While studying, I came across the following quote:
"A more serious disadvantage is that successive points on a cumulative distribution plot are correlated — the cumulative distribution function in general only changes a little from one point to the next, so adjacent values are not at all independent."
What does it mean to say that data points on a plot are correlated?
Newman is not being precise here, but the way I understand it is that the error on the empirical cCDF (relative to the true cCDF value) will be correlated for nearby points. Whereas for a histogram, the error of the empirical frequency of nearby bins are independent, except for some global effects. (The error in the complementary cumulative is related to the sum of the errors above that point in the histogram, so you can see why it is correlated.)
As he indicates, this makes least squares regression a suboptimal method for extracting the power law, though there are deeper problems than this.