Determine how to geocode voter addresses

[joshsmith]

We need to determine how we're going to geocode voters' addresses.

[joshsmith]

Here's a table that shows estimated voter file costs, estimated number of addresses in those states, and then the cost to geocode 100% through Google (without Premium pricing) and the cost to geocode 14% through Google (optimistic best case of using SmartyStreet's 86% Zip9 addresses).

State	Voter File Cost	Est. # of Addresses	Cost 100%	Cost 14%
Alabama	$30,000	1,838,683	$919.34	$128.71
Alaska	$20	251,899	$125.95	$17.63
Arizona	$32,500	2,370,289	$1,185.14	$165.92
Arkansas	$2.50	1,129,723	$564.86	$79.08
California	$30	12,542,460	$6,271.23	$877.97
Colorado	$1,000	1,977,591	$988.80	$138.43
Connecticut	$300	1,355,849	$677.92	$94.91
Delaware	$10	335,707	$167.85	$23.50
Florida	$5	7,158,980	$3,579.49	$501.13
Georgia	$5,500	3,518,097	$1,759.05	$246.27
Hawaii	$500	449,771	$224.89	$31.48
Idaho	$20	579,797	$289.90	$40.59
Illinois	$500	4,772,723	$2,386.36	$334.09
Indiana	$5,000	2,481,793	$1,240.90	$173.73
Iowa	$1,950	1,226,547	$613.27	$85.86
Kansas	$200	1,110,440	$555.22	$77.73
Kentucky	$450	1,694,996	$847.50	$118.65
Louisiana	$5,000	1,707,852	$853.93	$119.55
Maine	$2,200	553,823	$276.91	$38.77
Maryland	$125	2,146,240	$1,073.12	$150.24
Massachusetts	$0	2,530,147	$1,265.07	$177.11
Michigan	$22.50	3,823,280	$1,911.64	$267.63
Minnesota	$51	2,107,232	$1,053.62	$147.51
Mississippi	$2,100	1,088,073	$544.04	$76.17
Missouri	$100	2,360,131	$1,180.07	$165.21
Montana	$5,000	405,525	$202.76	$28.39
Nebraska	$500	725,787	$362.89	$50.81
Nevada	$500	999,016	$499.51	$69.93
New Hampshire	$8,200	518,245	$259.12	$36.28
New Jersey	$2.55	3,186,418	$1,593.21	$223.05
New Mexico	$4,800	761,938	$380.97	$53.34
New York	$0	7,234,743	$3,617.37	$506.43
North Carolina	$0	3,715,565	$1,857.78	$260.09
North Dakota	$5,000	287,270	$143.64	$20.11
Ohio	$0	4,557,655	$2,278.83	$319.04
Oklahoma	$0	1,444,081	$722.04	$101.09
Oregon	$500	1,516,456	$758.23	$106.15
Pennsylvania	$500	4,958,427	$2,479.21	$347.09
Rhode Island	$700	410,058	$205.03	$28.70
South Carolina	$160	1,780,251	$890.13	$124.62
South Dakota	$2,500	323,136	$161.57	$22.62
Tennessee	$2,500	2,475,195	$1,237.60	$173.26
Texas	$1,250	8,886,471	$4,443.24	$622.05
Utah	$1,050	886,770	$443.39	$62.07
Vermont	$0	257,004	$128.50	$17.99
Virginia	$5,000	3,022,739	$1,511.37	$211.59
Washington	$0	2,629,126	$1,314.56	$184.04
Washington DC	$2	263,649	$131.82	$18.46
West Virginia	$6,000	741,390	$370.70	$51.90
Wisconsin	$12,500	2,288,332	$1,144.17	$160.18
Wyoming	$0	222,846	$111.42	$15.60

[joshsmith]

The table above may not be the most accurate. You can view the latest at this Google spreadsheet.

[joshsmith]

If we use SmartyStreets for the majority of addresses (70-86%), then we can get reasonably accurate. We can use Google to fill in all the gaps.

I'm thinking for a given Address we might have at minimum columns like geocoded_at, geocoding_source, accuracy_level.

[joshsmith]

For addresses that are Zip9 accuracy in SmartyStreets, we may want to calculate the size of that Zip9's geographic area and assume a certain level of inaccuracy above some threshold, and slowly work to improve addresses as we go.

[joshsmith]

Over time, we can potentially collect data from canvassers on the ground and use that data from their mobile device to get a clearer picture of geocoded data. Our ultimate aim for address data on its own should be to provide the best to-the-door accuracy for a canvasser.