Before 2013, the U.S. government was running many global surveillance programs in the total ignorance of the people through its biggest security agency, the National Security Agency (NSA). In June 2013, an American whistleblower, Edward Snowden, exposed the truth about the illegal surveillance programs to the eyes of the world by leaking classified information in the media. The revelation of classified details from United States government surveillance programs prompted a cultural discussion about national security and individual privacy. Some people were interested in the existence of the a chilling effect subsequent to the revelations, but most of them agreed that no such effect exists. However, Dr. Jon Penney provided evidence of regulatory “chilling effects” of Wikipedia users associated with online government surveillance. Indeed, he showed that traffic to Wikipedia articles on topics that raise privacy concerns for Wikipedia users decreased. Wikipedia can be seen as a good sample of the population, but...

...is it representative of the population's behavior ?

How are Wikipedia users behaving?

As it has been developed in Penney's article, we observe a difference in the way the users are searching for Wikipedia articles before and after the revelation of Snowden. Referring to the graph below, we observe a reltively large decrease in the total number of views of terrorism-related articles. More importantly, before 2013, the total number of views tended to increase over time. After 2013, as can be seen in the following graph, the trend is going down. This is proof that the revelations from Snowden had an immediate effect and suggest that there is also a long-term effect. To verify that the previous hypothesis is not a consequence of a general trend, he compared the result with similar articles that don't raise any privacy concerns. These articles are security-related and cannot be defined as group-control but as comparator groups. The goal of this group is to show that for other security-related topics (that should interest the same category of person) that that do not raise privacy concerns, the revelations had less or no effect.

The previous graph shows that the effect is much higher for the terrorism-related articles than for the control group. Thus, we can infer that the hypothesis might be verified for Wikipedia users.

What about Google users ?

How are Google users behaving?

Google requests are more representative of people's habits, indeed the number of users is way larger on Google than Wikipedia. A similar analysis should be done on Google's data to assess the presence of a chilling effect. First, let's see the interest over two different periods. In Figure 2, there is no evidence of chilling effects. However, this plot is not representative of the underlying trends! What if there were an immediate chilling effect after the revelations but no long term effects? Let's continue our analysis!

In order to represent the trend of the interest in the terrorism-related subjects over time, we are using a regression for both periods, similarly to what has been done for Wikipedia.

We observe a slight decrease of 8% (even more if we look at the points, the regression overestimate the interest in the beginning) in the interest after the Snowden revelations. It could be considered as a short-term chilling effect. However, very fast the trend is catching up with his old value and after 3 months, the chilling effect is no longer apparent. Now, I'm sure you're wondering: what if it's not really due to Snowden's revelations?

Don't worry, we've got it covered just below.

How are other topics varying ?

Maybe you've had enough about terrorism. Let's change the subject for a moment and speak about Domestic Security. Domestic security-related queries have a closely related content with terrorism and will likely attract similar users. After a similar analysis of this domestic security-related queries, let's see what are the trends with our new subject.

The previous graph shows a bigger drop in the intereste than the control group. It supports the hypothesis that the chilling effect is affecting terrorism-related topics more than other topics such as domestic security-related topcis.

Oh no... we forgot to check if some data could mislead our analysis. Indeed, the way one interacts with Wikipedia and Google is very different. Wikipedia is used to do some research, to learn new things, to check a fact, etc. We rarely go on Wikipedia to look at the attack article just after an attack happened. We would most probably just write Attack on Google. Consequently, Google requests are strongly related to the news. It is obvious when we take a closer look at some research terms such as Terrorism (most probably related to some attacks) that some events induce some huge peak of interest around these events. This sensitivity to world events induce a lot of noise in the data and could lead to false interpretation. Finally, after the removal of these subjects that we call outliers, we should have a new look at the trends!

How surprising! Did you notice that the terrorism trends now seem to be completely independent of surveillance programs' exposure? Indeed, when we remove the data that are related to events from the news, we don't see any chilling effect at all. An answer to this mystery could be that we have two categories of people. The first category could be called the inquisitives, they tend to be interested only in hot topics but wouldn't be interested in this subject normally. The second group includes people that would search for this topic in everyday life (for research purpose for example). We hypothesize that the first category of people, the inquisitives, is affected by the chilling effect, while the second one is not.

Do government activities have an effect on how people use Google?

As hypothesized in the last section, a given group of people could be subject to the chilling effect while others not. However, we need to qualify our point. Indeed, we don't have statistical proof that a chilling effect, even a short-term effect, exists. The main limitation consist in the noise induced by events. News can create big peaks of incentive on specific subjects and thus a lot of noise due to other factors. As a way to go further, it could be interesting to verify the hypothesis we made as well as finding out if an hypothetic "chilling effect" spread only in the U.S.A or worlwide.