Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: THEend8_
Final Data Analysis Project (FDAP)
Abstract
Abstract: Should be typed, doublespace, 1 paragraph, maybe a table if useful, and any equations. Describe the data, your tentative model(s), source of prior information and purpose of your analysis. One or a few data plots can be useful.
Please supply the following information about your project:
● Project name. (< 1 sentence).
● Purpose of analysis. (1 sentence).
● Outcome name/definition.
● Predictor variable names/definitions.
● Sampling model.
– Some issues that you may want to explain: Are outcomes correlated? Is there nesting/clustering? Too many predictors? Missing data?
– Only discuss issues relevant to your data set.
● What are the parameters, and what do they mean? Parameter names and descriptions.
● Prior information source. (principal investigator, yourself, literature search)
● Prior densities/model. (For example: product of independent normals for the regression coefficients and inverse gamma for σ2 , etc.)
Type. A short phrase is often sufficient. Be brief!
The purpose of this update is to ensure that you are on track with your project. See me if questions or issues.
I will get back to you if I see a problem/difficulty or if something is unclear.
Data
You will need to identify, find or assemble a data set to analyze early in the quar- ter. Places to search for data or data sets are a) assembling or collecting your own data; b) your current or previous work; c) occasionally your advisor may have a suggestion; d) Your doctoral thesis or master’s paper data; e) consulting client; f) CDC or NCHS data set (https://www.cdc.gov/nchs/data_access/ ftp_data.htm) or census data. Getting a real data set is much preferred. Real means that someone cares about the conclusions of your analysis.
If you get a data set from the web you will need to do additional thinking to decide on the purpose for your analysis and to find a source of prior information. You may not take a data set from the notes, from a previous statistics course like 200A or from Hoff’s, Gelman’s or Congdon’s books, nor from the UCI machine learning repository, nor Kaggle.
The Final Report
Your report must:
● be at most 4 typed double spaced pages,
● be appropriately supplemented with graphs and tables as needed in the appendix. The appendix is not part of the 4 page limit.
● include a sensitivity analysis of the main results to key assumptions.
● include proper informative priors for at least some parameters. The jus- tification for your prior parameters will form the basis for part of the grade.
● report plot(s) of the posteriors of interest.
● include your JAGS code in the appendix – helps me debug sometimes.
● interpret the numerical results in terms of the underlying problem. This is vital for any data analysis.
● label appropriately all tables and graphs
● refer to all tables and graphs in the main text.
The outline of your paper might include many of the following sections:
1. Problem motivation & goal(s) of the analysis.
2. Description of the data set.
3. Description of prior information. What it is and how you translate that to a prior density.
4. Choice of model(s) and prior(s).
5. Mathematical specification of the model, that is, using “twiddle notation”,
for example yi ~ N(µi ,σi(2)).
6. Numerical and graphical results.
7. Convergence issues.
8. Problems encountered.
9. Sensitivity Analysis.
10. Conclusion/discussion.
11. References.
12. Appendix: R and JAGS Code, Necessary Figures and Tables. A regression-type model or hierarchical model is recommended.
Grading The project will be graded on the writing, including English, orga- nization, neatness and flow; and content, including justification and sensibility of the analysis. Allowance will be made for the complexity of the problem. Simple data sets will need to have substantially more thorough sensitivity anal- yses as compared with complicated data sets and models where getting a single posterior requires a substantial amount of work.
Feedback Please see me if you aren’t sure whether a particular analysis/data set/problem is appropriate or doable. We’ll talk through your goals and interests and the data set.
Turn it in Please turn in a PDF to Bruin Learn for the abstracts and the final project.
Each person must analyze a different data set. Joint custody of data is not allowed.
More on Data Sources
Some suggestions on data sources:
● Collecting your own data.
● Your research group.
● NCHS/CDC https://www .cdc.gov/nchs/data_access/ftp_data.htm.
● Census data.
● Google’s dataset search engine: https://datasetsearch.research.google . com (I haven’t checked this out.)
● Data.gov: https://www.data.gov
● https://data.mendeley.com (I haven’t checked this out.)
● Our world in data https://ourworldindata.org/
● Michigan ICPSR https://www.icpsr.umich.edu/web/pages/ They try to be careful about collecting sufficient information.
For any data set, particularly from the larger sources, the data may be lacking in documentation, and that will make it hard to use. So please check that it is usable! For most any of these sources, you need to provide a “reason for being”: What is the purpose of your data analysis?
For some data sources you have to decide on the variables you want to use – the data source is a huge collection of variables for (say) countries, states, counties, cities. For another example, Wikipedia provides lots of data sets (one variable at a time!) for many or all countries.
Some places are not good sources for data because they may have (a) poor curation – variables are not clearly defined, zeros or missing data are not identified, (b) the “story” isn’t described, the reason the data was collected in the first place is not given. Example of poor sources include Kaggle and UCI machine learning repository and most textbook datasets. The large repositories (or link repositories) are scrapping data and probably are not checking on data and story/curation quality.
You also can’t take a data set from one of your classes or the corresponding text- book, or from the Gelman et al BDA book. Sports data sets, while major league fun, can be difficult to work with: make sure the predictors you were thinking of aren’t actually a part of your outcome and that observations are independent (hint: they usually aren’t!).