It appears that the code here updates the location data by re-geocoding based on the primary and secondary roadway involved.
But in in a number of cases, there was already more precise location, or a house number along the primary street which would be more accurate to geocode from instead of the nearest intersection. To illustrate the difference. Here are screenshots of similar area before and after the re-geocoding.
Before: Location data in the original source data

There you can see crashes all along 3rd and 10th and along the bypass / College Mall Road.
After: Location data re-geocoded to intersection of the primary and secondary roadway

After the processing, the crash locations have become clustered around the nearest intersections.
Some alternate approaches could be:
- Leave pre-existing location data alone.
- Or if there's a compelling case to re-geocode the location data, first try a house number on the primary street, and only re-geocode to the intersection as a last result.
- Consider adding a data column that disclaims how re-coding as done. Maybe an enum with values like "original", "house_address" or "intersection". Also, if geocoding accuracy is returned from the geocoding service, considering adding a column to disclaim the accuracy.
This issue relates to #2 and #6, which are also geocoding issues.
CC: @carsonology
It appears that the code here updates the location data by re-geocoding based on the primary and secondary roadway involved.
But in in a number of cases, there was already more precise location, or a house number along the primary street which would be more accurate to geocode from instead of the nearest intersection. To illustrate the difference. Here are screenshots of similar area before and after the re-geocoding.
Before: Location data in the original source data
There you can see crashes all along 3rd and 10th and along the bypass / College Mall Road.
After: Location data re-geocoded to intersection of the primary and secondary roadway
After the processing, the crash locations have become clustered around the nearest intersections.
Some alternate approaches could be:
This issue relates to #2 and #6, which are also geocoding issues.
CC: @carsonology