Erroneous classification of neuronal activity by the respiratory modulation index

Failure to record respiratory activity in the mesencephalon of the chronic cat led us to analyze the formula (the respiratory modulation index, RMI) used by Hugelin and his colleagues to discriminate respiratory neurons. Using computer simulations, we compared RMI with the analysis of variance (F) and the non-parametric Friedman's test (chi 2). Samples were drawn repeatedly from simulated distributions of neuronal activity and were allocated to successive bins representing the respiratory cycle. Allocations of bins were made randomly so that only a chance relationship existed between the simulated activity and respiratory cycle. These simulations revealed that the RMI erroneously yields values indicative of a respiratory relationship and does so as a function of sample size and the variability and shape of the distribution of non-respiratory activity. Although some of the simulated conditions violated assumptions of the F test and, to a lesser degree, the chi 2, these statistics erred at rates close to the chosen 5% level. When respiratory activity was stimulated, chi 2 and F were more sensitive than RMI in detecting the relationship. We conclude that the high incidence of respiratory activity reported by the Hugelin group is based upon a faulty statistic and is highly questionable.