Models selected using different information criteria.

The state-space log-linear models containing interactions up to the th-order () are applied to the data from simultaneous spike sequences. The spike data is generated from a time-dependent full log-linear model (see dashed lines in Figure 4C) repeatedly for trials; either (left) or (right) trials. For this data set, we compute three information criteria (AIC, BIC, and PDIO) and find the order of the model that minimizes these information criteria. We repeated the selection of the model order times, using each of the criteria and using the spike data that contains respective number of trials ( or ). The count of the order of spike interactions that minimizes the applied information criteria is increased by 1, and finally expressed as a frequency. The asterisk marks the most frequently selected model.