In Aspen Hill, Maryland (USA), the geodata associated with addresses
is in error.

While addresses at intersections are well-placed, addresses between
intersections are poorly placed.

An example: the geodata for 13900 Parkland Drive is accurate. The
geodata for 14000 Parkland Drive is accurate. The geodata for 13920
Parkland Drive is very inaccurate.

Google's geodata for this area seems to have been derived by
calculation, rather than observation.

Interpolation between the intersections might be a close approximation
suitable for dense cities. In such dense environments, you could
safely assume that for each block, you might have 50 addresses on each
side of the street, and interpolate geodata between the intersections
with such a method.

In the suburban environment of Aspen Hill, Maryland, along the
specific block in question, there are actually 11 addresses on each
side of the street.

The geodata assigned by the interpolation method will calculate that
the address 13920 Parkland Dr is, thus, 20/50ths of the distance
between 13900 and 14000 Parkland Drive. The geodata maps a point only
2/5ths of the distance between those addresses.

Using a database of all addresses in Aspen Hill which I have, the
geodata maps to show that all street addresses in this area were
calculated via interpolative method rather than by direct observation
and recordation.

I propose that Google may wish to select certain areas, which users
may wish to recommend, and apply re-calculating algorithms to
regenerate their geodata/address associations.

Such an algorithm might compare the lowest address on a block and the
highest address on a block, finding the difference, and re-calculating
with that difference as the factor for interpolation, rather than
applying an arbitrary factor of 50 addresses per block as the factor
to interpolate.

Although still not as accurate as direct observation and user-edited
placement, this would be far more accurate than the present system.

In closing, allow me to thank the Google developers for adding the
"edit marker placement" feature.

Regards,

Hi,
  The observations that you have made are indeed very perceptive.
While Google Map Maker is not yet open for editing for USA, your
observations might be helpful for assigning addresses on Map Maker for
other countries as well. Would you mind adding your observations on
the Issue Tracker: http://code.google.com/p/google-mapmaker/

Map Maker Help: http://maps.google.com/support/bin/topic.py?hl=en&topic=23521

Thanks & Regards,
Map Maker Guide.

On Nov 21, 11:53 pm, thardman <thard...@thomashardman.com> wrote:

On Nov 21, 10:53 am, thardman <thard...@thomashardman.com> wrote:

This is assuming that those actual addresses are available*, which may
not be the case, depending on the source of the data. Some data
contain only ranges (like 800-899). In other cases, the algorithm you
suggest may not actually be correct. In some, perhaps recently-
planned, cities, addresses are assigned via the same distance/
interpolation method that is used to calculate the position from an
address. That is, if a road segment that is 5000 feet long is
allocated addresses from 800-899 (800-898 left and 801-899 right),
when a parcel is assigned an address, they measure the distance from
the start of the segment to the main access point (usually a
driveway), divide by the length of the block (5000), multiply by 100,
and add that to the starting address (800), offsetting by 1 if
necessary to put it on the correct side of the street. Sometimes
rounding is also used, to come up with easily remembered numbers.

Other times, particularly in residential areas, they might use a
standard increment between house numbers on a block (often 10),
regardless of driveway location or minor differences in lot size.

Without knowing exactly which scheme is used in which area, I still
think the distance interpolation is the most reasonable default
choice.

If users want to contribute locally observed "address strips", I'm all
for that. Also, if users want to do the work to find the local
assignment scheme and give the info to Google to implement, they
should be receptive to that. This seems to require, in the cases I've
looked at, interviewing someone in the city planning/public works
department, which is usually tasked with assigning addresses. They are
sometimes generally governed by local laws that specify a general
convention (odd/even, name types), but the detailed house number
assignment seems to be up to the particular agency's internal methods.

Google should be able to mark specific addresses/ranges with any of
these source types and then tell the user how the location was
derived.

*I've recently looked at maps in my area of San Bernardino County, CA,
and it is appears that Google has actual lot boundaries and exact
addresses that are likely sourced from local city/county databases.

P.S. Apologies to the reader for the disorganized, perhaps rambling,
nature of this article.