Note: this course is code-agnostic, but will presume R. If you use any other
coding environment, make sure that it is fully reproducible (for instance, a
Jupyter notebook). The .gitignore file in the repository is specifically for
R. You may wish to modify this file if you are using Python, Julia, or any other
such language.
Please clone your repository as part of this organization (sta440-sp21).
The project proposal will be due Tuesday, February 9. The introduction, data, and EDA components will be due Tuesday, March 2. The methods component and statistical analysis plan will be due Tuesday, March 23. Preliminary results for the project are due Sunday, April 4. All components are due 11:59PM on their respective dates.
You will be randomly paired with two students for peer review on Monday, April 5, for the preliminary results. Peer review and reproduction of your assigned partners' works will be due Friday, April 23.
Additionally, you must present and defend your project in class (April 6th through April 22 meetings). The final report is due Saturday, May 1.
For each of the written components, you must submit a .pdf document to Gradescope that corresponds an .Rmd file (or similar equivalent) on a GitHub repository in order to receive credit for your project. If a .pdf is not uploaded to Gradescope by the submission deadline, your latest commit prior to the deadline will be used as your submission.
No prior video presentation is needed -- the presentation will be made live in class.
Your GitHub report repository and commit history will also be evaluated by the instructor. The GitHub repository must contain the reproducible R Markdown document corresponding to the submitted reports, and will be checked throughout the course of the project.
- This is an individual assignment.
- Everything in your repository is for your eyes only except for the instructor or TAs.
- As always, you must cite any code you use as inspiration. A failure to cite is plagiarism.
By submitting an assignment, you pledge to uphold the Duke Community Standard:
- I will not lie, cheat, or steal in my academic endeavors;
- I will conduct myself honorably in all my endeavors; and
- I will act if the Standard is compromised.
The data used for your project should be uploaded to either the GitHub repository or the Duke Computing cluster (depending on the size of the dataset). These data must be shareable, as others will be reproducing your work.
Students may provide their own data, with the understanding that these data should not have been analyzed previously as part of a course assignment or internship (if you would like to take a prior analysis to the next level, please provide the full prior analysis write-up with your proposal for evaluation). If you are having trouble finding data, please contact the instructor as soon as possible.
There are two main purpoess of the project proposal:
- To help you think about the project early, so you can get a head start on finding data, reading relevant literature, thinking about the questions you wish to answer, etc.
- To ensure that the data you wish to analyze, methods you plan to use, and the scope of your analysis are feasible and will help you be successful for this project.
In writing your proposal, include an introduction (brief context, the motivation for your research question, any relevant literature citations, and the general research questions and hypotheses) and a description of the data that you will be using (including any data use agreements and where the data came from).
Note that your proposal does not have to be your final analysis! It is just meant to get you started. It's possible that the scope of your project will change throughout this semester.
The project proposal will not be factored into your final project grade. However, if it is not completed by the due date, your final project will receive a penalty of 5 points per day (after the grace period ends).
The proposal can be no more than 3 pages at 11 point font or larger (figures and tables not included). Any text beyond 3 pages will not be considered.
The in-class presentation and final write-up for this project is worth 100 points total, broken down as follows. You will have the opportunity to make up points on this component.
| Component | Points |
|---|---|
| Introduction | 10 |
| Methodology | 30 |
| Results | 20 |
| Discussion | 15 |
| Appendix | 10 |
| Repository | 5 |
| Presentation | 10 |
Note: Submissions missing code used for the manuscript submission in the GitHub repository will automatically receive a penalty of 50 points.
The write-up must be no longer than 12 pages at 11 point font or larger (figures and tables not included). Any text beyond 12 pages will not be considered. Note that you may include technical derivations, assumptions, etc. in a supplemental appendix that does not factor into this page limit.
The introduction should introduce your general research question and your data (where it came from, how it was collected, what the observational unit is, which variables were used in the analysis, etc. Feel free to create subsections as needed (for instance, for the dataset, exploratory visualizations, etc.). In writing your introduction, make sure you address the following points:
- Is/are the main goal(s) of the analysis easy to identify and appropriate for addressing the overall research problem?
