You might want to research how online polls can be slammed by one and only one person with the right skill set or mere access to msdos/linux shell scripts.
That's probably true of some polls--namely, the ones coded by incompetents. Any web developer worth his salt will log IP addresses and forbid multiple votes from a single IP, or if possible will require a logged-in user with a unique email address in order to vote. Those measures can be circumvented, but not by your average script kiddy.
It's no secret that RP supporters "Freep" political polls. When I see a link to a poll, I often pop over and vote for RP myself, and I'm sure others do too. I don't vote multiple times, but out of curiosity I've cleared cookies and reloaded, and when I've checked, I get the message, "Your IP has already voted" OWTTE.
On the other hand, I'm sure that "Hillraisers" also "Freep" polls for Billary. It's impossible to say whether the result is a wash or not, but you would generally expect the results to be vaguely correlated with the candidate's actual popularity: if Giuliani supporters outnumber RP supporters by 50:1, and a poll is posted to forums where Giuliani's and RP's respective supporters hang out, you'd expect the proportion of folks rushing over to vote would be somewhere in the ballpark of 50:1 in Giuliani's favor. If you firmly believe that some hacker is voting multiple times, then you'd expect pro-Giuliani hackerbots to outnumber pro-RP hackerbots by roughly 50:1, and again you'd expect a relative wash.
Since that's clearly not happening, one might cast about for an explanation. Keeping the made-up ratio of 50:1 for Rudy, and supposing that RP beats Rudy by 3:1 in internet polls, the following are some possible explanations:
1) Perhaps the average RP supporter is 150x more likely to go vote in a poll he sees posted online. (This seems doubtful, though.)
2) Perhaps skilled hackers are 150x more likely to support RP than Rudy. (I can't begin to guess whether that's plausible or not.)
3) Perhaps each RP supporter, on average, votes 150 times in Internet polls for each Rudy supporter's single vote. (They must have lots of time on their hands.)
4) Perhaps voters who support Rudy are 150x less likely to use the Internet in the first place. Maybe they're all teamsters, and after a hard day at the loading dock they're just too worn out to bother going online.
5) Perhaps Rudy supporters aren't even reading political sites. Maybe they're all at MySpace hooking up, while every single RP supporter is busy reading LewRockwell.com and spamming Digg. So maybe the RP supporters are 150x more likely even to know the poll is out there.
6) Perhaps Internet users are much more likely to be libertarian than the average person. This is a variant of #4 above, more or less.
7) Then again, it's possible that the relative support isn't 50:1 after all.
The disparity between online unscientific polls and offline (presumably scientific) polls is too great to be ignored; something is going on there. But I would hesitate to offer glib explanations. I suspect that #5, #6 and #7 all play a part in the true explanation. Proving it would be tough. Translating that into election predictions would be even tougher; we'd have to figure out the relative likelihood of Internet users voting, compared to the general population, just for starters.
--Len.