a big sample size needs a lot more math to get to beeing close to represantive.

you need a random sample, thats realy hard to get if you track a big none random part.

Ah, no. Sure, you can do your modeling so wrong, that your better sample turns worse data, but that's a problem of the model not of the sample. Also it is possible that a smaller sample better covers more different groups of (in this case) consumers. But we have no sign to believe the VGC is better. If you don't know more, it is best to assume the bigger sample is also the better sample. You can argue the NPD sample has systematical error, because they don't get some user groups. But for that you can model (having other user groups in your sample still is better). But it's bordering on impossible that VGC has the more diverse sample.

the most people here take the numbers as facts. modeling leads to errors in most cases.  and we dont know how good vgcs numbers are because we know shit about em.  and we need some more informations about npd too...

i dont think its such a good tracker that you shoudl take the numbers as facts, those are just sophisticated estimations....