’s wages are on average about 80% of their male counterparts’ wages. Our regression tool can be used to examine questions about whether this is driven by differences in age, education, occupation, parenthood, etc. 3. Predicting an outcome or outcomes. Sometimes we are most interested in making a prediction. For example, when a judge decides whether to release a defendant before trial, she wants to predict the probability that the defendant will show up for the trial and not get into trouble beforehand. When a tax auditor is trying to decide whether to audit a filing, they need to predict the expected amount of tax fraud they will uncover. a. What variables impact our predictions of expected outcomes This is a variant on the main question, where we are interested not only in predicting an outcome, but in understanding how the predictions change as we compare different values of explanatory variables. E.g., what is the relationship between firm size and expected amount of tax fraud Data OPTION 1: I recommend using data from the IPUMS project: (https://www.ipums.org/). This is an amazing collection of micro datasets from both current and historical periods for the US and across the world. If you use one of the IPUMS datasets, you will need to build in some time to get an account approved, and allow for a round or two of learning what data you want to download and clean. This source of data will require a bit more work than the choices below, but is likely to provide the greatest learning opportunity. OPTION 2: You can also use one of these rich data sets that are available on the course web site along with documentation: ● The Consumer Expenditure Survey (CE) is collected every quarter by the US Bureau of Labor Statistics from a large nationally representative sample of American households. It includes numerous measures of income, purchases, household characteristics, and individual characteristics. This data is used to by the government to compute poverty thresholds for social programs and revise the Consumer Price Index (CPI) market basket of goods and services. Businesses and academic researchers use the data to learn about consumer spending habits and trends. I’ve posted a large subset of the CE data collected in the first quarter of 2013. ● The General Social Survey (GSS) has been collecting data on American knowledge, opinions, and behavior every two years since 1972. It asks many of the same questions every round giving us our best source of data for examining trends in these factors. It also asks new questions on timely issues every round. I’ve posted the data and documentation for the 2012 wave. OPTION 3: With my pre-approval, you may also use data you find yourself, though this is often a more difficult proposition. If you want to go this route, you need to obtain my written approval by Friday, October 7th. (To do so, you will need to reach out to me in advance of that date with your idea). You will not be allowed to conduct any experiment or survey that would require approval from a Human Subjects Review Board. Part 1: The Research Question & Annotated Bibliography You should write about 400-600 words discussing the question you hope to answer. Why is this question interesting What do you think the answer might be and why I encourage you to look at the documentation for the data sets listed above for ideas. Here are a few examples of feasible questions: ● How do characteristics of owners influence the revenue firms make ● How does formal education affect people’s knowledge of current events ● What kinds of families have multiple cars ● Does smoking affect grades in high school ● What characteristics of countries determine how many gold medals they win at the Olympics ● How do different NBA player performance statistics (e.g., points or rebounds per game) influence salaries There are four issues you may have to address in your research question: 1. Population of interest: You might want to focus your study on only certain groups, geographic areas, etc. This may be only a subset of the population covered by the dataset. You want to be clear about what you population of interest is. 2. Causality vs. correlation: Are you trying to estimate how changes in one thing (call this an independent variable) affect another thing (call that a dependent variable) This would be a statement about causation, and as many of you have already heard, correlation does not necessarily imply causation. It is much easier to describe how two things might be correlated, but it’s much more interesting and useful to estimate causal effects. 3. Time ordering: In general, if you are estimating a causal effect, your determinants should be established before the measures of what they might be affecting. If you are looking at the effects of formal education on knowledge of current events, you should make sure formal education has been completed before you measure knowledge of current events. On the other hand, it doesn’t make sense to look at how college education affects whether people smoke since most people who smoke start before they finish high school, and few people quit smoking in college. 