it's an example of a concept I teach my methods students I call garbage in --> garbage out"peak" as in logged in at one time. Steam user count sits ~50million.
So uh...the rest of your post...yeah.
you can continue finding data for games/gamers not on steam and try to extrapolate more date from it but the glaring problem is as I wrote: we don't know the response rate of the survey
at best we now have a comparison of, even assuming an impossible 100% response rate and discounting multiple accounts for the same person, 50 million steam responses as against nearly 50 million non-steam players.
I don't know what you think you "proved" by your response to my post but you certainly didn't add any support to the claim that the 75% of steam respondents in that survey represent the average, majority, or anything at all of computers in the world that may or may not run games.
it's just garbage data
and when you try to use garbage data in a research model you'll simply end up with garbage conclusions