Introduction
Using Google Trends to predict elections and other phenomena
Predicting phenomena with Google Trends
Predicting elections with Google Trends
Study | Election(s) | Width(s) | Distance | Categories used | Data type used | Multiple GT datasets | Explain / mention search term selection |
---|---|---|---|---|---|---|---|
Peer-reviewed | |||||||
Raubenheimer et al. [22] | 2020 New Zealand cannabis referendum | 1 week; 3 months | 1 day before event | No | Hourly daily | Yes | Yes |
Prado-Román et al. [3] | 2004-2016 presidential elections US and 2004–2019 presidential elections Canada | 1/2/3 month(s) | 1 day before event | ? | Daily | No | No |
Mavragani and Tsagarakis [18] | The 2014 Scottish the 2015 Greek the 2016 UK Hungarian Italian and the 2017 Turkish Referendum | 1 week; 1 month | 1 day before event | Yes (no further explanation why) | Hourly daily | No | No |
Mavragani and Tsagarakis [23] | Greek referendum 2015 | 8 days; 1 day; 12 hours | 1 day before event & N= 0; on Referendum day | ? | Hourly | No | Yes |
Polykalas et al. [19] | 2005-2013 German Elections (only parties CDU/SPD) | 1 month | 1 day before event | ? | Daily | No | Yes |
Harkan and Eryanto [24] | 2019 Indonesian Presidential Election | 23 Sept 2018 - 16 May 2019 | 1 day before event | No | weekly? | No | No |
Polykalas et al. [14] | Greek elections 2007-2012 & Spanish national elections 2008-2011 | 1 month | 1 day; 1/2 week(s) | ? | Daily | No | Yes |
Yasseri and Bright [25] | Iranian election 2013 German election 2013 UK election 2010 | ? | ? | ? | Daily? | No | No |
Granka [6] | US Presidential Elections 4’ 8’ and 12’ | 1 month | 2 months | No | weekly | No | No |
Vergara-Perucich [26] | Chile: Presidential election 2006–2021 | 121 days | 5 days | ? | Daily | No | No |
Non-peer reviewed | |||||||
Sjövill [20a] | Swedish general election 2010-2018 | 1/2 weeks; 1 month | 1 day before event | No | Daily | No | Yes |
Wolf [15] | US Presidential Elections 16’ swing states | 2/4 Days | Depending on poll data | No | Daily hourly? | No | ? |
Askitas [17] | Greek referendum 2015 | 1 week | 1 day before event | No | Hourly | No | Yes |
Askitas [16] | Irish “Gay Marriage” Referendum 2015 | 1 week | 1 day before event | No | Hourly | No | Yes |
Data and methods
Google Trends as a data source
Search terms and category filter
Category filter
Search terms
Election date | Search queries |
---|---|
2009-09-27 | CDU + CSU + Angela Merkel; SPD + Frank Walter Steinmeier; Grüne + Jürgen Trittin + Renate Künast; Linke + Gregor Gysi; FDP + Guido Westerwelle |
2013-09-22 | c(“CDU + CSU + Angela Merkel”, “SPD + Peer Steinbrück”, “Jürgen Trittin + Katrin Göring Eckardt + Grüne + ”, “Linke + Sarah Wagenknecht + Gregor Gysi”, “FDP + Philipp Rösler”); c(“Bernd Lucke + Afd”, “CDU + CSU + Angela Merkel”, “SPD + Peer Steinbrück”, “Jürgen Trittin + Katrin Göring Eckardt + Grüne”, “FDP + Philipp Rösler”) |
2017-09-24 | c(“CDU + CSU + Angela Merkel”, “SPD + Martin Schulz”, “Cem Özdemir + Katrin Göring Eckardt + Grüne”, “Linke + Sarah Wagenknecht + Dietmar Bartsch”, “FDP + Christian Lindner”); c(“Alice Weidel + Alexander Gauland + Afd”, “CDU + CSU + Angela Merkel”, “SPD + Martin Schulz”, “Cem Özdemir + Katrin Göring Eckardt + Grüne”, “FDP + Christian Lindner”) |
2021-09-26 | c(“CDU + CSU + Armin Laschet”, “SPD + Olaf Scholz”, “Annalena Baerbock + Grüne”, “Linke + Janine Wissler + Dietmar Bartsch”, “FDP + Christian Lindner”); c(“Alice Weidel + Tino Chrupalla + Afd”, “CDU + CSU + Armin Laschet”, “SPD + Olaf Scholz”, “Annalena Baerbock + Grüne”, “FDP + Christian Lindner”) |
Collecting GT data
Google Trends samples
Google trends data windows
Collecting polling and election data
Prediction models and benchmark
Results
Comparing predictions across GT data windows, across parties
Comparing predictions across model classes
Comparing predictions across model classes and elections
Model class | Election | Better predictions: % | Better predictions: n out of total |
---|---|---|---|
MC1: GT | 2009 | 0 | 0 out of 150 |
2013 | 66 | 99 out of 150 | |
2017 | 97 | 145 out of 150 | |
2021 | 0 | 0 out of 150 | |
MC2: GT + election weight | 2009 | 0 | 0 out of 150 |
2013 | 0 | 0 out of 150 | |
2017 | 32 | 48 out of 150 | |
2021 | 0 | 0 out of 150 | |
MC3: GT + weekly polls weight | 2009 | 20 | 30 out of 150 |
2013 | 29 | 43 out of 150 | |
2017 | 73 | 109 out of 150 | |
2021 | 20 | 30 out of 150 |