Jsmith - Most reputable polls use some form of random sampling in deciding whom to call. They also take demographic information about whom they reach, and combined with demographic data for the region they're polling, they'll weight the results. I've read (I believe at www.pollster.com that polling firms never release raw data, they always weight it, trying to balance race, gender, party affiliation and perhaps other variables. And different firms do differ in how they weigh things, which often explains much of the difference in their reported headline numbers.

I think there are several possible explanations why current polling shows McCain beating Obama more often than not:
1. Obama is certainly well known among Democrats now, but likely less well known than McCain among independents and Republicans, but they're also sampled for general election polls.
2. The general public perception of McCain is still of a "maverick" (i.e. not Bush) Republican. So he is not as adversely affected by the administration as he likely will be by November.
3. I do think the negative attacks of the primary have hurt Obama.
4. It's quite possible that many, or perhaps most, pollsters have not adjusted their weightings to account for changes in voter registration, relying instead on older data.

I suspect a little of all of them may be at work, as well as other factors I'm not thinking of. The relative good news is that once the nominee is selected and the general election campaign begins in earnest, the first three factors should get better for Obama. And the fourth is irrelevant for the election itself, it just affects how well pollsters will predict the election.

So I'm still overall optimistic that Obama will win in November, but the current polling suggests McCain is ahead right now, and it's something to be aware of. I'm confident David Axelrod is already well aware of all these numbers, and is working hard on improving them.

If you have a truly random sample from a population, you don't even need to poll 1000 people to get a 4% margin of error. And it's a heck of a lot cheaper to poll 500 to 1000 voters than to poll 5000 or more. And the added accuracy from a larger sample size generally isn't viewed to be worth it, especially when opinion can and does change quickly. You're better off doing a lot of relatively small sample polls with a bit wider margin of error than doing fewer larger sample ones with a smaller margin of error.

Now whether the sample is truly random or not is a question pollsters spend a lot of time analyzing. In the mid-20th century, a major pollster famously blew a presidential election (I think it was Truman/Dewey in 1948) because they did a telephone only poll. Most people at the time did have phones, but those that didn't were much more likely to vote for Truman than Dewey, so while the sample was a random selection of phone owners, the bias of ignoring non-phone owners caused them to blow the election.

The other big variable in primary polling this year has been turnout models. Especially in primaries, a big part of getting good polls is figuring out who is actually going to vote. A year like this, with a lot of new voters, is tougher for pollsters to get right, because the composition of the voting population is changing, and if their weightings don't reflect that, they can more readily miss the results.

I still think the main reasons Obama is a little behind right now are likely to be corrected as this becomes a two-person race. Once he's only being attacked by McCain, and once he's focused on making the sharp contrasts he has with McCain, I think Obama's numbers will improve sharply. But now, I'd guess a that to a sufficiently number of independents and even Democrats who haven't been moved to vote in this year's primary, Obama is that black guy whose name sounds vaguely like Osama, whereas McCain is the "maverick" Republican who distances himself from Bush. Those perceptions will rightly change in the fall, most likely to Obama's benefit.