Greek Elections September 2015 – closing the experiment

The third election in 2015 has come to an end in Greece making A. Tsipras the first Greek premier to have signed an austerity deal and survive politically. I used Google Trends data to monitor some conditions in the voter body. In one method (the naive method) I just used party names. It was clear to me that this approach clearly did not identify voter intend but just general footprint of each party in the urban chatter. Impressive here was and remains to be that Golden Dawn was really high dueling with SYRIZA for first place. They are now the third party in the new parliament but one thing is for sure: they are no longer marginal.

The second method (the ballot method) arose in the last week and was using “ballot <partyname>”. This did a better job at identifying voter intend and using it I could call SYRIZA at 36% 7 hours before voting ended. SYRIZA finally received 35.47% of the vote with 99.44% of the total counted. Not bad. There are discrepancies for ND and PASOK but the rest was more or less right. I do not know why this method showed PASOK that high.

Both methods suffered from selection and identifications issues. For example some parties have only an acronyma (e.g. KKE or SYRIZA) others have both (eg ND or Nea Dhmokratia, XA or Xrysh Aygh etc). The longer the name the less accurate its footprint it seems. When the names are very short they are ambiguous. The acronym XA in addition to Golden Dawn also identifies the Athens Stock Exchange for example and ND identifies shorthand for state laws in addition to the conservative party. Data was low in all cases.

The bottom line is that provided there is enough data and a good identification strategy the method works very well. I may come back and try it in the upcoming US election, so stay tuned.

This slideshow requires JavaScript.

This entry was posted in comment, data, nowcasting, politics and tagged , . Bookmark the permalink.