Figure 3

Symptom or sign complexes for A) H1N1-2009, B) Influenza B. Symptoms or signs are ranked by average frequency over the study period over all strains. The expected proportion of symptom pairs assuming symptoms develop independently is the product of the two marginal distributions. The discrepancy between this and the observed proportion of symptom pairs is assessed via a binomial test, with differences that are significant at the 5% level represented by coloured cells, with the thickness of the cell wall indicating the p-value (thin means 1% <p < 5%; medium, 0.1% <p < 1%; and thick, p < 0.1%). Colours encode the excess probability ratio, a measure of effect size, with blue indicating a lower proportion and red a higher proportion observed than expected, and shade indicating the ratio of the empirical proportion that the pair of symptoms appear together relative to what would be expected if their occurrence were independent. With 420 tests, the conservative expected number of false discoveries is 21, assuming all symptoms or signs occur independently".