Just do what Nielson does: Pay people for their data.

I don't think it's that difficult; if pay-for-survey skews results toward overvaluing the opinions of the poor and/desperate-for-money, well, then it would be the first time in history.