Dalthien said:
That's the same problem that I have with the site. I'm sure that a fair amount of time and research goes into your numbers, along with a fair amount of mathematical analysis, and I'm not discouraging you from continuing that in any way - but it seems that it would only be right (and in the interests of full-disclosure and fair reporting) to make sure that the numbers on your site are sourced properly, and that any of your own numbers are clearly labeled as such.
Again, I think that you do a pretty good job with your numbers, and you obviously put a lot of time into trying to follow trends and project these trends into the future - but that is exactly what Media Create and Famitsu do as well, and they are professional firms that make their livings at this. From what you were saying, it sounds as though you apply ajustment factors to Famitsu's and Media Create's numbers - but they have already spent a lot of time and money researching the best adjustment factors that they already apply to the data. Their numbers are not 100% perfect, but they are as close as anyone is going to get. Media Create tracks more than half of retail in Japan, and has researched the best possible methods for estimating the remaining retail that they don't track. And Famitsu isn't far behind Media Create with their retail tracking. If these firms weren't very close with their numbers - then they wouldn't be paid huge dollars for their reports. And for the most part - Media Create and Famitsu track very close to each other. Dengeki is often a bit further apart with their numbers from Famitsu and Media Create - but I usually just focus on Famitsu and Media Create anyway.
Anyway, the point is that I think it would be appropriate for you to fully label and disclose where all of your numbers come from (including your own numbers) so that people who visit your site know just where the information is coming from.
I agree with the majority of what you say, and I agree that in the main Media Create and Famitsu tend to track pretty close to one another in terms of sales, especially with games that have large first weeks and drop off quickly so it makes sense that they are pretty accurate for many games.
Things have improved a lot over the years and I believe that Media Create and Famitsu sales are much much closer to the actual numbers now than they ever were in the past.
Have a look back over:
http://www.ga-forum.com/showpost.php?p=2483252&postcount=176
for some strong examples of the kind of disparities I am talking about. They do differ from game to game, but from the trends I have then the scale factors I use give a good first guess to the actual numbers, and they tend to be pretty close to the numbers that are given to us.
For example, Pokemon Stadium Gold/Silver back in 2001 did the following:
226,185 week 1
115,548 week 2
39,242 week 3
780,017 in total (stopped tracking about mid 2003).
Nintendo said in the CESA gaming report 2004 that upto Dec 2003 (again by this point you presume that the game is no longer in production / selling etc) that they shipped 1.14m.
So what I do is go back and say 1.14/0.80 (taking 800k from the Famitsu number plus a little for maybe surplus stock and so forth) = 1.425
So week 1 sales = 226,185 * 1.425 = 322,000
week 2 = 115,548 * 1.425 = 164,500
and so on (only I have first average Famitsu with Dengeki and MC so you'll find different numbers on Everything and Nothing) but the fundamental idea is that the sales are adjusted to match the shipment numbers given out by manufacturers. This is why I said anything over a year old on the chart is accurate (and by accurate I mean the best possible guess you'll ever get to how the game ACTUALLY sold based on fitting the weekly trends given by MC/Fam/Deng to the total amount of the game sold). Of course this is based upon the assumption that number of copies sold = number of copies shipped, but for a game released 5 years ago that has stopped selling a long time ago I think this is a fair assumption.
Using this data, there are some strong trends over the years, basically that MC/Fam/Deng have got better and better at estimating the sales as I said above, but for Nintendo games especially they are ~10% too low on average. So for a first guess I adjust number up by about 10% and then adjust this scaling factor when more accurate shipping numbers and so on become available.
The only numbers that I 'estimate' are when the game drops out of the top 30 / top 50 and this is usually based upon monthly info so if a game sold 8,000 in October and it's a 4 week month and then 6,000 in Nov which is a 4 week month also then it would be reasonable to suggest that weekly sales would be about 2 and a bit thousand at the start of Oct and a bit less than 1.5k at the end of Nov and drop between the two, so maybe 2.2k, 2.0k, 1.8k, 1.6k (giving 8k for Oct) and 1.6k, 1.5k, 1.5k, 1.4k for Nov (giving 6k). Alternatively if sales are across a month like Aug - Sept then there is always a slight boost around end of August so numbers will be 'estimated' to reflect this.
Either way, the numbers are being fitted to what we know.
Nothing is being 'guessed' as such, just filling in the blanks that you would normally have to pay lots of money to purchase the accurate versions of.
The idea of adjusting numbers by a factor (like the 1.425 Pokemon Stadium example above) may be more questionable to some people but as I said in the other thread what other way would you deal with the fact that Famistu have tracked 0.78m copies of a game being sold that Nintendo have shipped 1.14m copies of?
I do believe that the numbers on that site (certainly pre 2005) are a much much more accurate depiction of how games sold than the famitsu charts of the time showed. Famitsu only tracked ~5m copies of Pokemon Red / Green / Blue sold compared to 10.24m copies that Nintendo shipped in Japan. They knew they were miles out and just added random numbers to the game some weeks (the total would increase by 750k from one week to another) in an attempt to rectify the fact their tracking was way off.
Here you go:
Famitsu chart for two weeks ending 4th May 97:
http://www.sfoxstudio.com/sell/1997/19970421.htm
Famitsu chart for week ending 11th May 97:
http://www.sfoxstudio.com/sell/1997/19970505.htm
The first chart shows Pokemon on 1,789,550 copies and a week later it's on 2,457,181 after only selling 56,611 copies that week, a 611,020 difference that Famitsu have just added in (or more likely they have added 1/3 to the sales).
Notice also Mario Kart 64 on ~756k and Donkey Kong Country 3 on ~662k, a far cry from the 2.24m Mario Kart 64 shipped and 1.77m DKC 3 shipped.
Exhibit B:
Week ending 28th Sep 97:
http://www.sfoxstudio.com/sell/1997/19970922.htm
Pokemon on 3,419,775
Week ending 5th Oct 97 (week later):
http://www.sfoxstudio.com/sell/1997/19970929.htm
Pokemon is on 4,209,464 after selling 39,689 in the week between.
Famitsu have just stuck 750,000 onto the total !!
All of this was no doubt after Nintendo released press statements such as:
http://www.virtualpet.com/vp/news/111897b.htm
Reuters
09:12 a.m. Nov 18, 1997 Eastern
By Miki Shimogori
The sales of the Pocket Monster, or Pokemon, adopted for Nintendo's Gameboy player totalled 6.94 million units since its first launch in February last year and Nintendo said it plans to launch a new series of the software on Friday.
So 6.94 vs ~ 4.6m from Famitsu (and that's after adding on 611,000 + 750,000) so numbers were basically miles off.
As I say, these days Famitsu and MC are much much better (presumably larger coverage, much better scaling as you said etc) but they do still tend to be around this magic 10% too low, but obviously until these games stop selling and we have final shipment and final sales numbers we can't prove it either way *bookmarks thread*
