| Great questions! For background, I was part of the Flu Trends and Consumer Surveys teams before I left Google to start Iodine. Google Consumer Surveys attempts to create a representative sample of the internet population according to the breakdown from the Current Population Survey (CPS) published by the Census. This is done at survey delivery time. Additionally, after the survey is performed, the stats are reweighted by this distribution since the targeting system isn't always able to deliver a fully representative sample due to other constraints (publisher volume, survey volume, etc). Of course, this only helps with ensuring that the stats represent Internet users. However, there are many reasons to believe that this sample is a reasonable proxy for many types of questions. For examples, see: http://www.google.com/insights/consumersurveys/static/consum... http://www.people-press.org/files/legacy-pdf/11-7-12%20Googl... http://fivethirtyeight.blogs.nytimes.com/2012/11/10/which-po... Additionally, there can be a bias in many traditional survey platforms (RDD telephone surveys being the current gold standard) due to the adoption of mobile phones among other things. With respect to recall-reporting problem, for a subset of users, we asked not only about the medication but what they were taking the medication for. We do this in the form of an open ended question and it helps increase our confidence in what the user reports. Details of our exact methodology are forthcoming! |