Optimal smoothing parameter selection for smoothing spline (csaps)?

According to the documentation for csaps, "the interesting range for p is often near 1/(1+h^3/6), with h the average spacing of the data sites". (Note that p is the smoothing parameter in $p\sum\limits_{j=1}^n y_j-f(x_j)^2+(1-p)\int|f''(t)|^2 dt$; see csaps documentation). I have searched the literature for an explanation of why this is so, but I am unable to find a reference for this formula. I did find a paper (Woodford, C.H., 1970, An algorithm for data smoothing using spline functions: BIT Numerical Mathematics, v. 10, no. 4, p. 501-510) that suggests that p^(-1) should be an estimate of the standard deviation of the ordinate y_i; however, I fail to see how the standard deviation of the ordinate corresponds to the average of the abscissa. How was the formula for p derived, and/or what is the reference for the formula?