NPD Sales.

[NJ5]

HappySqurriel said:

One thing to consider that I see few people ever mention is how the sales numbers tend to be estimated ...

Most statistics people learned in school use the assumption that the data set they collected is a perfectly random sample, and thus the bias or error in their dataset is based on the probability of (randomly) having a disproportionate number of responses in your dataset; this is the main reason why the assumption is that your dataset becomes more accurate as your sample size improves. As a demonstration, the probablility flipping a coin 4 times and getting heads 3 out of 4 times is fairly good, at the same time the probability of flipping a coin 400 times and getting heads 300 of the 400 times is not very good.

When you're dealing with a statistic like VGChartz and NPD are, you're forced to make the assumption that your data set is not a perfectly random sample and you will be forced to compensate for the bias in your numbers. Consider for a moment that the type of person who buys their games and consoles at Best-Buy/futureshop is quite a bit different from the type of person who buys their games at Gamestop/EBGames or Walmart.

Now ... Being that a large portion of both VGChartz and NPD's numbers are based on projections (which are, essentially, based on assumptions) it is quite possible that either of the numbers can be off. If NPD (for example) underestimates the sales of the Wii at Walmart, and VGChartz (for example) overestimates the sales of the Wii at Walmart, both numbers can be off by quite a bit and they (certainly) wouldn't match up.

That is probably what ioi means by "retailer bias" in his methodology page... IMO, it must be one of the hardest problems in sales tracking. Another one is that stores often do promotions which will significantly alter consumers' behavior, screwing up assumptions about retailer bias.