Normally I'd be quippy about the plural of anecdote not being "data", but this isn't even plural. This is a single anecdote. The claim you have made is "I have encountered fraud, personally, so the system is a disaster."

Well run systems experience fraud. It's something you generally want to minimize, but like, it's not necessarily an indicator that the system is broken. Like... AWS has tons of fraud. AWS is still very much not a disaster. (Well, it kind of is a disaster, but mostly because it's a machine that chews up humans via oncall, which is unrelated to their fraud.)

As claimed, light reading confirms this observation to be data not anecdata.

"NY's COVID unemployment fraud topped $11B, partly due to system failures..."

https://www.yahoo.com/news/nys-covid-unemployment-fraud-topp...

Look at more recent data. https://www.osc.ny.gov/files/state-agencies/audits/pdf/sga-2...

It looks like fraud rate was typically 5-10%, which might be high, might be "fine". In 2020-21 and 21-22 fraud rate jumped way up to 20%, which is obviously way too high.

But in 2023-24 fraud rate is back down below 10%. We don't have 24-25 data yet, but it looks to me like we had a couple of unusual years during the pandemic, but audit controls seem to have reigned a lot of that back in.

I'd say, evaluate this year's data and then decide if this was a blip or not, then revise your mental model with data.

Shouldn't that be roughly constant across the year though? We don't need to trust the numbers to be exact, to observe a precipitous drop.

Should fraud rates in public taxpayer-funded systems stay constant or go down over time, assuming no new fraud type is invented?

Depends what if anything is being done to combat it.

One of the following is true:

- The numbers somewhat-accurately reflect the trend of employment

- Fraud levels were reduced 66x in one year

If it was the second one, that's a sufficiently massive reduction that news stories would be written about it. There would be stories about this great victory over fraud.

A quick search showed no particular anti-fraud measures or claims of effectiveness unique to that time period.

Unemployment numbers will never work for counting the number of people actually unemployed. They do work for seeing unemployment trends. A ton of people suddenly applying for unemployment is a pretty clear signal.

Edit: typo