“…In many applications, lagged methods for effective connectivity in fMRI are applied to the deconvolved BOLD time series, with an example of GC (David etĀ al., 2008; Goodyear etĀ al., 2016; Hutcheson etĀ al., 2015; Ryali, Supekar, Chen, & Menon, 2011; Ryali etĀ al., 2016; Sathian, Deshpande, & Stilla, 2013; Wheelock etĀ al., 2014). However, as demonstrated in FigureĀ 5b, the natural variability in the neuronal dynamics results with an upper bound on the accuracy of the lagged based methods: even assuming a perfect deconvolution (which is never the case in practice), which would allow for perfect retrieval of the neuronal time series from the BOLD time series, for the TRĀ =Ā 0.70Ā s, the accuracy rate of ā would be <100% (for the parameter space we are exploring in this study, this accuracy would be on the level of 90%).…”