Firing Rate

Three estimates are in common use (Dayan & Abbott (2001), ch. 1), and each depends on a time window chosen by the analyst. A mean rate over a session or an epoch gives one number per unit. A binned rate counts spikes in successive bins. Narrow bins resolve fast changes but are noisy, and wide bins are smooth but reduce temporal resolution. The bin width can be chosen from the data (Shimazaki & Shinomoto (2007)). Aligned to an event, the binned rate is the PSTH. A smoothed rate convolves each spike with a kernel (usually a Gaussian), whose width has the same role as the bin width, and provides an estimate from a single trial.

Firing rate is one description of a spike train. Under the spike-count hypothesis, a neuron signals only through its rate over a time window. However, reducing a spike train to counts can only lose information, and in sensory pathways spike timing can add to the count at a precision of a few milliseconds (Quian Quiroga & Panzeri (2009)). Considered together with the spike waveform, firing rate is also evidence of cell type (Sun & Ibbotson (2026)).

A firing rate assumes that the spikes belong to one neuron. A unit that merges two neurons reports their summed rate, and a unit affected by electrode drift shows a rate that decreases for reasons that are not biological. Firing rate is therefore computed after spike sorting and interpreted together with the unit's quality estimates. The event rate that Radiens reports per channel is a different measure: it counts threshold crossings on a site (a multi-unit measure) and is used for live signal quality.