Well it does appear as Source has mentioned that VGChartz estimates have improved. Not only has the average difference declined for the last 3 months as he mentioned but, as that would indicate, the individual differences are improving.
In May the site had 2 systems within 10% of NPD and 2 off by more than 30%.
In June it was 3 systems within 10% and 1 off by more than 30%.
Last month it improved to 4 systems within 10% and none off by 30%.
Not as valuable as per system sales but total system sales have improved markedly from being 7% off to 4% to spot on this month (not including GBA). So if the site can keep getting the total number right then it's just a matter of getting the shares correct, admittedly the tougher and more important of the two.
Now the bad side. Although average error improved last month a lot of it was due to the GBA moving from way off to almost exactly right. Remove the GBA and and the average error was slightly worse (14.8% vs. 14.5% last month). Also despite the improvement in average error the variance increased. That's because the ones that were right were more right but the ones that were wrong were more wrong.
The only systems VGChartz seems to have issues tracking are the DS and 360. Not counting the GBA they account for almost half of the average percentage error over the last 3 months (7.4% of the 15.4% average error). Which since there's only 6 systems that would of course mean those 2 typically have twice as much error as the other 4.
Anyways, so long as VGChartz doesn't backslide its doing a pretty good job for a free service in my book.