4. Confounders: Are there variables that might influence both your independent and dependent variables and thus induce a correlation e.g., People with high IQs might be more likely to get more formal education and also keep up better with current events. People from poor families might be more likely to smoke and also receive poor grades in school. You will need to account for these potential confounding variables in your analysis. The question you choose is not set in stone–You will be able to modify or change your research question later on if you need to. Also, you should write up 2-3 pages (750-1500 words) discussing related literature. This can be papers whose findings are related to your research question, or which motivate or speak to your question. In your literature review it is very important that you connect every article you cite explicitly to your empirical project—if you cannot articulate concisely the relevance of an article to your research question, you should not be citing that article in your literature review. The grading rubric for the Research question write-up will be based on the following: Rubric for Research Question & Annotated Bibliography Clearly stated research question (20 pts) Policy or social science importance of question (10pts) Feasible to answer (20pts) Relevant literature (35pts) Appropriate consideration of: Population of interest, Causation vs. correlation, etc (15pts) Final Grade Part 2: Data Description Create your analysis sample from your raw data. What are the relevant variables There should be 10-15 variables incorporated into your paper. How many observations do you have of each You should have at least 100 observations and ideally quite a few more. Describe each variable’s distribution using a tabulation if it is categorical or summary statistics and a histogram if it is continuous. Are your continuous variables normally distributed At least one of your variables should be an indicator variable (only takes a value of 0 or 1). Describe the distribution of one of your continuous variables conditional on one of your indicators being zero and being one. Do not show histograms for categorical variables. Talk about the variables themselves rather than the name or number they might have in the survey (e.g., Don’t talk about q55 but instead talk about income or grades or smoking status). Formally test differences in means for some of your variables in two subgroups. e.g., Are families with children more likely to own more cars than families without children Do countries in colder regions of the world win more gold medals per capita at the Winter Olympics than countries in warmer regions The results of these tests should give some insight into your research question and motivate the regression analysis you perform in part 3. Your report should be primarily English sentences with supporting tables and figures and should be no more than 2500-3500 words, plus Tables and Figures. It should include the title, research question, motivation, and literature review that you developed earlier, and you should incorporate any new ideas you’ve had since then and any feedback you received. Your tables should look pretty and should not simply be copied and pasted from Stata output. In addition to your report, you should create and hand in a single Stata .do file that performs all of the descriptive analysis. Your grade for this part will be computed using the following rubric: Rubric for Data Description Question (10 pts) Clear (3 pts) Important (4 pts) Feasible (3 pts) Literature Review (5 pts) Sample (15 pts) Clear explanation of source of data (5 pts) Describe unit of observation (5 pts) Number of observations (5 pts) Description of continuous variables (30 pts) (at least 3) Include relevant variables (10 pts) Means and standard deviations (7 pts) Histograms (7 pts) Clear English sentences & write-up (6 pts) Description of categorical variables (25 pts) Include relevant variables (5 pts) Tabulations (aka fractions of sample in each category (10 pts) Clear English sentences & write-up (10 pts) Statistical Tests (15 pts) Appropriate (10 pts) Well explained (5 pts) Final Grade Part 3: Final Project Use an appropriate combination of multiple regression models and (if appropriate) more advanced methods to answer your research question. You should estimate regression models with one to three different dependent variables and control for potential confounding variables. Carefully interpret the signs and magnitudes of coefficients in your preferred regression specifications. If there are additional potential confounders that you do not have in your data, discuss how omitting them may bias your results. Discuss potential intervening variables if you have any. ● Include the title, research question, motivation, and literature review you developed earlier. ● Include the equation for your preferred regression model specification. ● Include tables of regression results (including coefficients and standard errors) in your final report. ● The maximum length for your report is 15 single spaced pages of text (7500 words), and up to an additional 5 pages of tables and figures. You will likely produce more during your research process, but it is important to know how to identify the most important results. (Your list of references can be in addition to the 15+5 page limit.) In addition to your final report, you should submit a single .do file that performs all of your tests and regression analysis, and if it is not too large upload a copy of your analysis data file (this is the file you will run your analysis on, after cleaning and organizing your data). If your analysis data file is very large (say, N > 30,000), instead upload the output from “describe” and “summarize” results for the dataset. Final Project Grading Rubric Question (10pts) Clear (3 pts) Important (4 pts) Feasible (3 pts) Literature Review (5 pts) Well-connected, sufficient motivation (5 pts) Basic Summary Stats & Descriptive results (10 pts) Appropriate to motivate/give context for research (5 pts) Well executed and described (5 pts) Regression models (50pts) (at least 3) Dependent make sense (5 pts) “Main” independent variables make sense (5 pts) Include potential confounders Hope yes (10 pts) Include big intervening variables Hope no (5 pts) Interpret signs and magnitudes (10 pts) Interpret statistical significance (10 pts) More advanced methods (up to 5 bonus points) Appropriate & Well explained (5 pts) Discussion/Conclusion (25 pts) Main take away (10 pts) Limitations of approach (10 pts) Policy relevance (5 pts) Final Grade