The fact that reports on "misinformation" don't look at the denominator when considering the volume of impressions is a great example of selectively reporting statistics to support a preconceived notion.
Not to mention bot farms. The most-viewed source of “misinformation” in the linked 2020 report simply… doesn’t exist anymore. Have we just been hearing about crappy sites buying views these past few years?
The videos they find also show view counts. This allows them to estimate (very roughly!) the view counts across all videos, because it allows them to see the distribution.
The issue with that is if there is one or a handful of videos that have a significant portion of all youtube views. Most likely they will not be in your sample which could lead to a big underestimate.
It should be obvious, but "misinformation" is an arbitrary political designation and therefore a constantly moving target. "You should wear an N95 mask to avoid getting COVID" was misinformation in March 2020 but not 3 months later. "Vaccines may not prevent COVID transmission" was misinformation from 2020 until sometime in 2022. "COVID infection may confer significant immunity from future infection" was misinformation for about the same period. The "lab leak hypothesis" was a "racist conspiracy theory" from 2020 until 2023, when it was officially endorsed by YouTube's political sponsors. And so on.