Madison Comparison

[spencerrecneps]

Category	Original BNA	pybna (normal parking)	pybna (low parking)
Overall	50	38	46
People	56	43	53
Core Services	49	27	36
Dentists	38	15	21
Doctors	39	18	23
Grocery	56	33	43
Hospitals	49	12	13
Pharmacies	41	22	27
Social Services	60	27	35
Opportunity	50	35	42
Employment	44	31	38
Higher Education	54	25	26
K12 Education	54	37	46
Tech/Vocational College	42	8	13
Recreation	65	36	39
Community Centers	43	19	25
Parks	75	67	72
Trails	69	n/a*	n/a*
Retail	56	39	47
Transit	24	6	9

* In the past we have discussed a better version of the trails score. I've left it out completely here but it would be easy to add a measure for access to trailheads--which is what we've discussed in the past--as long as it can be taken from OSM.

[spencerrecneps]

There are three big differences in methodology that are affecting these scores. #3 would apply most significantly to the normal parking scores listed above.

1. Original BNA does not account for detour and would thus have higher scores as documented in % detour differences #47.
2. There is no trails score in pybna pending further discussion about the best way to capture access to trails. Since Madison has a lot of great access to trails this is probably affecting the overall score somewhat.
3. Madison is heavily impacted by the change in our treatment of bike lanes. Many streets with bike lanes flipped from low-stress in the latest online scores to high stress in pybna. Bike lanes adjacent to parking #48

I also ran the pybna scores using destinations I downloaded from OSM at the time of testing, which means there could have been some variation in the destinations adding to the scoring differences.

[spencerrecneps]

I built some testing capabilities into pybna that allow us to verify the LTS scores on a variety of road conditions. I'm attaching that here for you to reference. As you can see, we went through each possible condition and checked what the score should be based on our LTS chart against what pybna was scoring. We corrected any discrepancies as part of the validation process.
lts_assessment.xlsx

This is likely another source of variation if there are any cases where the original was inaccurately rating roads or intersections. We did some testing of the original scores but the new method is far more rigorous and gives me much more confidence that our LTS scores are being accurately applied. (Whether the underlying data is accurate is, of course, an entirely different problem.)

[spencerrecneps]

Here are exports of the two Madison runs.
https://www.dropbox.com/s/2wj42i5zutdoolb/madison.zip?dl=1