Cloning the FERC Form 1 DB

FERC Form 1 is… special.

The Form 1 data is published in a particularly inaccessible format (proprietary binary FoxPro database files), and the data itself is unclean and poorly organized. As a result, very few people are currently able to use it at all, and we have not yet integrated the vast majority of the available data into PUDL. This also means it’s useful to just provide programmatic access to the bulk raw data, independent of the cleaner subset of the data included within PUDL.

To provide that access, we’ve broken the pudl.extract.ferc1 process down into two distinct steps:

  1. Clone the entire FERC Form 1 database from FoxPro into a local file-based sqlite3 database. This includes 116 distinct tables, with thousands of fields, covering the time period from 1994 to the present.

  2. Pull a subset of the data out of that database for further processing and integration into the PUDL data packages and sqlite3 database.

If you want direct access to the original FERC Form 1 database, you can just do the database cloning, and connect directly to the resulting database. This has become especially useful since Microsoft recently discontinued the database driver that until late 2018 had allowed users to load the FoxPro database files into Microsoft Access.

In any case, cloning the original FERC database is the first step in the PUDL ETL process. This can be done with the ferc1_to_sqlite script (which is an entrypoint into the pudl.convert.ferc1_to_sqlite module) which is installed as part of the PUDL Python package. It takes its instructions from a YAML file, an example of which is included in the settings directory in your PUDL workspace. Once you’ve created a datastore you can try this example:

$ ferc1_to_sqlite settings/ferc1_to_sqlite_example.yml

This should create an SQLite database that you can find in your workspace at sqlite/ferc1.sqlite By default, the script pulls in all available years of data, and all but 3 of the 100+ database tables. The excluded tables (f1_footnote_tbl, f1_footnote_data and f1_note_fin_stmnt) contain unreadable binary data, and increase the overall size of the database by a factor of ~10 (to ~8 GB rather than 800 MB). If for some reason you need access to those tables, you can create your own settings file and un-comment those tables in the list of tables that it directs the script to load.


This script pulls all of the FERC Form 1 data into a single database, but FERC distributes a separate database for each year. Virtually all the database tables contain a report_year column that indicates which year they came from, preventing collisions between records in the merged multi-year database. One notable exception is the f1_respondent_id table, which maps respondent_id to the names of the respondents. For that table, we have allowed the most recently reported record to take precedence, overwriting previous mappings if they exist.

Sadly, the FERC Form 1 database is not particularly… relational. The only foreign key relationships that exist map respondent_id fields in the individual data tables back to f1_respondent_id. In theory, most of the data tables use report_year, respondent_id, row_number, spplmnt_num and report_prd as a composite primary key (According to this FERC Form 1 database schema from 2015.

In practice, there are several thousand records (out of ~12 million), including some in almost every table, that violate the uniqueness constraint on those primary keys. Since there aren’t many meaningful foreign key relationships anyway, rather than dropping the records with non-unique natural composite keys, we chose to preserve all of the records and use surrogate auto-incrementing primary keys in the cloned SQLite database.