I’m a fan of using whatever tools we have available for collecting data, especially when the methods don’t reveal any personal information of individuals. When I first started writing about the potential effects of the coronavirus lockdown and social distancing efforts, I took a look at some Google Search trends, focusing on the United States, in part because that’s where I am, and also because that’s where we’re seeing a lot of new cases.
Google Trends is a powerful tool for population level psychological analysis. We can’t easily poll a large portion of the population, day after day, and see how they are feeling. But we can look at search habits. The tool has been used in quite a bit of research. Both Nuti et al. 2014 and Seung-Pyo et al. 2018 analyze a number of studies that utilize Google Trends data.
The former study looked at 70 studies conducted between 2009 and 2013 and suggested a seven fold increase during that time period. Of those studies, 24% covered topics of mental health and substance use. The latter analyzes 657 research papers. The research suggests that reliance on Google Trends increased substantially in the 10 years leading up to the study and that in recent years, trend data is used in combination with other big data sources.
Unfortunately, not enough data was available to make any inferences about the changes in population level psychology at the time, but I did notice a few things. First, searches for stress related terms, such as anxiety and depression follow a roughly seven day cycle, with lows appearing on the weekends and highs appearing around mid week. I’m sure that I’m not the first to see this trend, but it is still an interesting one, and something that doesn’t seem to be discussed in enough detail.
That mood is influenced by the season has been fairly understood for some time. The disorder was first identified by name in 1984 by Normal E. Rosenthal et al. The condition is known as “seasonal affective disorder.” But seasonal mood disorder is a long term shift in emotional state, and isn’t quite well understood. This pattern occurs over a short period of time, and seems to be clearly linked to the work week. But how do I know the trend is even there?
I’m a big fan of doing more than just visual analysis. And I want to make sure that I can remember as many different statistical methods as possible. So I decided to run to Python, one of the easiest and most powerful tools for data analysis. The main analysis I decided to select was autocorrelation analysis. The idea here is that if a time series data set more or less repeats after a given time period, the series should correlate with a shifted time series that’s offset by that period. In other words, if the data set repeats every seven days or so, there should be a strong correlation around the seven day offset.
Autocorrelation for 0 should always be one, since a time series has to correlate exactly with itself. The next peak is at seven days, which is exactly what was expected. The autocorrection plot was created using the tsaplots module in statsmodels.
Expectations for Shifting Trends
With the workweek essentially cancelled for many people, I expect to see a reduction in the cyclical nature of these search trends. It already appears to be happening, and now we’re starting to see searches for “anxiety” trend higher. This shift appears to have started sometime around March 13th. I also expect any reduction in stress from a lack of workweek to be overwhelmed by the increase in stress caused by the lockdown and many people being unemployed. I can’t seem to embed Google Trends data in the article itself, but you can access the results for “anxiety” using this link: Google Trends for Anxiety.
Even if the exact impact on anxiety has yet to become identifiable using trend data, one trend did become clear fairly quickly is that the country is trying to figure out how now to be bored. From an initial observation, searches for “boredom” have nearly tripled since the lockdown began. And boredom can lead to more anxiety, as well as depression.
Aside from direct changes to stress levels, I am also concerned about what the changing stress levels could do to us, as a society. If stress ends up increasing sufficiently, it can lead to numerous detrimental outcomes, increasing depression, illness, aggression, societal unrest, and more. We’re already seeing these kinds of things happen in Italy (Bloomberg). Keeping an eye on search trends can help us predict if and when it’ll happen here.
Originally published in Longevity on Vocal.Media
I don't have a master index for publish0x or Vocal articles yet, but I do have some index articles on Medium, which link to a whole bunch of different articles that I've written.