Model selection results for ISI probability distribution of experimental PC.

Natural logarithm of the marginal likelihoods of the first seven models in the hierarchy are shown for N = 28966 spikes (the full data set). Numbers following the ± sign are the standard deviation of the marginal likelihood, estimated using importance sampling (see Materials and methods). Since we show log likelihoods instead of likelihoods, the standard deviations transform into asymmetric errors around the mean, and both asymmetric errors are shown. The model with the highest marginal likelihood, M = 5, is highlighted. Note that the model with M = 6 cannot be ruled out, as it has very similar marginal likelihood.