I'm trying to understand how using each part of an IP address influences the accuracy of geocoding. For example, if
15.201.xxx.xxx
is used instead of
15.201.123.xxx
what is the approximate distance in KM? I'm using the IP2Location Lite database to examine who has visited a website. The README states the following:
1. IP2Location Lite Edition is free package with accuracy up to Class C (192.168.1.X) only. It is restricted for non-commercial use.
However, I'm having difficulty finding documentation on what the accuracy of this actually means. Does this accuracy vary over time also?
Answer
"Class C" is a historical reference meaning that your location data is summarized to the /24 level. For example, even if 203.0.113.1 and 203.0.113.254 happen to be in very different physical locations, the database will have only one lat/long for the entire 203.0.113.0/24 network.
Potentially useful generalizations about accuracy:
If you summarize to the /24 level (203.0.113.x), you can be reasonably confident that more than 80% of your data points will fall within the correct general metropolitan area.
If you summarize to the /16 level (203.0.x.x), you can expect that more than 80% of your location data points will be on the correct continent and probably even in the correct country.
If you summarize to the /8 level (203.x.x.x), your location data will be almost completely meaningless.
IP addresses that are likely to geocode reasonably well (within 25 miles or less) at the /24 summary level include:
- College/university campuses
- Corporate headquarters
- Cable modem connections serviced by major carriers like Comcast
- Fiber-to-the-premise connections serviced by major carriers like Verizon and Frontier
IP addresses that are reasonably likely to geocode quite poorly (off by 200 miles or more) at the /24 summary level include:
- 3G and 4G wireless phones/hotspots
- Corporate branch offices
- DSL connections serviced by CLECs
No comments:
Post a Comment