Hi Gary, I was not complaining about a three digit postal code boundary
file, but instead the six digit postal code point file, which I was using
for a number of cities in Canada. The six digit postal code file should have
located a point on the block face of one street segment. That is, I expected
all vector segments to have a postal code point assigned to the vector
segments midpoint coinciding with the correct street address. Unfortunately
a number of postal codes were off by three or four blocks and a couple of
them were off by two kilometers. That level of inaccuracy is simply
inexcusable. Also I understand that there might be some situations in which
a few postal codes may have the same geographic location, however that was
not the situation in this case.

The main point of my argument is that I bought data in good faith believing
it to be accurate. If I was informed or warned that the postal code file had
some inaccuracies I could have made a decision to geocode on another data
source, e.g., street files, etc.

Yours truly,
Derek

From: Gary <[EMAIL PROTECTED]>
To: Derek Wilson <[EMAIL PROTECTED]>
Cc: [EMAIL PROTECTED] <[EMAIL PROTECTED]>
Date: February 12, 1999 9:08 AM
Subject: Re: MI Re: Just how accurate is that data we spend our money on?


>Derek,
>
>    I don't know if this applies to the specific problems that you
>encountered, but I do know of a few caveats to be aware of when dealing
>with US zipcode files (this may or may not apply to other nations'
>postal codes):
>
>First, you need to understand that attempting to place a geographical
>limit on a zipcode is, at best, an estimate.  Zipcodes are not defined
>by the USPS in geographical terms.  They are a compilation of carrier
>routes, which, in turn, are each a list of addresses.  Vendors that
>attempt to place boundaries around this are making a guess as to what
>the area covered by those addresses are.  By attempting to create these
>boundaries in a way that avoids overlap between them, they automatically
>introduce errors.  However, most people do not want to see their
>zipcodes overlapping, so they accept the trade-off.
>
>Secondly, as soon as you generalize an area (which, as discussed above,
>a zipcode really isn't, at least not in the same sense as say a census
>tract) to a point location (such as a centroid), you are accepting a
>generalization.  What is a 100% accurate point that describes a
>zipcode?  Is it the centroid of the generalized boundary that we chose
>to represent that zipcode?  Is it the center of mass of all the
>addresses that make up that zipcode?  Is it the location of the Post
>Office that services that zipcode?  By accepting a point to represent a
>zipcode, you need to understand the assumptions that come along with
>that.  Your data vendor should be able to provide you with a description
>of the methodology that they used to place those points.
>
>Finally, in regards to the many-postal-codes-at-one-point problem, this
>is often the case for US zipcodes.  Take for example the case of postal
>codes that represent a PO Box zipcode (the USPS generally takes all the
>PO Box addresses at a post office and assigns them to one or several
>zipcodes of their own).  In this case, the data vendor actually does
>have a clue about where to geographically locate those zipcodes.  Since
>the PO Boxes are physically located at the Post Office and we have no
>way of identifying where the owners of the PO Boxes actually reside, it
>seems logical to place the zipcode point at the location of the Post
>Office.  If you have several zipcodes at a single post office, then
>those points will most likely be coincident.
>
>While I may not have addressed your specific problem, I hope that this
>has been helpful.
>
>Good luck,
>Gary.
>
>Derek Wilson wrote:
>
>>  Hi, does anyone have any warnings, comments or concerns regarding
>> data purchased from a vendor. The last postal code point file I
>> purchased had numerous duplications of lat and long values and a
>> number of points were simply not in the right location. When I
>> contacted the vendor to complain, they tried to downplay the
>> inaccuracies. HOWEVER, if someone in our office who had local
>> knowledge of the area had not discovered the errors our client would
>> have considered us at fault. Try explaining to a client "oh, you don't
>> understand. You see there were errors in the data we used to geocode
>> your customers. That's why the locations are off by 3 or 4 blocks, or
>> that's why a number of customers with different postal codes appear on
>> top of each other." AND the thing that really kills me, is that the
>> data vendor made no mention that there data is not 100% accurate. I do
>> know to expect a 100% accuracy is to expect too much, but when you
>> achieve an error rate exceeding 10% then some type of warning should
>> be included with the data you sell. Please let me know your thoughts
>> on my rants,Derek
>

----------------------------------------------------------------------
To unsubscribe from this list, send e-mail to [EMAIL PROTECTED] and put
"unsubscribe MAPINFO-L" in the message body, or contact [EMAIL PROTECTED]

Reply via email to