Anticipating epidemic transitions with imperfect data
Fig 3
Lead up to an outbreak of measles in California.
Weekly case report data from the US CDC’s MMWR compiled and released by Project Tycho [34]. Estimates for measles reporting efficacy in the USA are around 45% [27, 35, 36] In addition to the raw weekly case reports, the bottom row shows the data aggregated into bi-weekly and four-weekly reports. Changing the aggregation period changes not only the values of the variance (top row), but also the trend, quantified using Kendall’s τ. Only data to the left of the shaded area are included in calculating the variance. The moving average is calculated using a two year window, i.e bandwidth b = 52 for weekly reports; b = 26 for bi-weekly; b = 13 for four-weekly. For details see the Methods section.