Richard A. Dunn and Nathan W. Tefft, "Replicating the Levitt and
Porter Estimates of Drunk Driving", Journal of Applied Econometrics,
Vol. 35, No. 6, 2020, pp. 786-796.

The included scripts (zipped in dt-files.zip) implement the
estimation methods developed by Levitt and Porter (2001), hereafter
LP. The scripts use data from the Fatality Analysis Reporting System
(FARS) from the National Highway Transportation Safety Administration
(NHTSA). You can manually obtain the National FARS datasets between
1975 and 2017 from NHTSA's website (NHTSA, 1975-2017).

In addition, compressed versions of all the FARS files from 1975 to
2017 are included in the directory FARS. These files are compressed
using the 7z format instead of the original zip format, because 7z
achieves a much higher level of compression.

Both Python and Stata versions of the replication scripts are included
(the Python scripts are also available as a Github repo at
https://github.com/ntefft/lp).

To replicate LP in Python (recommended), run the following scripts, in
order:

1. retrieve.py to retrieve and save the raw data from NHTSA's FTP
site. If this automated download process stops working (because of
NHTSA web site changes), you can manually obtain the National FARS
datasets between 1975 and 2017 from NHTSA's website (NHTSA,
1975-2017). The extract.py file (below) will look for them in the
"replication\data" folder.

2. extract.py (in the "replication" folder) to extract and harmonize
relevant estimation variables across survey years. The combined
accident, vehicle, and person files are stored in the replication\data
folder.

3. replicate.py (in the "replication" folder) to generate summary
statistics and estimation results that replicate LP.

To replicate LP in Stata, run the following scripts, in order:

1. ddwc_rep_fars_extract.do (in the "replication analysis" folder) to
extract and harmonize relevant estimation variables across survey
years. You can store the accident, vehicle, and person files in the
same folder or create a "data" folder for them and adjust the do file
accordingly.

2. replication_paper.do (in the "replication analysis" folder) to
generate summary statistics and estimation results that replicate LP.

References

Levitt, Steven D., and Jack Porter. "How Dangerous are Drinking
Drivers?" Journal of Political Economy, 2001, 109(6), pp. 1198-1237.

National Highway Transportation Safety Administration (NHTSA). 
1975-2017. Fatality Analysis Reporting System. U. S. Department of 
Transportation, https://www.nhtsa.gov/node/97996/251 (accessed May 15, 
2020).