- Ground-motion model is:
where Y is in cm∕s2, δ

_{i}= 1 at ith receiver and 0 otherwise, for horizontal PGA a = 0.918 and b = -0.00846 (σ not given) and for vertical PGA a = 0.865 and b = -0.00741 (σ not given). c_{i}given in paper but are not reported here due to lack of space. - Data from three vertical arrays in Japan so predictions at surface and at different depths down to 950m.
- Different definition of M
_{JMA}for focal depths > 60km so exclude such data. Focal depths between 2 and 60km. - Exclude data from earthquakes M < 5.0 because errors are larger for smaller events.
- Exclude data for which predicted, using a previous attenuation relation, PGV < 0.1cm∕s in order to find precise attenuation rate.
- Most data from earthquakes with M ≤ 6.0 and most from X ≤ 100km.
- Records low-pass filtered with cutoff frequency 25Hz for records from 2 sites and 30Hz for records from 1 site.
- Use two-stage method because positive correlation between M and X. Also apply one step; find it is biased and two-stage method is most effective method to correct bias.
- Check residuals (not shown) against M and X find no remarkable bias.