- Is the rationale for the data analysis explained well?
- Does the manuscript describe the context of the work and its relation to existing literature?
- Are the variables (response and predictors) clearly identified and discussed?
- Does the manuscript explain how the data were collected and/or how they were derived?
- Is any EDA helpful and informative in addressing the main project goal(s)?
The methodology section should clearly explain the model(s) used in your analysis. You must clearly state your model formulation using appropriate mathematical notation and justify their use, and address any model assumptions or diagnostics needed. In writing the methodology, make sure you address the following points:
- Is the proposed analysis appropriate given the main goal(s) and dataset?
- Why was this particular methodology chosen over competing choices?
- Are the specific methods described in enough detail that the work could be replicated by other researchers without access to the original analysis code?
- Is it clear which approaches/models were used to evaluate specific goals?
- What assumptions are needed for the model(s), and how do you plan to assess whether they hold?
- What sensitivity analyses, if any, are planned, and how do they relate to your analysis approach?
You may include technical derivations and evaluation of model diagnostics and assumptions in the appendix to your manuscript; they do not belong in the body of your work.
Showcase your results! Provide model output (such as coefficient estimates and quantification of uncertainty) in tabular and/or visual format. Make sure that each set of results presented supports the goal(s) of your manuscript.
Discuss the implications your results have in terms of your goal(s) and research question(s). As well, provide a reasoned critique of your methodology and provide suggestions for improving the analysis or what additional data might have strengthened the paper. In doing so, make sure you address the following points:
- How do your results address or fail to address the goal(s) of your manuscript?
- Does the manuscript provide clear, correct, and effective interpretation of the analysis results?
- Are all conclusions made directly supported by the results?
- Was your methodology fully appropriate? What alternative techniques might have been useful?
- Are there any issues with reliability or validity of the data?
- What would you do differently if you had to approach the study again? What additional data sources, variables, or techniques might help you create a stronger manuscript?
This is where you present any technical derivations (if needed) and demonstrate that the assumptions for your models are met. Examples of derivations might include explicit variance terms of frequentist estimators or derivation of full conditional distributions for Gibbs samplers, etc.
As for examples of assumptions, if you are creating a linear model, this would be a good place to discuss the assumptions (e.g., by providing residual plots and comments); if you are performing a Bayesian analysis, this would be a good place to show diagnostic plots (e.g., trace plots, etc.). This section may be as long or as short as needed.
In this section, I am evaluating whether the commit history is appropriate and contains clear descriptions of each committed change. This repository must contain all code used, as well as any ancillary external files which you may have used in your analysis.
The final .pdf report must be typeset and reproducible from your analysis code (end-to-end scripting). This .pdf must match your submission to Gradescope. As well, any slides used for your vidoe presentation should also be included. If these slides were generated using a reproducible method (such as xaringan or Beamer), you must include the code used as well.
Note: Submissions missing code used for the manuscript submission in the GitHub repository will automatically receive a penalty of 50 points.
You must make a live video presentation of 15-20 minutes in class (between April 6 and April 22, inclusive). You may use as many slides or visualizations as you wish. This presentation should include all components, including the preliminary results and discussion. The instructors, TAs, and your peers will be asking questions of your project -- you must defend your project verbally in response to these questions. Treat this as additional peer review (the video presentation is meant to mimic a formal paper presentation at a conference).
- Clearly state your hypothesis (or hypotheses) - think about how your paper might create actionable insight for others.
- Make sure you use best visualization practices as discussed in class for all visualizations and/or tables.
- Write clearly and effectively; confusing the reader is never a good thing!
- Make sure the paper is professionally presented and free of distracting errors or other issues such as poor organizsation, problems with grammar, spelling, or punctuation, and layout concerns such as small font in visuals, excessive and inappropriate decimal points, etc. (this is not an exhaustive list!).
- Quality over quantity - do not calculate every statistic and procedure you have learned for every variable, but rather choose the most appropriate technique or set of techniques to address the goal at hand.
- Focus on methods that help you begin to answer your research questions.
- For the presentation, keep things simple and streamlined. Focus on the general message you want to get across; simply copy/pasting from your paper is unlikely to receive a good grade!