Hacker News Reader

Comment by QuercusMax | Hacker News Reader

The part about birthdates being on first of the month or divisible by 5 seems pretty weak to me. Records weren't great back then and many very old people may not actually know their true birthdate.

View on HN

>Records weren't great back then and many very old people may not actually know their true birthdate.

Exactly why they can't be trusted!

Taking data for a whole country and drawing conclusions about a specific region may not be so good. For example, in the paper there is a lot of talk about bad data in the US but when they turn to the blue zone in the US there is no issue with the data. It's considered well kept.

For other places, like Sardinia, they don't explain whether the variations (1st of a month, etc) are happening at the local level. Can the data there be trusted? We are left to draw conclusions because many gaps aren't explicitly filled in.

The blue zones project limited the places they would label a blue zone to those that had good data. Many places were rejected because the data wasn't good. Looking at those places, like the UK or France, and stating they have untrustworthy data actually confirms the blue zones work of rejecting claims there due to untrustworthy data.

There is some statistical thing about fraud and the frequency of certain numbers being made up. I don't recall it specifically but made up amounts, dates, number have certain clusters of numbers vs what normally occurs.

edit; didn't get all that was in my head out :-) So, it could be made up or support the fact that actual docs or good record keeping weren't a thing.

https://en.wikipedia.org/wiki/Benford%27s_law

thanks, could not remember it and my google fu was poor trying to look for it :-)

I think you're thinking of Benford's law: https://en.wikipedia.org/wiki/Benford%27s_law

It's pretty strange, basically if you have a document with a bunch of numbers (say, some company's quarterly report), look at the leading digit of all the numbers. For some reason, numbers with a leading digit of 1 show up more often than those with a leading digit of 9 (i.e. 1,234,567 is more likely that 987,123).

You'd think there wouldn't be any particular pattern in the leading digit, I mean why should there be? But observational data seem to suggest a pattern.

So Benford's law can be used as a leading indicator of fraud. If you apply it to a quarterly report and there's an unusually high distribution of 8's, for example, then while you can't be certain that it's fraud, it might be flag to an inspector/regulator to take a closer look.

It's pretty obvious, really. A quick little sketchproof; say you have a metric, like headcount, revenue, expenses - whatever. These tend to grow exponentially-ish over time. When that's the regime, the number spends 1/3rd of it's time between 1 and 2 of its leading digit, and 2/3rds growing from 2 to the next power of 10. Similar logic applies if you're counting things that follow any power law - population of cities etc. Wherever you have a power law distribution, Benford's law applies.

But when humans enter data, they tend to fake numbers with a uniform distribution - to appear more random. That's how you catch it.

If you are going to make up a fake birthday and you think you might have trouble remembering it, picking a nice round number might make it easier.

In Norway, the authorities had a problem with our SSN equivalent a few years ago - for decades, it had been SOP to assign any immigrant with unknown birth date the birth date January 1st.

Eventually, they ran out of valid SSNs for Jan 1st births in some years. (The number is on the form DDMMYY XXXYY, where XXX is assigned sequentially and YY are control digits.)

Hence, for any given date the system can accommodate 1000 people, plenty in a country with some 1000 births a week. Until you start assigning a certain date to people with unknown DOB, that is. They are now assigned a random date.

In the absence of that sort of SSN, it has a higher impact than you'd think. It comes up enough in the UK to be something we have to plan around.

It's a particular problem in refugee communities, especially where there may be common names. No certificate, often. What happens is that the first time they need to know their birthdate is when they have interaction with a healthcare system, and the doctor (or the admin staff), when told "Oh, some time in 1931, I think", puts "1/1/1931" into the records.

All it takes is two of "Samuel Goldstein, born 1/1/1931" in the same suburb and you've got a serious risk of misidentification when one of them has a heart attack and turns up in an ambulance. Misidentification of patients might be relatively uncommon, but the danger when it happens is severe.

You'd have to compare with a control group. If being 100+ was strongly correlated with having a birthdate on the first of the month or such then... maybe be suspicious.

That's exactly the point - it means that someone "guessed" the date. But if they guessed the day, the year isn't reliable, either, especially at these time scales.

Here's an example of how a similar principle can be used to observe electoral fraud: https://en.wikipedia.org/wiki/2011_Russian_legislative_elect...

Reply on news.ycombinator.com