-----Original Message-----
From: Jerry Ratcliffe [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, April 17, 2001 11:33 PM
To: [EMAIL PROTECTED]
Subject: [CRIMEMAP:826] re: calculating distances


A modest contribution to the calculating distances discussion...

Kim Rossmo estimated that 10 feet was the possible limit of accuracy for 
mapped crime locations. Most locations for crime mapping purposes are 
derived from the geocoding of addresses using TIGER line data - not the 
most accuracy process. Some work I've recently completed in the Eastern 
Suburbs of Sydney (analysing over 20,000 address cadastral plots and their 
geocoded locations) estimated that a geocoded point is within 64 metres of 
the centroid of the cadastral boundary for the address 95% of the time. In 
75% of the cases the accuracy can get down to 36 metres and 26m only 50% of 
the time. More worrying, if you are doing a strategic analysis 
incorporating census variable and using census tracts then the geocoded 
point and the address plot centroid are in different census tracts up to 
7.5% of the time. The situation is worse if you use the default geocoding 
settings in MapInfo or ArcView (the size of a land plot and the size and 
shape of the house is of course relevant here but in the study area I used 
land plots are very small and average less than 25 metres by 25 metres, 
minimising the impact of this error factor).

A theoretical accuracy of 10 feet is possibly achievable in the distant 
future, but not with the current geocoding processes that most of us in the 
US and Australia are stuck with. A more realistic accuracy limit may be the 
size (minimum radius?) of a small building. If we can get a geocoded point 
inside the boundary of the building we wish to geocode we are doing well. 
Unfortunately this is still rarely the case with TIGER line type geocoding. 
The UK AddressPoint data is significantly more accurate, but significantly 
more expensive.

These limits to TIGER-line geocoding raise questions when people are 
calculating journey to crime distances with levels of analysis and accuracy 
that are spurious when the original geocoded data has an accuracy measured 
in tens of metres and not tens of centimetres. Common sense does need to 
prevail. I would therefore concur with Kim that worrying about a few metres 
or feet here or there is irrelevant and while spherical geometry is 
interesting it has little to contribute to crime analysis over more simple 
methods of distance calculation.

Jerry R

PS If you are interested in this study mentioned here, there is a paper 
coming out at the end of the year in the International Journal of GIS, and 
I may (or may not!) be talking about this in Dallas at the CMRC conference.
Dr Jerry H. Ratcliffe

Postal address:
      School of Policing Studies, Charles Sturt University
      NSW Police Academy, Goulburn NSW 2580 Australia
Physical address:
      Room 2186, Haydon-Allen Bdg,
      Australian National University, Canberra
Tel:  +61 (2) 6249 4139
Mob: 041 48 48 456
www.csu.edu.au/faculty/arts/policing
www.bigfoot.com/~jerry.ratcliffe

----------
I'm as honest as the day is long,
The longer the daylight, the less I do wrong.


~~~~~~~~~~~~ List Sponsor ~~~~~~~~~~~~
Digital GEO data for 125 countries
LAND INFO International produces digital geographic data for over 125
countries.   DEMs, satellite imagery, topo maps, vector map layers,
flood maps, and more.   Visit http://www.landinfo.com/indexdm1.htm and let
our specialists find the right solution.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

_______________________________________________________________________
List hosting provided by Directions Magazine | www.directionsmag.com |
To unsubscribe, send e-mail to [EMAIL PROTECTED] and
put "unsubscribe MapInfo-L" in the message body.

Reply via email to