Youngs et al. (1988)

See Section 2.71.
Ground-motion model is: $ln(Sv∕amax) = C6 + C7(C8 - Mw )C9$
Response parameter, S_v, is velocity² for 5% damping
Develop relationships for ratio S_v∕a_max because there is a much more data for PGA than spectral ordinates and use of ratio results in relationships that are consistent over full range of magnitudes and distances.
Calculate median spectral shapes from all records with 7.8 ≤ M_w ≤ 8.1 (choose this because abundant data) and R < 150km and one for R > 150km. Find significant difference in spectral shape for two distance ranges. Since interest is in near-field ground motion use smoothed R < 150km spectral shape. Plot ratios [S_v∕a_max(M_w)]∕[S_v∕a_max(M_w = 8)] against magnitude. Fit equation given above, fixing C₈ = 10 (for complete saturation at M_w = 10) and C₉ = 3 (average value obtained for periods > 1s). Fit C₇ by a linear function of lnT and then fix C₆ to yield calculated spectral amplifications for M_w = 8.
Calculate standard deviation using residuals of all response spectra and conclude standard deviation is governed by equation derived for PGA.

²In paper conversion is made between S_v and spectral acceleration, S_a, suggesting that it is pseudo-velocity.