The Financial Times carried a good discussion on the failure of the Google Flu Trends model to predict.
High quality global journalism requires investment. Please share this article with others using the link below, do not cut & paste the article. See our Ts&Cs and Copyright Policy for more detail. Email ftsales.support@ft.com to buy additional rights. http://www.ft.com/cms/s/2/21a6e7d8-b479-11e3-a09a-00144feabdc0.html#ixzz2xLIVMnX4
I am not able to provide an excerpt, but will paraphrase.
In the instance of the model, the explanations are that Google itself changed the conditions and introduced encouragement for additional searches (that in turn showed an epidemic larger than real), and that media hype re flu had driven up the searches. Any model will fail when one of these two conditions is met.
The discussion by Tim Harford then goes on the discuss sample bias. The n=all assumption, Harford shows, is true in all cases, or in most. This is what I discussed in relation to Google flu trends a few weeks back. It is because of our understanding of the problem that we have, from the beginning, focused on mobile transaction-generated data. That, we said is the only datafied data stream that could tell use about the poor. And we have from the beginning strived for n=all, though we have not quite achieved it yet because it’s not that easy to process data from multiple mobile operators.
So the Google Flu Trends failure should not harm our work. It will, hopefully, deflate the hype.
1 Comment