Ah, statistics - those wonderful numbers that carry a 14.32345% chance of accurately representing what you think they do.
The 18% figure under discussion might be more accurately called "the % of players who were logged in and chose to share statistics that chose a male character on the xbox360 within the first week after the game released" - or something along those lines.
Personally, I don't ever login when gaming unless I specifically want to access some online component of a game. And I decline any statistics sharing when given the choice. I've often wondered how long devs continue to collect data about their games, and how much weight they are given in future design philosophies.
Of the ME youtubes I've watched, I would guess that well over half of them feature a female Shepard. But - I dunno.
I don't think vgchartz tracks digital sales, and they're not particularly known for their accuracy to begin with.
Sans any kind of insider information, we're all really just guessing about how vgchartz acquires and processes its data.
Bear in mind, digital sales are not restricted to PC. I've purchased several of the games I own - including the ME trilogy - through PSN. As for consoles, the number of people who have played a game on console =/= the number of copies sold, since used copies are frequently traded. Also, there's this thing called Playstation Plus, which allows subscribers to play games they've not purchased - I'd expect Xbox has a similar service.
(Not at you specifically)
I've seen some interesting discussions about how vgchartz accumulates data, and how accurate it might be. It stands to reason that they adjust actual retail information reported to reflect digital sales, even if they're only using estimates. I've seen numbers floating around about digital sales relative to hardcopies, and I
believe those figures come from the publishers themselves - if not directly, then from financial analysts sifting through their published financials.
I guess I generally tend to view vgchartz as reasonable estimates, if not entirely 100% accurate.