Abstract:
|
The spike count correlation (SCC) is the Pearson correlation of the spike counts of two neurons recorded over multiple repetitions of an identical stimulus. SCC may reflect A) the correlation in the time-averaged drive to these neurons, which we call "firing-rate correlation" (FRC), B) point-process noise, and C) more precisely-timed correlation effects. We describe how SCC may not accurately indicate FRC, due to distortions caused by B) and C). We assume a hierarchical model where the spike counts are jointly generated at firing rates which are random. We propose bivariate (correlated) Poisson mixture models and nonparametric estimators to separate SCC from FRC, and test for the presence of within-trial correlation via jitter methods. We show that SCC and FRC can have different orderings across stimuli and pairs of neurons, and we formalize this phenomenon probabilistically and call it reversal problem. We apply our methods to data from macaque V4 area and find that within-trial correlation affects SCC in a limited way, while attenuation due to trial-to-trial variation dominates the dataset. Goodness-of-fit results and analysis of sensitivity support the proposed models.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.