Google Insights is so great it can predict the outcome of the 2008 election! Or so I think

Unlike previous attempts at predicting, based only on the number of searches for a certain candidate’s name (see Techcrunch for the tale of such an experiment), I used Google Insights’ archives of the 2004 election.

There was a very strong and very reliable correlation between the searches for ‘John Kerry’ between January and July, 2004, and the actual outcome of the election on a state-by-state basis. All I did was to use the 2004 equation (right) with 2008 data concerning the searches for ‘Barack Obama’. Now, the Google interest index is based on relative search volume. That means that in states where people Google a lot, searches for Obama are relatively less important. So I corrected the first results by looking at the difference between searches for McCain and Obama.

And… Obama wins!

Now, all that isn’t very serious (come on, Michigan, you can’t vote for McCain!). John Kerry did not arise as much curiosity as Obama does now, so that Obama searchers may not be Obama voters. Plus, Obama voters are likelier to own a computer and internet connection, overstating his victory margin. We’ll see how relevant this model was on November 4.

But that shows how powerful Google Insights could be, once again (more examples, in French, here and here).

Everybody’s talking about Google’s venture into content-production (there are only 600 blog posts on ‘Google Insight’ – 20 times more for ‘Google Knol’), even though we all know Google Knol’s bound to fail.

Google search history is where the value is. Google knows what we want. We tell the machine our every wish, every day. We do so without the inhibition a pollster can induce, so the data is more reliable.

If Google doesn’t kill pollsters right away, it’s going to sell them data. This will bring in much more cash that Google Knol ever will.