What percent of all active IP + User Agent combinations are not unique?

Question

It's fairly common practice to define a unique user by a 1-to-1 hash of IP + User Agent. But not all unique users have a unique IP + User Agent combination. Thus: collisions - events misattributed to users who did not fire those events.

This is especially problematic when trying to track unique users who use similar machines and settings all under the same roof of a large organization like a school or enterprise.

My question is: What is the global background probability of IP + User Agent concurrency at any given time?

I assume the rate is different for different times of day and regions. But I'm curious what the standard is.

Answer 1

It is not clear what your goal is. You do not tell us. If you are trying to identify unique users, you are using two of the most unreliable data elements possible. Yes IP addresses are often used, but in conjunction with other data elements. Here is what you are missing.

IP addresses are often dynamic and can change with every session. As well IP addresses can be shared amongst many users. Also consider the limited number of web browsers and the user agents provided. It is extremely common that IP Address + User Agent will result in collisions as you suggest. It is extremely probable in fact. Also consider the amount of times that a user agent is bogus. It seems these days that user agents are bogus more times than they are accurate. Using the user agent for a pivotal data element is a horrible idea because of it's unreliability.

For some of my work, I use IP address and first access Epoch time but depending upon what you are trying to do, this may not be possible or practical. This works okay for a user id but there are far better methods. I do not use this method to track users per se.

If you are writing code, there are modules available to create a unique user id. This may be your best bet again depending upon what you are trying to do. I know that this feature is available in PHP if that is your language, though I cannot tell you what that would be since I do not use PHP. Perhaps someone can comment.

Answer 2

Actually, it has been shown that UserAgent has enough properties that can differ from computer to computer to cause the composite UserAgent string to be unique to few or a single computer/device. To see how unique your personal device appears on the Internet, visit https://AmIUnique.org. At this time, they only have about 3,000,000 records, but most of my devices appear to be uniquely identified.