Here’s to you, dumb-ass reporters.
Yeah, that’s right.
Today’s dumbassery comes to us thanks to the USA Today/Gallup poll, and more specifically the coverage thereof. In a stunning twist, so the poll says, John McCainnow has a 54-44% lead over Barack Obama. That ten-point lead is among those defined by Gallup as “likely” voters. Among registered voters, they claim McCain leads Obama 50-46%.
Wow!
Wait. Hold on, dumb-ass reporters. Let’s not go sucking McCain’s old dick just yet.
You see, even with the best sampling frames, even if you could achieve the Holy Grail of sampling (Simple Random Sampling, said with heads reverently bowed), then, if you have a 3% margin of error, as Gallup says they do, then you’d still get a fucked-up result, on average and with respect to true population values, in three polls out of every one hundred taken. See? That’s because of something called random sampling error. A sample, of course, doesn’t measure what everyone in the population thinks. It can only estimate those unknown population parameters. Done correctly, samples are a very powerful way of estimating population parameters from a (relatively) few number of people. No, really, it’s quite amazing. However, because of the way combinations work, you will get some inaccurate sample results, just due to random sampling error. If you took every possible sample of size 1,022 out of the population of millions of registered voters (which would be impossible–there areway more combinations there than there are protons and neutrons in the universe), eventually, you’d get a sample of 1,022 people who are all Republicans. It’s unlikely, but it’s bound to happen. Even more unlikely, but still bound to happen, you could get a sample of 1,022 people who were all named “McCain.” Assuming, of course, that there are at least 1,022 registered voters who have that last name. The point is, given enough fucking iterations, you would eventually draw every sample that is possible. Most of the time–indeed, the vast majority of the time, those samples will have statistics that are pretty close to the population parameters they’re esitmating; that is, they will be representative samples. But every once in a while–and you have no fucking idea when this will happen–you’ll get a nonrepresentative sample. Oops. When that happens, the polls are considered outliers. But you can only make that determination, for an SRS, if you have lots of other polling data.
Head hurting yet?
Just bear with me.
But Gallup, nor any other nationwide polling firm, never conducts simple random samples (it’s not that they wouldn’t want to–it’s just that it’s impossible). Instead, they construct sampling frames. And this article doesn’t tell us just how they constructed that frame. Did they include more Republicans? How did they weight or oversample different groups? Without that information, it’s hard to accurately assess whether or not this poll is simply an outlier, or if the sampling frame itself is fucked. Or if there’s some combination of those two things going on.
However, what we can do islook at this:
Hmmm. Notice anything?
So, yeah. I think we might have an outlier here.
Not, of course, that any of the news outlets are reporting that.
Oh, fuck no.It’sapost–convention/SarahPalinbounce!
You fucks. You don’t get a seventeen-percentage-point swing (just in the Gallup poll) in a week. Not without something extraordinary happening–and I’m not talking about that coma-inducing GOP convention, either. How about you, I don’t know, look for a little context for your goddamn stories? Or would that be biased?
And you people thought all I was good for was swearing and pretty pictures.