menu

Google Trends on the wall, tell me who…

Google has become our oracle, to whom we turn when we need information, advice or even relief. We forget, however, that good oracles, the true ones, answered more about ourselves than about the world around us. Well, since 2004, Google has decided to also tell us things about ourselves through “Google Trends”.

Illustration by Klifton Kleinmann.

Google Trends is a project that allows to see how popular a certain keyword is among all the searches done through Google. In other words, it shows how often a particular search term is entered relative to the total search volume. This tool allows us to restrict the search to a particular region and to a particular time period. For example, if we search for “tomato” in United Kingdom in 2015, Trends analyzes a percentage of all searches in Google in UK in 2015 and tells us at what point tomato was popular among all the searches that english residents have done in 2015. Google Trends also allows the user to compare the volume of searches between two or more terms, between two or more regions and between two or more periods of times. Then, we can determine if the word “tomato” was more popular in 2015 or in 2014, if it was more popular in UK or in Australia (see Figure 1) and if it was more or less popular than “carrot”.

Figure 1.Screenshot of Google Trends, comparing the popularity of “tomato” in the United Kingdom and Australia.

Google Trends has gained popularity among scientific communities from various disciplines. Medicine, for example, uses the tool to know in real time the evolution of an epidemic, or to detect one before the affected reach emergency rooms. A lot of patients look for symptoms or home remedies before going to a doctor. Having access to data on web searches, such as provided by Google Trends, anticipates the detection of several epidemics and provides a wider margin of maneuver to the health authorities.

In 2008, Google developed "Google Flu Trends" (now closed), a utility that monitors multiple keywords related to flu and indicates the level of incidence of the illness among the population in a specifc region in real time. Flu Trends can detect regional outbreaks of flu between 7 and 10 days earlier than the conventional monitoring systems of disease control and prevention centers.

Flu Trends only monitors two diseases, but several authors have used Google Trends to detect outbreaks of many other diseases or to detect seasonal patterns of certains symptoms. Detecting seasonal patterns of certain diseases is often difficult with traditional data because, for instance, some diseases do not require medical visit. Researchers like David Ingram, Camilla Matthews, Camille Pelat, David Plante and Louise Rossignol, among others, have investigated several seasonal patterns of disease using Google Trends. Two of the diseases that they studied: urinary tract infections and snoring seem to show marked incidence peaks. We have more urinary infections in summer and we snore more in winter and during early spring. Regarding urinary infections, the authors not just use Goolge trends data but also medication sales data and found the same stationary pattern, showing some evidence that Google Trends can really capture actual health condition with the advantage that Google Trends data is much easier to consult.

Medicine is not the only discipline that has managed to use this data to tell us things about ourselves. Several economists have used Google Trends and have confirmed that the searches for certain keywords are an excellent indicator of various elements of economic activity almost in real time. Traditional indicators of the same elements will appear in official statistics months or even years later. Let us consider real estate statistics: It is not surprising that, if purchasing decisions are made well in advance, Google searches can really predict the future, since it is clear that if words like "buy house" or "real estate" increase its popularity among Google searches, it is likely that within a few weeks actual home sales data reflect an increase. Economists Hyunyoung Choi and Hal Varian studied various sources of data on car sales, unemployment, consumer confidence to buy or tourist destinations. They found that the estimation models that incorporate Google Trends predict the actual values between 5 and 20% better than models that do not incorporate them.

Economists Tobias Preis, Helen S. Moat and Eugene Stanley used Google Trends to study financial markets and found that changes in the trends of financial related searches on Google is very useful in finding "early warning signs" of future shocks in markets. This type of data is interesting per se because it provides valuable information to better understand ourselves, but also allows the authorities to act with more room and more information to possible shocks that greatly affect societies as a flu epidemic or a financial crisis.

When I realized this potential, I also started to search interesting data on Google Trends. First, I tried to better understand what was exactly the information that Trends provides1. The next box explains the algorithm used.

In order to preserve the anonymity of users, when we use Google Trends, what we obtain is not the total number of searches of a certain word, but a normalized index of the number of searches:

For example, for a unique keyword searched over a period of one year in a certain region, Google Trends first determines its popularity ratio for each week. The popularity ratio shows the number of searches of the keyword over the total number of searches. This ratio is then normalized proportionally such that 100 corresponds to the highest popularity ratio of the keyword in the whole year period. Thus, a normalized index of 50 means that the keyword was searched half as much as in the week where it was equal to 100. Normalized indexes allow then to measure the evolution of the number of searches of a keyword in time, to compare the popularity of a particular keyword in several regions of the world or to compare the popularity of several keywords.

Once it was clear what information Google Trends provides, I started to search keywords that relate with social movements like “change”. I searched it in Spain between April 2011 and January 2012. Since there was a very important social movement in May 2011 in Spain –15M or “The Outraged”–, I expected a pick or a change in the trend around May 2011. The results that I found are represented in Figure 2.

Graph showing the evolution of the popularity index for the word "change" in Google searches between April 2011 and January 2012. A peak can be observed in November 2011.
Figure 2. Source: Elaboration of the author from Google Trends data.

Surprisingly, I did not found any type of peak nor clear change in the trend around May 2011 but there was a peak at the end of October. Immediately, I start searching what happened in Spain in October 2011 and I did not found anything relevant. Then, I tried to see the evolution of searches of “change” in a longer period, between January 2010 and April 2015. The results can be seen in Figure 3.

Graph showing the evolution of the popularity index for the word "change" in Google searches between january 2010 and april 2015. Periodic peaks can be observed at the end of March and October each year.
Figure 3. Source: Elaboration of the author from Google Trends data.

Periodic peaks can be observed at the end of March and October each year, matching the time change that occurs in Spain between winter and summer time each year in October and March.

I tried then with another word with similar meaning but that is less ambiguous: “alternative”. The results are shown in Figure 4.

Graph showing the evolution of the popularity index for the word "alternative" in Google searches between January 2010 and April 2015. A clear peak can be observed in May 2011. Other peaks are present.
Figure 4. Source: Elaboration of the author from Google Trends data.

Now, we can clearly see a peak in May 2011. We see that the patterns are far less seasonal than “change”. This simple exercise shows why one must be extremely careful when working with data. Data from Google Trends are a very nice and powerful tool, but it should be used properly or it can lead to misunderstandings or even false conclusions.

Google has become a very good oracle that not only can tell us an infinity of things about the world that surrounds us but also can show us a lot about ourselves. We must be careful with the questions we ask; we must, like with the good oracles, the true ones, choose our questions wisely.

References

  • Choi, H., & Varian, H. (2012). Predicting the present with Google Trends. Economic Record, 88(s1), 2-9.
  • Ingram, D. G., Matthews, C. K., & Plante, D. T. (2014). Seasonal trends in sleep-disordered breathing: evidence from Internet search engine query data. Sleep and Breathing, 1-6.
  • Preis, T., Moat, H. S., & Stanley, H. E. (2013). Quantifying trading behavior in financial markets using Google Trends. Scientific reports, 3.
  • Rossignol, L., Pelat, C., Lambert, B., Flahault, A., Chartier-Kastler, E., & Hanslik, T. (2013). A Method to Assess Seasonality of Urinary Tract Infections Based on Medication Sales and Google Trends. PLoS ONE, 8(10), e76020. doi:10.1371/journal.pone.0076020

Notes


  1. In the following explanation of the index I will just present the case in which only one keyword is searched.