Nilatai wrote:
You do realise that would lead to more bias than random sampling, right?
No, it wont. That's why political polls now say "among likely voters". As I stated pages ago, there are pro's and con's to each form of sampling. There isn't no one best way of sampling, but for a polling like this, random isn't the right answer.
How do you not see that?
If you poll 1000 people on President Obama's approval and the majority of the poll consist of Democrats, his approval rating will be higher than if the majority poll consist of Republicans. The only way to get a fair observation of his approval rating is to have a poll closely relating to the actual population.
If your subset doesn't represent the total population, then the conclusion of that sample represents another population.
That's like asking 1000 people (750 men/250 women) on their opinion on who should pay for dates or any other relationship question. That's not a fair analysis. It doesn't necessarily have to be 50/50, but it should be as close to reality as possible. Else, you're just making conclusions from a made up population.