This octopus guys say they went to all 47 counties.
Actually a sample of 2000 is good enough to determine (with say 95% confidence level) error margin of 2% the sentiments of whole population on a certain issue.
Increasing the sample size does not yield any significant improvement in error margin. The trick is making sure that the sample is random and representative.
Politics aside, when these opinion polls are conducted professionally they can help the candidates fashion their messages to target audience. A candidate can hire the pollsters to determine their appeal to say a certain age group/geographic over and above the national opinion.
"..I am because we are. "― Ubuntu, Umtu,