- Ground-motion model is
^{4}: - Response parameter is acceleration for unspecified
^{5}damping. - Use two site categories:
- S = 0
- Rock: rock (V
_{s}> 600m∕s), very thin soil (< 5m) over rock or soil 5 to 20m thick over rock. - S = 1
- Deep soil: deep soil in narrow canyon (soil > 20m thick and canyon < 2km wide) or deep soil in broad canyon (soil > 20m thick and canyon > 2km wide).

- All records reprocessed using common procedure. Interpolated to 400 samples/sec, low-pass filtering with corner frequency selected for each record based on visual examination of Fourier amplitude spectrum, instrument corrected, decimated to 100 to 200 samples/sec depending on low-pass corner frequency, baseline correction using 0 to 10 degree polynomial, high-pass filtered based on integrated displacements.
- Only use response spectral data within frequency band 1.25f
_{h}to 0.8f_{l}to avoid effects of filter roll-off. Hence number of records used for regression at each period varies, minimum number is less than 100 records for 0.01s. - Well distributed dataset in terms of magnitude and distance.
- Supplement data with records from Gazli, Friuli, Tabas, Taiwan, Nahanni and Spitak.
- Consider source mechanism: reverse ⇒ F = 1, reverse/oblique ⇒ F = 0.5, others (strike-slip and normal) ⇒ F = 0).
- Consider hanging wall effect: if over hanging wall HW = 1, otherwise HW = 0.
- Note that interpretation of c
_{4}is not clear for their distance measure but yields better fit. - Model nonlinear soil response by f
_{5}. - Model uncertainty as magnitude dependent.
- Fix some coefficients to be independent of period so that response spectral values vary smoothly with distance, magnitude and period.
- Smooth coefficients using piecewise continuous linear fits on log period axis. For highly correlated coefficients, smooth one coefficient and re-estimate other coefficients.

^{4}f_{3} given in Abrahamson and Silva (1997) was modified to ensure homogeneity and a linear variation in f_{3} with magnitude.

^{5}It is probably 5%.