Figure 5.

Simulation using the Bayesian forecasting procedure to predict performance over time for subject “A,” starting at a later time during the sleep deprivation period. Details are the same as for Figure 2 (left column panels), except that here the initial homeostatic state was different because the simulation was started at 24 h awake (but for the purpose of the simulation the amount of prior wakefulness was considered not known). Timewise, the top panel in this figure corresponds to the fourth panel in the left column of Figure 2. The second through sixth panels display the updated 24 h performance predictions as time passed, shown in 8 h increments. Note that there were only 9 actual observations (gray circles) to compare to in the last panel, because data acquisition stopped at 88 h awake in the laboratory experiment.