An Interesting Take of the Delphi Method

The Delphi Method is that people in aggregate make better predictions than the experts.

This can apply to a group of experts, or it can apply to a large sampling of the general population.

Well, someone just applied this methodology by using Google searches:

Google has published details of the levels of search interest for political candidates as election day approaches. But there seems to be little clear connection between Google hits and the ballot box.

The main advantage the web has for tracking patterns is the sheer number of people using it. For example, it sounds insane that anyone wanting to find the Ultimate Fighting Championship (UFC) website would do anything other than try ufc.com as their first tactic. Yet the number of people who search for “UFC” and related terms in the days leading up to a pay-per-view event is nearly always a good indication of how many people will buy the event. The reason is that even if only a tiny fraction of people actually have to search for the term, it still adds up to enough people that variations over time are measurable.

Does this work for politics? Well, here’s the current levels of Google search interest among web users from the relevant state in three notable Senate races, along with predictions for the winner from polling analysis site fivethirtyeight.com (source of the image above). The first figure shows each candidate’s current share of the total searches for the listed candidates, while the second is the predicted share of the vote.

Florida: Marco Rubio 45.4/44, Charlie Crist 32.9/32, Kendrick Meek 21.7/24

Nevada: Harry Reid 54.4/47, Sharron Angle 45.6/50

Pennsylvania: Joe Sestak 51.1/48, Pat Toomey 49.9/52

I have no clue as to whether or not this is a valid technique, but if it is, Reid will be reelected, and Sestak will beat Toomey.

I guess that we will find out tomorrow, but I am inclined not to believe that this is a valid predictive technique.