- Ground-motion model is:
where y is in g, V

_{ref}= 1130m∕s, C_{1}= -4.407, C_{2}= 1.142, C_{3}= -0.012, C_{4}= -1.450, C_{5}= 0.140, C_{6}= 0.623, C_{7}= -0.389, C_{8}= -0.057, C_{9}= 0.188, τ = 0.322 (inter-event), σ = 0.530 (intra-event), τ_{S}= 0.230(site-to-site), σ_{R}= 0.477 (record-to-record sigma), τ_{P }= 0.337 (path-to-path), σ_{0}= 0.338 (remaining unexplained variability) and σ_{T }= 0.621 (total). - Use V
_{s,30}to characterise sites. 110 ≤ V_{s,30}≤ 1056.71m∕s. Good distribution of data between about 200 and 800m∕s. - Classify events into 3 mechanisms:
- S
- Strike-slip. F
_{N}= F_{R}= 0. - N
- Normal. F
_{N}= 1 and F_{R}= 0. - R
- Reverse. F
_{R}= 1 and F_{N}= 0.

- Focal depths between about 1 and about 32km with most between 5 and 20km.
- Use data from the Taiwan Strong Motion Instrumentation Program from crustal earthquakes between 1995 and 2009. All selected events have at least 50 records.
- Baseline correct and filter data using procedure of Pacific Earthquake Engineering Research Center.
- Most data from M
_{w}< 6.5 and R > 20km. - Develop maximum-likelihood method (the path diagram) based on mixed-effect model to quantify path
effects for each station. Divide record-to-record residuals into small brackets in a rose diagram for 6
source-to-site distance bins (< 50, 50–100, 100–150, 150–200, 200–250 and 250–300km) and 8 azimuth bins
(every 45
^{∘}). Then estimate mean residual for each path bin. Obtain 8 × 6 = 48 path bins (path-to-path residuals) at a site and hence compute repeatable path term for all path-to-path residuals for all stations. - Examine path diagrams for 8 stations in the Ilan Plain as an example. Find similar results.
- Examine path-to-path residuals w.r.t. azimuth and distance. Find no recognizable trends for azimuth. Path-to-path residuals become smaller as distance increases but note could be due to having fewer data at long distances.
- Estimate τ
_{P }and σ_{0}for each station. Find geographical patterns in these values. - Also apply Closeness Index (CI) approach (Lin et al., 2011a). Find similarities and differences (e.g. CI
approach leads to slightly higher σ
_{0}estimates) between results from the two methods.