An interesting result not related to the title is some scary disagreements between popular geoip databases and the active probes the authors used(§6.2):
> All five of the IP-to-location databases are more likely to agree with the providers’ claims than either active-geolocation approach is. As discussed earlier, we are inclined to suspect that this is because the proxy providers have influenced the information in these databases. We have no hard evidence backing this suspicion, but we observe that there is no pattern to the countries for which the IP-to-location databases disagree with provider claims. This is what we would expect to see if the databases were being influenced, but with some lag-time.
The paper does call out the geoip databases in question (DB-IP, Eureka, IP2Location, IPInfo, MaxMind) -- but not the VPN providers, which is unfortunate.
Also, the code looks quite good: https://github.com/zackw/active-geolocator