To be clear the paper claims they developed their own methodology. And that this methodology lines up with the previous source: "This figure is consistent with research conducted by Ratliff in 2014 (n. 5) and Chainalysis in 2017.6"

Whether you want to believe them is another issue, may simply be that they took the source and built a curve around it.

Happy for any other sources or surveys you care to share.

It sounds like you think scams and user error are not equivalent. Let's say user error losses are 150 bitcoin per day instead. That would suggest scams are 10X losses from user error.