Wednesday, September 15, 2010

What Anomalous Data Can Teach Us

Last week I wrote that it would be foolish to assume the bump in favor of Democrats in last week's Gallup generic congressional ballot meant anything. Sure enough, here's this week's Gallup poll, which the Democrats don't seem to be ballyhooing quite as much:

Registered-voter preferences for Congress since the beginning of August have averaged 48% for Republican candidates and 43% for Democratic candidates, identical to this week's results. While there have been a few instances in recent months when the Republicans were not ahead to at least some degree -- including in mid-July, when the Democrats were up by six percentage points, and last week, when the parties were tied at 46% -- the broad picture has generally been positive for the Republicans.

Generic Ballot Splits 48% for GOP, 43% for Dems

That's Gallup for "Last week was an anomaly. You weren't silly enough to believe it, were you?" Still, it seems that plenty of people were. Here's this week's graph:
Image credit: Gallup poll graph converted to PNG format by Cujo359
[click on charts to enlarge]

So, last week's result was the anomaly, and the previous week's record-setting Republican advantage was just a bit exaggerated. Go figure.

That high pitched sound you hear is Democratic politicians whistling past the graveyard.

Speaking of silliness, as I wrote in the update to last week's poll article, I made a rather silly mistake. The numbers I pulled from Rasmussen for their generic congressional ballot were all in reverse chronological order, i.e,, newest data point first. That made the graph reversed, too, and made me misinterpret.

Since then, I've played a little with the spreadsheet, and now it generates a graph similar to the Gallup chart, including the time span and color selection:
Image credit: Plot by Cujo359 from Rasmussen data

The spread in Rasmussen's data is now 9 percent in favor of the GOP, down from 12 percent last week.

What the combination of these two polls suggests is that there was a spike in the Republican's popularity two weeks ago, and that it may have abated a little. There's certainly reason to think that when you look at how Republicans have been behaving recently. I've made the point often enough that they're the only ones standing in the way of their controlling the House next year. They've given me no reason to think they still can't accomplish the miracle of losing to these Democrats.

Still, the Democrats are in big trouble. Both charts show enough variation to suggest that they still could lose in record-setting numbers to Republicans this fall. The enthusiasm gap is as big as ever, and the Democrats haven't done anything to close it besides trying to makes us more frightened of Republicans than we are of them.

There are some lessons we should learn from this:

  • Don't believe politicians when they're talking about polls.

  • Don't believe the people who support those politicians when they're talking about the polls.

  • When discussing any information like this, one of the most important things to ask is: What's normal? What is the normal variation in poll data from one time to the next? What change is evident in the long term? All of these things are important for knowing what a change in a given data point is likely to mean.

  • The chart is not the data. It's a way of making sense of the data. Understand what it's saying.

  • Mistakes are opportunities to learn something.

  • In the end, it's really up to you to use your own skepticism about everything you read, because even people who don't have an axe to grind can be wrong.


As many bloggers do, I try to always document where I obtained data from polls, and to explain any assumptions, guesses, and methods I use to draw conclusions from it. That's how it should be, and if there's something that isn't clear from that conclusion, then it's OK to ask, or to be skeptical. In fact, I really expect the latter.

No comments: