1. Conduct the following analysis in a well-documented R markdown file. You should include text in the document discussing what and why you are doing the analysis you are doing so that someone unfamiliar with your project could follow along. It may make sense for you to add to and edit your existing R markdown file that you used for analysis.

  2. If you have not already:

    • In README.md, include a brief description of what is in your repository and instructions of how to repeat the analysis.
    • Separate the code that generates your the dataset you will primarily use for your analysis from the code you will use in your analysis.

  3. When you have finished, create an issue in Github titled: “Review Project Assignment 3”. In the issue message write “cc @jrnold @CasAndreu”. This is how you submit your assignment, and lets the instructors know it is ready to review.

These are questions that you should try to answer in order of importance. Do as much as you can in the time you have. We can then evaluate where you are and there are still time to wok on

  1. For your variable of interest:

    • Run a bivariate regression
    • Plot the results of the bivariate regression, and interpret.

  2. Interactions

    • Does your theory require interactions?
    • If so, include those and interpret them using the methods discussed in Golder et. al.

  3. Omitted variable bias

    • What potential omitted variables can you think of?
    • Are you controlling for any variables that are mechanisms by which your variable of interest may be affecting the outcome?
    • Run a model with those controls. How much does this affect the
    • Evaluate the possible degree of omitted variable bias using the proportional selection of observables and unobservables assumption (e.g. the test used in Nunn and Wantchekon)
    • If there are any particular variables that you can think of but cannot control for, can you reason about how they would affect the coefficients of the variable of interest.

  4. Interpret the substantive effects

    • Calculate a first difference for a reasonable change in your variable of interest and its standard error.
    • What is the average marginal effect of your effect and its standard error?

  5. Measurement error

    • Are you concerned about measurement error in any of your variables?
    • How would that measurement error affect your results?

  6. Heteroskedasticity

    • Check for heteroskedasticity using a plot of studentized residuals vs. fitted values
    • Conduct a test for Breusch-Pagan test for heteroskedasticity. Does it suggest that there is heteroskedasticity?
    • Do you have time-series data? Calculate heteroskedasticity and autocorrelation robust standard errors.
    • Do you have groups in data? What are they? Include cluster-robust standard errors.
    • Estimate your model using bootstrapped standard errors: be sure to use the correct data generating process.

  7. Outliers

    • Check for influential observations using Cook’s distance and visual methods.
    • If there are outliers, can you explain them?
    • Ensure that your results do not rely on an outlier. If your results depend on dropping observations, how much would they change if you included those observations? Justify the omission.

  8. Model comparison

    • Compare the MSE, \(R^2\), and adjusted \(R^2\) of your models. Does it make sense to use this statistic for model comparison?
    • Compare your models using cross-validated MSE.
    • Are there other criteria that you should use to compare the model that are particular to your research question?

  9. Non-normal errors

    • Create a QQ-plot of the studentized residuals.
    • Does this plot appear approximately normal?
    • Is the mean a reasonable summary of your outcome variable?
    • Can you respecify your model so that the errors are more normal?