The unit of analysis was percentage of clinic patients with ILI and percentage of laboratory tests positive for influenza. To ensure this assessment of surveillance data was comparable to the study that validated Google Flu Trends, we performed a secondary analysis replicating methods from that study with our dataset. The mean coefficient of correlation between Google Flu IM-54 Trends and CDC ILI Surveillance was calculated over nine US Census Regions for the 2007�C08 influenza season. The mean coefficient of correlation between Google Flu Trends and CDC Virus Surveillance was calculated similarly for comparison. We also performed a secondary analysis to determine the sensitivity of the primary analysis to high-leverage, outlier observations. First, we performed simple linear regression to evaluate the association between either Google Flu Trends or CDC ILI Surveillance rates with reference viral surveillance data as standard rates. The effect of outlier PNU112455A observations was assessed with differences in the beta statistic. Individual observations were considered influential if they had a DFBETA greater than the absolute value of 2 divided by the square root of the total number of observations in the model. Subsequently, all influential observations were excluded and correlation coefficients were recalculated as had been done in the primary analysis. Correlation coefficients from the sensitivity analysis were compared to the same statistic from the primary analysis to determine whether any relevant changes in the strength of correlation had occurred with the removal of influential observations. Last, Spearman Rank correlation coefficients were employed for the primary analyses and noted to yield similar results. In terms of coefficients of determination, 88% of the variance is shared between Google Flu Trends and CDC ILI Surveillance, while only 51% of the variance is shared between Google Flu Trends and surveillance for laboratory-confirmed influenza. From September 2003 through May 2008, CDC ILI surveillance was more closely correlated with CDC Virus Surveillance.