Spanish Elections and Google Trends

The Spanish Elections are over and the results show it is a new situation for Spain in the sense that the dominance of PP and PSOE is now a thing of the past. Good luck to Spain for the difficult road ahead.

Concha Artola, economist at Banco de Espana and I tried to monitor the vote via Google search as I had done for the Greek and Portuguese elections in the past based on a method which worked very well to forecast the Greek and Irish referendums(See my papers: 1, 2).

We knew that while we identified the parties well our method was not and could not have picked up voting intend. The final results verify this. We would see Podemos and C’s high above PP and PSOE with Podemos being ahead of C’s. The reason is of course easy to imagine. PP and PSOE are old and well established parties and whatever is left of their following does not search for their program: these are die hard afficionados. Podemos and C’s are comparably young so people search to get to know them and so their searches are comparable. The fact that we could see Podemos ahead of C’s is then a prediction of the method. What our method could not do is to forecast the combined losses of PP and PSOE to Podemos and C’s.

I had observed similar failures when I tried doing this for the Greek and the Portuguese elections. This is not the case for referenda where both identifying the yes and the no is relatively simple but also both the yes and the no are equally “young options” (which people need to inform themselves about) contrary to PP being much older than C’s for example.

Perhaps in future exercises we can try weighing the searches by the length of a party’s history.

And Concha’s comments below more or less in line with my own assessment:

The first graph below shows the average of polls up to December 14th (Under Spanish law no polls are allowed to be published 5 days before the election)

The second graph shows our Google indicators as published on December 19th.


My reading is that we couldn’t track voting intentions for PP and PSOE properly while we got it right, much better than standard surveys, in the final results for Podemos and Ciudadanos. And the most important thing is that this trend was clearly visible since early December.

The reason of our failure in tracking PP and PSOE voting intentions is most likely due to demographics (age, education and may be income). A large fraction of voters to PP and PSOE are not young, and they might be less likely to be in the digital world.

Bottomline: Google is useful to get trends right but only among the people that do searches in the Internet. In other words Google cannot find your key unless you lost it by the lamp pole. Quite obvious!