Okay sure it's not a controlled environment but you must also consider the under reporting present within a system such as this. This has been estimated at 20-100x what's stated. The best estimate I've seen is 41x.
Edit: adding this- even controlled environments have issues with representativeness. For example, can a 1000 person sample actually detect issues that occur with 5-10 bps?

Okay sure it's not a controlled environment but you must also consider the under reporting present within a system such as this. This has been estimated at 20-100x what's stated. The best estimate I've seen is 41x.