Figure 2.

Left panel: Same figure as the left panel of Figure 1, but including CC_norm and its squared values for reference. Note that, for good predictions (values above ca 50%), $C{C}_{norm}^2$ and SPE are almost identical, both yielding very similar estimates of the proportion of “explainable variance explained.” This is as might be expected given the equality of Equation 7 and 28 when ŷ → y. However, as the prediction performance declines below 50%, $C{C}_{norm}^2$and SPE increasingly and sharply diverge. Right panel: Scatter plot of performance scores for predictions of neuronal responses from three different experiments. Each marker reflects the performance score of the prediction of the response of a single neuron. The black dashed line visualizes where SPE equals $C{C}_{norm}^2$. The values of CC_norm have been multiplied with their absolute value to demonstrate that negative values only occur for SPE, but not for CC_norm. The solid red line shows the values for the simulation of Figure 1, the dotted red line and the red cross mark the performance scores of the 60% noise simulation of the lower right panel in Figure 1. Corresponding to the overlap of SPE and CC_norm for good predictions, the red line approaches the dashed black line.