“…For this to be the case, the distributions need to have a large positive skewness γ (larger than, say, 2), or, equivalently, a large scaled variance s. Clearly, this coupling between the skewness and the spread is not present for a distribution with more than two independent moments, but in the discussion below we show that a large value of s does actually imply a large skewness and that a large variance without any skewness is not sufficient to obtain a low PT. In any event, for those few CNT systems for which the moments of the length distributions have actually been determined, the skewness and spread seem to be quite small, i.e., γ = 0.40 and s = 0.20, giving φ p (x)/φ p (x 0 ) = 0.84, 40 or φ p (x)/φ p (x 0 ) = 0.64, 48 or φ p (x)/φ p (x 0 ) = 0.50 for an exponential distribution; 43,44 so for these the effect of polydispersity is modest. Given the very large scatter in measured PTs of carbon-nanotube composites, 17 we conclude that the skewness in the distributions must usually, in fact, be much larger than this.…”