If you’re looking for a map of H1N1 cases across the US, you have several choices — each one with caveats because the data on H1N1 cases aren’t straightforward. In fact, there are no precise data on H1N1 specifically — only best estimates of “influenza-like illness,” which includes both H1N1 and seasonal flu. (The CDC aggregates several data streams to come up with its estimated flu numbers.) This limitation is reflected on all the maps: each one tracks influenza in general — not swine flu in particular — and is based on estimated numbers.
CDC offers FluView, which tracks “influenza activity” and generally has a time lag of a week or two. NPR, also using CDC data, recently created another map of national “influenza-like illness.” (Get the embed code here.) Flu.gov, in collaboration with HealthMap, offers its own best guess at general influenza levels. (For more on their data collection, see this post.) Finally, there’s Google Flu Trends, whose data might appear the most speculative: it bases its flu map on Google search data — essentially, on the number of flu-related searches.
We’ve been wondering how to assess the validity of Google Flu Trends. Google says that historical data correlate well with their Flu Trends results: “estimates based on Google search queries about flu are very closely matched to traditional flu activity indicators. Of course, past performance is no guarantee of future results.” But what about an outside opinion?
You might know Nate Silver from his blog FiveThirtyEight, which commanded a lot of attention during the 2008 election for its predictive power. Silver is a numbers lover. He made his name interpreting baseball statistics and then shifted his focus to politics. Occasionally he goes farther afield and just writes about some stats that intrigue him — as he did recently about Google Flu Trends. He seems rather bullish on it:
Yesterday, I had the opportunity to speak with a couple of engineers from Google who are working on a product known as Google Flu Trends. This is a very simple, yet elegant and important application of what might be termed predictive analytics; if there were awards given out for such things (the Jameys?), it would be a good candidate to win one.
Silver goes on to explain the advantage of Flu Trends, as he sees it, over the CDC-based maps:
The advantage of this is that whereas the CDC typically works on a 10 to 14 day lag before new flu statistics are published, the Flu Trends numbers can be turned around literally overnight. Flu Trends does not predict the future per se, so much as it “predicts the present”, as the engineers describe it.
Silver’s stamp of approval is of course no guarantee that Google Flu Trends is a perfect or even good current estimate. But, for what it’s worth, Silver offers a smart, independent opinion — which could be fodder for an interesting interview.