This week’s Science of Search column at TIME focused on the popularity of Congressman Ron Paul (R-Texas) online. We’ve known from past experience that you can’t predict elections based on search data (due mostly to the multiple intents behind political searches). Yet I’ve been puzzled over the last several weeks about the gaps between searches for candidates and their popularity in pre-elections polls.
Here’s the latest chart on top Republican candidates (note Thompson’s surge this last week coinciding with his announcement to run).

As I discussed in the column, perhaps the difference in polls versus searches could be based on demographics. Visitors to candidate’s websites that rank highest in search volume also skew younger which happens to be a demographic difficult to capture through phone polls.