Too Unique to Hide

Oh, it's a story about UK postcodes. To an Australian street addresses in the UK border on the insane.

When I was in the UK I had to travel to a B&B at night. I had a GPS but they hadn't given us a street address - just a postcode and house name. You couldn't enter that into our ancient, so I rang them. Initially they didn't give me an address, telling me it was useless. When I persisted they gave me a street name and suburb. I forget the suburb, but I will never forget the street name. It was "The Street".

I entered what I had in the GPS, and two hours later I arrived where it said to go. But there was no house with the name given. So we rang them again. Turns out there were three streets named "The Street" all within 1km of each other and we were in the wrong one.

We arrive at the correct "The Street", but still can't find the house. We ring them for help, but it's freezing and raining, or at least doing what passes for rain outside of London and they are reluctant. Admittedly it's a short street. We feel like fools, but as we still can't find it so keep calling, and eventually our hosts stand on the road in the rain and flag us down. Turns out the B&B is on a battle axe block and the car has to push past bushes in the entrance to get onto the driveway.

At breakfast I make the mistake of expressing my opinion of the UK's street address system. They look insulted and defend it vigorously. Then as we load the car I hear them having a loud "discussion". It turns out a new postman has been appointed who hasn't learnt the house names yet, so the mail hasn't arrived for a few days. They were discussing who should go to the post office to collect it.

As an aside, other countries can have considerably smaller post code areas than US or UK. For example, here's the size of a typical Swedish post code, probably less than a thousand people:

https://www.hitta.se/kartan!~57.68616,11.91682,16z/tr!i=5N3q...

I found it a little on the nose to complete the first interactive (about how very few points of data can track you between datasets), only to receive a Share on Twitter/Facebook.

If only people would practice what they preach.

There are several metrics in addition to the k-annonimity (https://en.wikipedia.org/wiki/K-anonymity) that has been going around the past few days. For example l-diversity (https://en.wikipedia.org/wiki/L-diversity) and t-closeness (https://en.wikipedia.org/wiki/T-closeness)

t-closeness, is defined as: An equivalence class is said to have t-closeness if the distance between the distribution of a sensitive attribute in this class and the distribution of the attribute in the whole table is no more than a threshold. A table is said to have t-closeness if all equivalence classes have t-closeness.

In short: the distribution of a particular sensitive value should not be further away than a distance t from the overall distribution.

Using the t-closeness metric circumvents issues associated with k-anonymity and ℓ-diversity. Briefly, k-anonymity states that a certain attribute class should be present in at least k records, which introduces ambiguity in the data set. However, if each of the k equivalence classes are the same, properties could still be resolved simply by elimination. The ℓ-diversity metric circumvents this problem by adding a further requirement: in addition to the class to being seen in k records, these records must have at least ℓ ‘well represented’ values. But if an attacker knows the real-world distribution of values, then attributes could still be disclosed with a certain probability, simply by combining different data sources

The implications of uniqueness through multiple seemingly innocuous data points are quite worrisome in a wider context. For example, in the digital sphere, if X person has a certain pattern of sites visited, browsing habits and so forth, and then tries to anonymize themselves as person Y for browsing in a way that has no direct connection to any of their regular online accounts, a third party correlating even a small amount of browsing and even vague location data by a bunch of unknown activity profiles could presumably then quickly identify that one of them, Y has a high chance of being X. Or if a bunch of online comment text, buying habits, random basic information or whatever from known person profile X were used to search for similar data points among random people, then correlated together, finding out that Y (who is X but decided to change their physical address, country of residence and so forth for personal reasons) matches very closely with X and is thus very likely to be X seems very easy.

Not a valid Postcode District...

Yes it bloody is...

Continuing anyway, I arrive at "NaN% of the time!"

I don't think I understand the number I got from this; after putting in my info, it says that if a company put in my demographic info and got a result, it "would be me 39% of the time". Shouldn't that fraction be 1/n for some positive integer n? Are certain people with the same demographic info more likely to be returned than others? Or does the data service give close but not exact matches some percentage of the time? What am I missing?

The company I worked for previously had access to a big dataset that was “anonymised”.

There was one poor individual who was the only person who lives in a certain post code area.

Ended up having to get the dataset owners to remove postcodes altogether (GDPR just came in and everyone was on high alert).

I guess it makes sense after you think about it for a minute. This is an illuminating execise to be sure!

It's funny how in my head, it seems like it would be way less likeley that you could be indentified with such a high degree of confidence just from those 3 peices of info.

None of the Belfast[0] codes are returning as valid; same for the rest of NI.

[0] - https://en.wikipedia.org/wiki/BT_postcode_area#Coverage

The scary thing is that this is what they call "anonymized".

Edit: changed.

If the day is removed from the date would this be sufficient to make the data anonymous or should the month also be removed?

Couldn't enter a month of birth . . .