There is a bit of a bias from the fact that there are more registered Democrats than Republicans (about 46% of population vs. 23% for Republicans), so given a random sampling of the population, the numbers work. I don't have the numbers, but I remember hearing that Democrats were also more likely to take part in these polls overall.