The inimitablehappier_bunny has told me some of ya'll are freaking out about polls and Obama's chances.
To this I must say:
I know something about polling methodology and statistics. This is because I possess two virtually useless degrees in sociology* (including a master's), and, as such, I'm thousands of dollars in debt.**
I would never hold myself up as an expert pollster, but I've done graduate-level study in social science methodology (which includes the creation of questionnaires used in polls) and statistics (which includes the interpretation of a poll's results). I'm not an expert, but I think it's accurate to say that I know a damn bit more than your average bear.
There are some key things everyone should know about polling.
Sampling
The population is the group you want to study. In national electoral polling, the population is all of the individuals who will vote in the 2008 election. Because we can't poll every single person who is going to vote in the 2008 election, social scientists use sampling to approximate the population.
In order for the poll to be accurate, the sample has to contain the same variations that exist in the population. We can poll 1,000+ people (the sample) and make estimates about tens of millions (the population), as long as those 1000+ people are really representative of the tens of millions.
Barack is hoping to make sampling horrifically hard for pollsters. He's trying to register tons of new voters, often young'uns, and will probably dramatically increase the number of African American voters. These new voters aren't on the pollsters radar, which means they aren't being included in the sample, and the poll is flawed if the sample doesn't reflect the population.
Margin of Error
Most people here are probably familiar with the term "margin of error," but I'm not sure that everyone knows just what it is and what it's measuring. No one anticipates that pollsters will get the sample to be 100% reflective of the population. The margin of error measures the sampling error in the survey. It's the little "+/-" number that shows up on all the polls. It's expressed with a confidence interval. In social sciences, the confidence interval is usually 95%.
This is just a fancy way of saying we're 95% sure the actual population falls within a certain range. For example, if McCain is polling at 50% and Obama at 48% and the margin of error is +/-3, we can be 95% certain the actual population supports McCain somewhere between 47-53% and supports Obama somewhere between 45-51%.
When the difference between the candidates' poll numbers is smaller than the margin of error, it is said that they are polling within the margin of error. In the scenario I just gave, Obama and McCain are polling within the margin of error and it's possible that Obama actually leads McCain 51% to 47%.
In these circumstances, people (myself included) like to throw around the phrase "statistically tied." Unfortunately, it's not that simple. The numbers that are reported (in our example, McCain 50% and Obama 48%) have a greater chance of being accurate than any of the other possibilities. Saying something is statistically tied is convenient shorthand for acknowledging that the sample may be flawed, but it's not really accurate. If anyone wants me to go into the reasons why the reported numbers are more likely to be true, I absolutely will, but be warned that it gets technical and involves parabolas!
Polls are never, ever predictive
I'm going to repeat that. Polls are never, ever predictive. When a poll is completed, it is only a snapshot of the electorate. Polls can be made obsolete overnight. Polls from a week after the Republican National Convention will have no relevance whatsoever on the electorate a week from now. And remember, polls take time to complete; a poll that is released tomorrow will contain people who were polled a few days to a week after the RNC was completed. Blogs live in immediate feedback world, but polls do not.
So, the good news: a snapshot of the electorate shows McCain is up. That shouldn't be surprising. The RNC completed a little over a week ago and candidates often get bounces from their conventions. But a poll from mid-September is going to have nothing to do with where were are come election day.
And the bad news: a snapshot of the electorate shows McCain is up. I'm not denying the accuracy of the current polls. I'm just saying don't let it freak you out and don't rest on your laurels. The polls will change as the circumstances surrounding the election change.
That means donate and volunteer. Make calls, help fund ads, register voters.
Help Sen. Obama kick some ass!
*I'm kidding about the virtually useless; I really like my major and do use it. ** I am totally not kidding about the debt.