It is the belief of The Jeopardy! Fan that these claims are inherently biased by a poor use of a dataset and completely flawed to the point that it is irresponsible to report on them.
The Casino.org methodology, while it acknowledges that the data is incomplete, appears to be working on the assumption that the incomplete data is representative.
As a Founding Archivist at J! Archive, I can categorically say that the data is far from representative, and not representative to the point that making any conclusions based on that data is improper and irresponsible.
One claim in the original study was that the 1991-92 season was “easiest”, because it had a ratio of “4.2 correct answers for every wrong answer”. That season, in J! Archive, has only 48 games out of 230 recorded. Of those 48 games, easily the featured set was Jerome Vered’s complete run of 5 games. It would therefore stand to reason that Jerome, who completely dominated and steamrollered his opponents in May of 1992, would boost that average significantly because his games take up so much of that season’s average. Its conclusion is thus flawed.
The other outrageous claim was comparing men with women. Again, the vast majority of the early games in J! Archive were put in there of players who later returned to the show in tournaments, either a Tournament of Champions, Million Dollar Masters, Super Jeopardy, the Ultimate Tournament of Champions, or the Battle of the Decades. Most of those “dominant” players were male. Thus, again, the dataset is inherently biased towards male players, and female players who might not have won enough (but still enough that it would have brought things back closer to its true average) are less included from earlier seasons (simply out of the fact that fewer of their games have appeared in private collections or video sharing sites over the years).
It is thus the opinion of The Jeopardy! Fan site that any story that reports on this study or dataset is inherently flawed and its conclusions are inherently invalid.