Search Trends And What They Reveal
I’m hooked on Google Trends. For example, it probably won’t astound you to learn that whenever search trends peak for flu symptoms, there’s a similar peak for vaccine.
(See also how at some points vaccine comes first and only then do flu symptoms arise. Conspiracy theorists, the ball is in your court!)
Probing this particular case further, I looked at the search trends for swine flu and vaccine. The former peaks in April 2009 and, to my surprise, vaccine actually declines over the following two months. However, it skyrockets during the third peak of the swine flu searches. What does this tell us? I’d like to speculate that people have become resilient to rumors, needing factual confirmation before rushing to the vaccine line. However, one thing is sure: This is a clear incentive for pharmacies and health care stations. What better time to promote their goods than the minute search trends for flu activity begin to gain statistical momentum?
By the way, like me, Google has tapped into Trends as a resource for identifying flu trends (see Google Flu Trends).
XThe Simmer Newsletter
Subscribe to the Simmer newsletter to get the latest news and content from Simo Ahava into your email inbox!
Search trends, correlations and conclusions
I’m a HUGE fan of Freakonomics by Steven Levitt and Stephen J. Dubner. It’s a book on economics, correlations, causality, data mining and other boring-sounding topics, but the authors have managed to spin a yarn so delightful that it makes for a very entertaining read. Freakonomics studies trends and how they seem to correlate. They draw conclusions that might at times stray beyond the scope of credibility, but it all appears kosher enough to a layman (which is probably why it’s a bestseller). Anyway, once you get really into populist economics like this book, you start to get paranoid.
You start seeing correlations everywhere.
For example, between 2004-2013 there’s a clear decline in the search trends for war, which is a good thing, I guess. I like to think that it’s an indication of the fact that war is not the foremost thing in people’s minds. On the other hand, there’s a clear rise in queries for what to do, the motto of a very bored generation. It would be so easy to make a tale out of these two distinct search trends. Maybe it would be something like “there’s no war going on, what should we do now?"; an unhealthy, completely irresponsible statement (and factually wrong) coupled with a very popular information query (e.g. what to do in Berlin?), coming together in a correlation which might mislead the misinformed.
This is, of course, completely ungrounded conjecture. But seemingly unrelated correlations can be a breeding ground for some pretty amazing conclusions, as can be seen in Freakonomics. War and what to do might be an example of search trends having correlation only in the wildest imaginations of blog authors. Conversely, a study was done on the search volumes for debt, and it showed that search trends for debt were a very clear indicator of market fluctuations, thus showing that studying search queries has concrete applications.
How are we doing, really?
What do search trends tell about the state of humanity? Can you really draw conclusions on the basis of what people type into the small input box day in, day out? I think you can, up to a point. First, you need a hypothesis before you start shuffling through the data. Just throwing in terms at random provides optimal manure for misguided conclusions based on your first interesting findings (see the comparison between war and what to do above). Second, a good quantitative analysis can go only so far without a qualitative component. In addition to crunching up the numbers, think of how the statistics relate to events around the globe during the peaks and troughs in the search volume. Think of why people did those very searches in those very months.
Once you accept the fact that you have no control over the quality of your informants, that your case study has little to do with ideal, clinical laboratory conditions, and that you are looking at search trends not absolutes (people do a lot more searches than just the ones you look at), you might see something of interest. Maybe this turns into a lucrative chance for you, especially if you identify something that has market value. Or maybe you’ve found something that tells you more about human nature than a single Superbowl commercial ever did.
Or maybe you forget all about statistical significance and quality, and you see something like this, and it makes you smile just a little bit:
