This project aims to analyze and provide some summary statistics for NIH grant award data from ExPORTER data files.

The data are cleaned and preprocessed in the Preprocessing.ipynb notebook. This is meant to be run against the raw data from ./data/exporter and the output is saved to ./out/csv with the format: out/csv/post_processed_{ISO_DATE}.csv.gzip.

This project does not house the ExPORTER data files (~200MB). As a convenience, we include a bash script ./utils/get_csvs which programmatically downloads all of the files listed in ./utils/reporter_files.txt. Once you have the exporter files downloaded in the ./data/exporter directory, the preprocessing notebook will be able to find them. The files list in reporter_files.txt is up to date at the time of this publishing. However, as new files are added weekly, you should add the new files to reporter_files.txt and re-run ./utils/get_csvs. A more efficient way might be to just download the new file directly to ./data/exporter with curl like so:

> curl https://exporter.nih.gov/CSVs/final/RePORTER_PRJ_C_FY2020_031.zip -o "./data/exporter/RePORTER_PRJ_C_FY2020_031.zip"

Any column meant to represent a date is converted to datetime64[ns] using pandas.to_datetime. Some columns that are strings are explicitly converted to StringDtype as it is the recommended way to store strings in pandas.

Some rows are dropped during preprocessing as they would be unhelpful for analysis. This is primarily because they:

• have no associated costs
• are duplicates of another award

When filtering out the above cases, about 3.433% of rows were dropped.

Although ~98.3% of ExPORTER data reference awards that are administered by an NIH agency, the data also includes awards from other government agencies (CDC, FDA, VA). Since this analysis is only relevant to NIH, rows that are not NIH awards are dropped. Also, we drop all rows where ORG_COUNTRY ("country in which the business office of the grantee organization or contractor is located") is not the United States.

Since all US award data have related zip codes for the organization, we added additional columns for lat/long coordinates. The "US Zip Code Latitude and Longitude" by CivicSpace Labs is licensed under Creative Commons Attribution-ShareAlike. Copyright 2004 CivicSpace Labs.

.
├── Analysis.ipynb                                               # this notebook
├── Preprocess.ipynb                                             # preprocessing notebook
├── CHANGELOG.md                                                 # When data or models are generated, a corresponding entry goes here
├── README.md                                                    # Project details
├── data
│   ├── exporter                                                 # exporter files (not included in repo) 
│   ├── geo                                                      # shapefiles for maps
│   │   └── cb_2019_us_state_500k.{cpg,dbf,prj,shp,shx,xml}
│   └── zips                                                     # Zip to lat/long
│       └── us-zip-code-latitude-and-longitude.csv 
├── out
│   ├── csv                                                      # output of preprocessing
│   │   └── post_processed_2020-05-29T12:36:20+00:00.csv.gzip    # sample (Note: this is not the entire dataset, see utils/get_csvs for script to get all files for yourself)
│   ├── json                                                     # any json output goes here
│   └── models                                                   # any saved models go here
├── styles                                                       
│   └── styles-nih.css                                           # external styles
└── utils
    ├── get_csvs                                                 # function to get data
    ├── reporter_files.txt                                       # list of reporter files to retrieve
    └── style_dataframes.py                                      # Helper methods to style dataframes  

9 directories, 19 files