Can Google Trends Predict The Election?


If more people are searching on Google for “Obama” than “McCain” does that mean he is more likely to win the election?

Not every voter in the U.S. uses Google, or even uses the Internet, for that matter. But enough of the population does use Google that its search patterns cannot be ignored by either candidate, the press, or anyone interested in the outcome of the election. Fortunately, Google lets anyone see the relative popularity of different search terms on Google Trends. The screen shot above is from a comparison I just did between “Obama” and “McCain” in the U.S. over the past 12 months. If search volume is predictive of election results and the elections were held today, Obama would win.

More traditional polls come to the same conclusion. A CNN-Opinion Research Corp. poll conducted June 4 and 5, shows Obama ahead by 47 percent to McCain’s 43 percent (Ralph Nader has 6 percent). And a Gallup poll shows a tighter race, with Obama at 46 percent verses McCain at 45 percent. Obviously, it is a close race and sentiment can go either way between now and November. And there is a likely correlation between search volume and news mentions, which are also compared in the graph above (by pulling in data from Google News).

What is great about Google Trends, though, is that you can drill down by state. An in-depth analysis of how predictive Google Trends was during the primaries (by Michael Giuffrida, a student in Virginia) shows that in at least half the cases for the Democratic primaries, Google Trends did a good job predicting the outcome. Update: Just to clarify, the analysis looks at both Democratic and Republican primaries. For the Democratic primaries 37 states were analyzed, and five of those had to be thrown out because of insufficient data. Of the remaining 32, Google Trends correctly predicted 27 of the primary elections, or an 84 percent success rate. For the Republican primaries, 29 elections were analyzed and Google Trends correctly predicted only about half (the data wasn’t as good for a variety of reasons).

Below are two of his comparisons of Google Trends and actual election results in Missouri (where Obama won) and Florida Nevada (where Clinton won). Google Trends appears to be more predictive the higher the search volume (i.e., the more data points). Some states had more searching than others, but you’d expect election-related searches to spike across the board as the general election nears. At the very least, both campaigns would be wise to use it as a sanity check on their own polling on a state-by-state basis, if they are not doing so already.

(via Slashdot).