Number of cells recorded in each condition.

aFor our measurements of output correlation (Fig. 3C), we include additional data from experiments performed as part of other studies in which receptive field structure was not probed. For model correlations and other analyses, we only used the subset of retinae and cells for which we obtained robust receptive field estimates.bWe used a stringent requirement that receptive fields (RFs) be of high quality for any analyses in which we used receptive field estimates (e.g. in Fig. 3D).cCells were only divided into classes if they had high-quality receptive fields and were recorded in response to stimuli chosen for detailed analysis.d confidence intervals were computed by boot-strapping. See Methods for details.