On Sunday December 20 2015, the Spanish electoral body is voting for a new government. Concha Artola, economist at Banco de Espana and myself are going to try to monitor the electoral process using hourly Google search data and applying the same technique as in my referendum papers (1, 2).
We use three methods to measure the footprint of each party in Google search.
- Simple Names: podemos, ciudadanos, iu,pp, psoe
- Mixed Names: podemos, cs + “Ciudadanos”, iu + “Unidad Popular”, PP + “Partido Popular”, PSOE+”Partido Socialista Obrero Español”
- Programa: programa podemos, programa ciudadanos, programa iu, programa pp, programa psoe
The graphs are shown below:
The “mixed” and the “simple” methods show PODEMOS well ahead of C’s and the “programa” method shows a close duel between the two.
The graphs are a departure from traditional polls.
Discussion: We used our intuition to choose the search strings. Ideally we would like to have the following being true:
- A) Search strings are relevant to the party and the electoral process.
- B) Search strings identify voter intent for the party they try to locate information on.
From a previous exercise we know that searches for a party do not necessarily imply voter intent. Due to the simplicity of choices (duality of yes/no) and polarization this was true in my referendum papers both in the Greek and in the Irish case. Subsequent attempts to apply this to elections (Portugal and Spain) showed that applying this identification strategy to multi choice elections has limitations which we may not always be able to lift because we are restricted by the Google Trends nomenclature. For example there is a bias introduced by the length of party names: if the names are short they may be identifying other things in addition to the party and if they are long then they may not be typed as often… Such was the case with the conservative party in Greece (ΝΔ) whose short name is also used to talk about laws in Greek. In the Spanish case an attempt to use iu+up+”Unidad Popular” for the corresponding party contaminated the results with searches for “pin up”, “step up” etc so we had to revert to using iu+”Unidad Popular“. The programa method is good at achieving A but not necessarily B.
We notice that among the top additional words in each case are also the names of other parties. This means people are still trying to inform themselves as in searches like “programa podemos”. An attempt to disentangle these (i.e. use “podemos -cs -iu -pp -psoe” instead of simply “podemos”) as in the Greek referendum paper diminished the series to zero. So currently we cannot do so. This is either a malfunction of Google Trends or it simply means there are not enough searches for any party without mention of some other party as well. If so we are in a dynamic and exploratory stage. This may change as we are approaching the election day.
Modulo the issues mentioned above we feel that A) above is by and large satisfied but not necessarily B). is all three methods. For what it’s worth though the footprint of parties in search is depicted in these graphs.
Looking at longer term graphs below we see that the splits of searches for the four major parties between being searched alone and with others are widening as we approach elections which means that voters are increasingly more actively in the market for political representation. The smaller the gap then the more convinced the voters of that party are but also the less that party is being viewed as an alternative by defectors from other parties.