One recent study by Johns Hopkins University in Maryland sought to determine whether user "tweets" or messages on Twitter can be used to forecast or predict the rate of influenza in the following weeks (Paul et al., 2014). They used a basic linear autoregressive model, which is similar to logistic regression except that the features are the influenza rates of the previous weeks and the coefficients are determined by least squares regression. They compared models that used only the CDC's published weekly rates with models that also used a Twitter influenza surveillance system that analyzes tweets to determine whether they are about influenza infections. Their study found that incorporating Twitter data helps reduce errors in the predicted influenza rate by 15 - 30% up to 10 weeks into the future. Similarly, a review study analyzing the use of social media for monitoring disease found that joint information from social media sites and national health statistics predicted outbreaks before standard outbreak surveillance systems (Charles-Smith et al., 2015).