I had an idea recently: Is it possible, with a reasonable accuracy to estimate how many copies a particular game sold according to its steam reviews?

The starting point would be comparing the copies sold (on pc/steam) vs. the number of steam reviews. This would be done for games which developers had released their sale numbers. And since i think (i might be wrong) that the number of people that review the game they bought on steam is quite the same across any title/genre or "fame" (but thats just and assumption on my side), i think it could be used and then projected on games that didnt release any sale numbers.

For example The Vanishing of Ethan Carter sold around 60 000 copies, around 46200 were on steam, steam reviews are 1538, that means that around 3,33% of players who bought the game wrote a review. I made additional calculations with different numbers from different games (those i know the sale numbers from developers) and came with the average of around 2% players that bought the game wrote a review, but thats based only on few games (6 i think), and sometimes i wasnt really sure if the number the developer released applies only to game (without platform specificity), only to pc, only to digital distribution, only to steam... I tried my best and the average number i got was around 2%. Do you think its a good way to "guess" the sales? Anybody thought about this before?

You might want to look at a few other variables that might come into play as well. I've found that people are more likely to add a review to a game if its extremely good or extremely bad. It needs to invoke a reaction from the player to usually get a review. This could skew those results a bit as mediocre games might have less ratings compared to terrible or great games.

If you have sufficiently large ranges, yes, you could probably make rough approximations.

I'd probably have buckets like:

< 1,000

1,000 - 50,000

50,000 - 500,000

> 500,000.

You might be able to add a fifth.

You might want to look at a few other variables that might come into play as well. I've found that people are more likely to add a review to a game if its extremely good or extremely bad. It needs to invoke a reaction from the player to usually get a review. This could skew those results a bit as mediocre games might have less ratings compared to terrible or great games.

Yeah i thought about that too, but overall do you think its a good idea and agree that it might be quite accurate?

did anybody thought about it before?

Don't forget that Steam reviews are only recently added, so things would be inaccurate if you applied this to games that fewer people have gone back and reviewed after the system was introduced.

Another idea is, since every Steam account has publicly-accessible view of what games they own, crawl over every Steam account and count how many users have each copy of the game.

This doesn't tell you at what price they got the game (maybe 90% off, or maybe it was when the game still cost $9000 dollars), or whether it came bundled with other games, but at least it'll let you know how many official copies are circulating about.

Ars Technica tried this, though measuring different measurements (time played, for example), but didn't realize that the 'time played' statistic is very inaccurate.

You might just ask Ars to give you their dataset for your research.

Don't forget that Steam reviews are only recently added, so things would be inaccurate if you applied this to games that fewer people have gone back and reviewed after the system was introduced.

Another idea is, since every Steam account has publicly-accessible view of what games they own, crawl over every Steam account and count how many users have each copy of the game.

This doesn't tell you at what price they got the game (maybe 90% off, or maybe it was when the game still cost $9000 dollars), or whether it came bundled with other games, but at least it'll let you know how many official copies are circulating about.

Ars Technica tried this, though measuring different measurements (time played, for example), but didn't realize that the 'time played' statistic is very inaccurate.

You might just ask Ars to give you their dataset for your research.

1) yeah i forget about that the reviews were not there always, any clue when they were added?

2) Go through each of the milion steam accounts and count the specific game, how :-)? You dont mean manually thats not in human ability, maybe with some program but im not a programmer, how did you mean that please?

3)Ars Technica tried this, wow didnt know about that, will check it out, thanks for links.

1) yeah i forget about that the reviews were not there always, any clue when they were added?

Only real recently. I think it was less than a year. Certainly less than two years.
You can google this information easily enough to find out: Twenty seconds of googling tells me it was revealed on October 19th, 2013 and entered beta on Nov 25, 2013. That's to say that people have only been using it for 11 months.

2) Go through each of the milion steam accounts and count the specific game, how :-)? You dont mean manually thats not in human ability, maybe with some program but im not a programmer, how did you mean that please?

Yes, with some web-crawling script reading the pages and parsing them, pulling the information out. It's not within my circle of experience, but it definitely can be done. And hey, Ars Technica did it. Or rather, one journalist who works for Ars Technica who was board did it in his spare time.

3)Ars Technica tried this, wow didnt know about that, will check it out, thanks for links.