RESEARCH TRENDS
Can we really use Google data for research?
By using the frequency of Internet searches, trends can be evaluated almost in real time. But such methods have their pitfalls.
“How long will Covid last?” This was one of the questions most frequently Googled in Switzerland in 2021. The data that Google can provide is valuable in the social sciences. Even central banks meanwhile take decisions using Google Trends. This is because, inter alia, developments in consumer behaviour are reflected in people’s search activity long before the impact is felt in sales statistics.
This method – called ‘nowcasting’ – is not only faster than conducting surveys, but also solves a big problem common to them. When asked questions, people tend to provide answers that they think their conversation partners want to hear. This desirability bias does not exist with Internet searches, which also makes it easier to assess the electoral chances of right-wing populists.
But this abundant data source has its own pitfalls. Researchers have found that the data Google makes available to its advertisers actually deviates considerably from publicly available data. The time window selected plays a major role in this. “The fresher the data, the bigger the deviations”, said Sergej Zerr, an IT researcher at the University of Hanover, in conversation with the Neue Zürcher Zeitung. Alessandro Rovetta works for the Italian software company Redeev, and his own analysis led him to conclude the following: “Any study performed through Google Trends is inherently not reproducible”.
Researchers from the Swiss State Secretariat for Economic Affairs, ETH Zurich and the University of Basel have now published a statistical method to eliminate inconsistencies according to data age, the time window selected and the size of the catchment area. This means that “the development of public sentiment as well as public interest in certain topics or products” can be carried out with Google Trends reliably in future.