Figure 4From: Mood dynamics in bipolar disorderIn-sample error ratio for nonlinear vs linear modelling. The original time series is represented as a vertical line and the errors for the surrogates are shown as a histogram. The panel on the bottom-right shows that the original series is relatively better forecast by the linear than the nonlinear method. This anomaly is caused by the surrogates having an artificially lower average correlation, shown in the equivalent panel in Figure 3.Back to article page