Volourn said:
"million+ copies? Where are those numbers from? Even if we consider the numbers of VGCharts to be somewhere within the realm of the accurate at all:
Dragon Age II (X360) - 0.70 mil
Dragon Age II (PS3) - 0.35 mil
Dragon Age II (PC) - 0.26 mil
Even if we double the PC sales, we won't be getting anywhere near the said 2 million. I'm not saying that VGC is accurate, but as far as I know this is just about the only source of information regarding the number of copies shipped, so, if there's some other, I'd be very curious to know it"
Just because it's the 'only' source doesn't make it worthwhile. vghchartz is shit. Plain and simple. It's untrustworthy with its made up numbers.
I'd trust your made up numbers more than vghcartz's numbers because vghchartz doesn't even have faith in its owna ccuracy. L0L
WARNING, THIS IS GOING TO BE THE MOTHER OF TL;DR POSTS!!!
My apologies (
), just really want to get this settled
The only argument I've seen people actually raise against vgchartz is that it uses market sampling rather than having access to raw sales. If that's what you mean by 'made up numbers', then you're a fool who can't into statistics/standard deviation.
There's nothing wrong with market sampling as a basis for measuring sales, so long as you have an accurate error margin. Error margins aren't guesses btw - if the sample is a random one (the 'random' is ultra-important) then for any sample and any population size you can mathematically calculate the error distribution. Ususally you use a 99.95% accuracy, so by saying 'sales = 2 million, +/-0.01 mill', you mean 'there is a 99.95% probability that the sales are between 1.99 and 2.01, with 2 being the most likely'.
That's not 'making up numbers', it's a mathematical calcuation that gives you the right range 99.95% (or higher, if you increase the error margin) of the time.
One immediate problem with Vgcharts is that it doesn't tell us the error margin. BUT you can get really accurate measures with quite a small data pool. I'd be very surprised if it they couldn't get a large enough sample for a measurement within, say, 100,000-200,000. That's 10%,which is a MASSIVE error margin for study - normally you'd expect a decent set of market data to have at most a 2-3% error. I'd say that they could probably get something within 3-5% without the necessary sample size getting out of hand, but the argument that you guys are having doesn't actually need more than a rough measurement.
Just like you can predict a large market with a rather small RANDOM sample, your numbers will get thrown right off if the sample ISN'T RANDOM, even if you have a massive sample size. GET THIS IN YOUR HEAD - SAMPLE SIZE DOESN'T MATTER NEARLY AS MUCH AS THE SAMPLE'S RANDOMNESS! You just about assume that sample size is not a problem unless the randomness is out (read up on standard deviation if you don't believe me - even wikipedia should have it).
Problem is, getting a truly random study ranges from really hard to impossible. Taking surveys by phone? Not random - some genders/ages are more likely to take part. Counting sales from a selection of stores - a SELECTION, you said? Observer bias. Counting sales from a representative sample of stores - are you REALLY saying that all the stores have the same distribution of ages, genders, incomes, education, ethniciities in that area? You get the picture. If it can't be computerised, it probably isn't perfectly random.
Now that's why medical studies DO require massive populations, whereas physics/chemistry studies do not. Every country differs, so you need global stats. Different wealth/literacy/culture affects how likely people are to go to hospital, so you need distribution across each of those. Different ages, again - it adds up to huge numbers, but that's because of the difficulty getting randomness in that field, not because of a 'making up numbers' issue.
Personally, I've got no idea whether computer game sales have a distribution that can be easily studied. They do reasonably well using the same methods for top 40 albums/singles/videos (studio only knows how many they've shipped, not the actual sales, and its not practicable to get data every week from every store, so they use sampling instead), but that always gets divided up by country, and often by musical subgenre. I'd be a lot more confident in Vgchartz stats if they split them up by nation.
So, here's the question Volourn, sorry for the length: what is actually said to be wrong with VGchartz sampling methods? Do they rely upon self-reporting (obviously a hideous method), is it geographical skewing, are they not able to defend their data when challenged, what's the issue said to be? I'm not disagreeing with you that they're crap, because there's plenty of ways that they could be (and it's cheaper/easier to produce dodgy but uncheckable facts than it is to hire a statistician and do what he says needs to be done). But if it is just 'lol how they know 2,000,000 when they only count 2,000', that isn't necessarily a problem.