Data triangulation makes it possible to determine an extraordinary amount of information simply by connections. Even when a data point is in itself a void.
I used to explain it as the "no point in staying off Facebook" effect.
It doesn't take many people who have you in their address book, and who join Facebook, for Facebook (other data hoovering services are available) to know pretty much everything there is to know about you. And that's without you ever creating an account.
Firstly it knows you exist - since your email address crops up in your friends address books (remember how Facebook offered to scan your contacts ?). From there it starts to know if you are a member or not. If you are not it can still know enough about you to punt your details to advertisers. After all, if your friends like x,y, and z, then it's likely you will to. And you will also fit into other monetisable demographics. The more people Facebook can draw connections from and too (1st, 2nd, 3rd line connections) then the accuracy of it's guesses about you improves.
The Data Protection Act deliberately totally failed to address this phenomenon.
Suffice to say is given enough data about enough people, you can tell a lot about the people you haven't got data on.
This would have remained a mere curiosity if we (i.e. mankind and techies) hadn't almost immediately (in social terms) then come up with a mechanism to trawl that data at the speed of thought using language modelling to analyse unbelievable amounts of information. But we did, and here we are.