Using Google Trends to Predict Elections?October 22, 2008
What a day and age we live in. The vast amount of information and analytical ability we possess due to our digitalized environment can often times be overwhelming, but there’s some good stuff to be found. Some stuff we do (okay, at least the geekier among us) with the information we have is to play with it. No ends needed, no definitive goal or tactical reasons, just playing with data like we had with Matchbox cars as a kid. That sounds weird, doesn’t it? Eh, oh well, so I like stuff like that–it could be worse, I could be a plushophile.
Occasionally though, playing with data leads to something neat, as I think it did this time. There I am, playing with Google trends, comparing ‘sex’ to ‘Jesus’, and ‘Tootsie Roll’ to ‘Thoreau’ (sex is way more popular than Jesus, and surprisingly, Thoreau beat down Tootsie Roll), when I decided to plug in the presidential tickets. It seemed banal at first, but then I got to thinking. I downloaded the chart and opened it in Excel, made it into my own chart with linear trend lines, and viola! An instant explanation of the past month in the campaign.
The big peak that you see is the VP debate. That orange skyscraper is Palin being searched a ton, probably the searchers looking for some sort of rational explanation of what to expect, and the polling results of the debate. Biden went up by about the same factor, although his search volume both before and after just can’t compare to Palin’s.
Biden, no matter his ability to stuff his size 11 loafer deep into his mouth, has had relatively little search volume for the entire campaign. This is not only good, it’s great. Biden isn’t a sexy choice (although don’t tell that to my mother and her friends, apparently he’s hot stuff to the AARP girls), but he’s one that’s seen as completely safe to the voter. There’s no reason to search for Biden stuff, he’s a total pro with a resumé longer than da Vinci’s. The sheer boringness of Biden is his greatest strength on the Democratic ticket. No surprises.
Palin has peeks and valleys reminiscent of that Himalayan adventure I’ve never actually taken. She’s the most dangerous person I’ve ever seen on a campaign trail, and not to the opponent. She can’t go three days without some major, idiotic gaffe that a first year poly-sci student wouldn’t make. The VP needs more power, the VP is in charge of the Senate, I can see Russia from my snowmobile, the Bush Doctrine (“In what way, Charlie?”), and any number of other mind numbing comments make her like a blind moose on meth running through the entire Republican party with chainsaws for antlers–you know someone’s getting hurt, but you can’t help but watch. There is reason after reason to search for her on the Internet, and not too many of them are good.
Now, McCain versus Obama.
Senator Obama’s trend line looks like a plane taking off guided by a demure and responsible pilot. McCain’s trend is rising, but ever so slightly. The best news for him is that by the end of the campaign he should be more searched than his Vice Presidential choice. It’s called upstaging, John, and you’ve been upstaged by a half-wit former sports anchor from a small TV station in Alaska. That’s gotta hurt.
These trend lines seem reflected in the polls. Senator Obama’s support seems to be solidifying just at the right time, while Senator McCain, although not losing percentage points (in most polls), he is losing ground.
Here’s the RCP tracking averages for the past month, with an approximate trend lines added:
McCain’s trend line in the searches seems to be better than the polling numbers, and Obama’s polling numbers seems to be shallower than his increase in search volume. Now, one can reasonably expect that closer to the election, more searches will be performed on each candidate, so logically, the trend lines need to be ‘normalized’ to some degree. I could have actually done some math for this, but I decided to merely overlay the trend lines, and adjust the search lines so that they have approximately the same midpoint as the poll trends–about where 50/50 would be.
Lo and behold, the overlay:
Wow, that’s a close match. The search trend lines are still not in exact line with what the polls are saying, but I’d say that they match up better than expected. Can this be? Can searches from Google show the direction of the campaigns? Honestly, I’m not completely sold on the idea, but I’ll tell you what- that’s an interesting correlation. I told you playing with data can be fun.