Bill.Stewman wrote:Thanks for the warning, but most of the geocoding seems OK.
I fully expect that "most" of the geocoding performed by that web site is okay. The problem is that for applications such as emergency preparedness, "most" is not nearly good enough.
Would the stake president want to send a team to rescue the Jones family, trapped on their roof by a flood, directing the team to a location that was a couple of miles from where the Jones house really is? The time to resolve the geocoding errors is long before the flood hits. That's why it's called "preparedness."
Bill.Stewman wrote:What are the better options to place each home more accurately?
One option may be to wait for the Church's own geocoding/mapping application, now reported to be in beta. It apparently will include a step for correction of geocoding errors and capturing the corrected data in the Church's databases. (The contours of that geocoding tool remain to be disclosed outside of the beta.)
Meanwhile, I don't know of a general batch-geocoding solution that is free, accurate and comprehensive. And I have looked hard. Getting most results and making maps is pretty easy. Getting fully accurate results is what is hard. (In my ward, we are lucky enough to have good public-domain geocoded data from the city. With some custom programming, we are able to geocode almost all addresses precisely and automatically, then manually resolve those that fail the quality threshhold. A few households with fundamentally invalid addresses will not be geocoded at all.)
The particular problem with batchgeocode.com is that it disards known errors and warnings it receives from its own source, the Yahoo geocoding API, when addresses are processed one at a time. So, for example, a street address may not be located, but Yahoo plots the Zip-code centroid instead. Or the street number may not be located, so Yahoo plots the center of the whole street, which might be several miles long. Or the street is listed in the ward database as "Wyoming Ave" even though there is no such street. Yahoo guesses that it really is "Wyoming Blvd," which is probably a good guess but may not be. Yahoo flags these as warnings in a way that a program can discern, but batchgeocode.com discards the accuracy detail without notifying the user at all.
The magnitude of this problem will vary with many factors, including how clean the addresses are in the ward directory. In many -- most? -- wards, the quality of that address data is not very good.
Similar problems can occur with other geocoding engines. Geocoding is inherently error-prone, and no batch process can be accurate unless it provides for detection and manual correction of errors -- both errors of omission and of commission.
You might look at HPaulsen's mapping program. One thing it does under the hood -- it is using Google's geocoding API -- is to check the quality code Google returns with each geocoding transaction, and exclude those addresses that fail to meet a threshold of reasonable certainty. The user then can attempt to plot these addresses manually. That application does not export KML files, but theoretically it could be enhanced to do so. (Whether Google would allow that under its Terms of Use is another matter. I'm not sure of the answer.)
Other solutions might include commercial geocoding service bureaus.