| UncleScrooge said: Research companies are trying to avoid picking a stupid sample, though. Let's say in a phone surves they accidentally call 2000 males at the age of 40 with roughly the same income. Obviously that would affect the survey's results. But research companies know preferences of certain demographics and social groups. For instance prior to the 2005 elections in Germany polls showed the conservatives polling around 45%. They only got 35% of the votes, though. The reason for this was not random polling, though (because research companies know about this and know how to meassure a screwed sample). They had simply underestimated the number of swing voters. As long as you know what you're doing there is no "random sample". VGChartz does the same. They don't get data from all stores so they have to calculate store preferences, obviously. Just like Media Create and NPD. |
That is just not how random sampling works. You can't specifically exclude anyone or else there is a very, very good chance that the results will be invalid. There is the notion of forcing the sample to match the demographics, but that is not as reliable as pure random sampling when it comes to generalizing the results. If you get a bad sample like the above you simple toss it all out and try again. Laws of probability dictate that you are likely to get a random sample on any given attempt so it is not that much more work. The only catch is you have to be able to know the sample was biased which in this case would not be immediately obvious.
