- Based on Lee et al. (1995). See Section 2.128.
- Response parameter is pseudo-velocity for 5% damping (also use 0, 2, 10 and 20% damping but do not report results).
- Before regression, smooth the actual response spectral amplitudes along the log
_{10}T axis to remove the oscillatory (‘erratic’) nature of spectra. - State that for small earthquakes (M ≈ 3) equations only valid up to about 1s because recorded spectra are smaller than recording noise for longer periods.
- Only give coefficients for 0.04, 0.06, 0.10, 0.17, 0.28, 0.42, 0.70, 1.10, 1.90, 3.20, 4.80 and 8.00s but give graphs for rest.
- Assume that distribution of residuals from last step can be described by probability function:
where p(ϵ,T) is probability that log PSV(T) - log (T) ≤ ϵ(T), n(T) = min[10,[25∕T]], [25∕T] is integral part of 25∕T. Arrange residuals in increasing order and assign an ‘actual’ probability of no exceedance, p

^{*}(ϵ,T) depending on its relative order. Estimate α(T) and β(T) by least-squares fit of ln(-ln(1 - p^{1∕n(T)})) = α(T)ϵ(T) + β(T). Test quality of fit between (ϵ,T) and p^{*}(ϵ,T) by χ^{2}and Kolmogorov-Smirnov tests. For some periods the χ^{2}test rejects the fit at the 95% level but the Kolmogorov-Smirnov test accepts it.