- Ground-motion model is:
_{1}= 3.672, c_{1}= -1.940, c_{2}= 0.413, h = 10.322, c_{3}= 1.34 × 10^{-4}, b_{1}= -0.262, b_{2}= -0.0707, s_{A}= 0, s_{B}= 0.162, s_{C}= 0.240, s_{D}= 0.105, s_{E}= 0.570, f_{1}= -5.03 × 10^{-2}, f_{2}= 0.105, f_{3}= -5.44×10^{-2}, f_{4}= 0, σ_{B}= 0.172 (inter-event), σ_{W }= 0.290 (intra-event) and σ = 0.337 for horizontal PGA and e_{1}= 3.511, c_{1}= -1.741, c_{2}= 0.324, h = 9.052, c_{3}= 1.28×10^{-3}, b_{1}= 9.04×10^{-3}, b_{2}= -0.0270, s_{A}= 0, s_{B}= 0.167, s_{C}= 0.204, s_{D}= 0.190, s_{E}= 0.350, f_{1}= -7.09 × 10^{-2}, f_{2}= 7.79 × 10^{-2}, f_{3}= -6.96 × 10^{-3}, f_{4}= 0, σ_{B}= 0.160 (inter-event), σ_{W }= 0.270 (intra-event) and σ = 0.314 for vertical PGA (standard deviations of coefficients from bootstrap analysis also given but not reported here). After trial regressions and following Boore and Atkinson (2008) fix R_{ref}= 1, M_{ref}= 5, M_{h}= 6.75 and b_{3}= 0. - Use five site Eurocode 8 (EC8) classes (150 stations in total):
- A
- V
_{s,30}> 800m∕s. 334 records. C_{A}= 1 and other C_{i}s are zero. - B
- 360 < V
_{s,30}≤ 800m∕s. 99^{32}records. C_{B}= 1 and other C_{i}s are zero. - C
- 180 < V
_{s,30}≤ 360m∕s. 182 records. C_{C}= 1 and other C_{i}s are zero. - D
- V
_{s,30}≤ 180m∕s. 17 records. C_{D}= 1 and other C_{i}s are zero. - E
- 5–20m of C- or D-type alluvium underlain by stiffer material with V
_{s,30}> 800m∕s. 37 records. C_{E}= 1 and other C_{i}s are zero.

About 30% of classifications based on shear-wave velocity profiles and rest from geological and geophysical data.

- Use four faulting mechanism classes using classification of Zoback (1992):
- Normal
- 593 records. E
_{1}= 1 and other E_{i}s are zero. - Reverse
- 87 records. E
_{2}= 1 and other E_{i}s are zero. - Strike-slip
- 61 records. E
_{3}= 1 and other E_{i}s are zero. - Unknown
- 28 records. E
_{4}= 1 and other E_{i}s are zero.

Note that ‘unknown’ could be dominated by earthquakes of another class (e.g. normal).

- Use data from the Italian strong-motion database ITACA, which has been updated in various studies, including additional local site information, and all the records have been individually reprocessed.
- Note that the L’Aquila 2009 earthquake (M
_{w}6.3) adds considerable data over a previously poorly-sampled magnitude-distance range. - Firstly select data with M > 4, r
_{epi}< 200km and focal depths h < 35km. This leaves 1213 records from 218 earthquakes and 353 stations. Note that for M < 4.5 only M_{L}is available for most earthquakes. Also many stations recorded only one earthquake. Therefore, exclude earthquakes without M_{w}, those only recorded by one station and those stations with only one record. Finally data from 150 stations. - Most earthquakes on normal faults in central and southern Appennines with h < 20km. Some reverse-faulting earthquakes in north-eastern Italy and northern Apennines with h > 15km. Strike-slip earthquakes generally in southern Italy with h between 10 and 30km.
- Only about 60 records from M
_{w}≥ 6 (roughly evenly spread for smaller magnitudes). Most data from 10 < r_{jb}< 100km and most records with r_{jb}< 5km are from M_{w}≤ 6. - Process records by: 1) baseline correction; 2) application of cosine taper, except for late-triggered records; 3) select bandpass filter cut-offs based on visual inspection of Fourier amplitude spectrum; 4) application of a second-order acausal time-domain Butterworth filter to zero-padded record; 5) double integration to displacement; 6) linear detrending of displacement; and 7) double differentiation to obtain final acceleration.
- Regress twice. Once to find inter-event and intra-event standard deviations and secondly to find inter-station and intra-station standard deviations, which are not explicitly given in the text but are shown on graphs.
- Analyze trade-off between coefficients by studying off-diagonal elements of unit covariance matrix (shown for 0.1 and 1.0s). Find strong trade-offs for some coefficients.
- Note that care should be taken when considering site coefficients for classes D and E because they are based on limited records from only a handful of stations (i.e. for class E, 33 out of 27 records are from Nocera Umbra and for D, 12 records are from Colfiorito and 5 from Norcia). Because of this, remove data from these classes and re-regress. Find little difference in median predictions (less than 10%) (not shown). Hence conclude that data from D and E are not having a large impact on results.
- Constrain coefficient for unknown faulting class to zero and sum of coefficients for other classes to zero
so that offset coefficient e
_{1}corresponds to average effect of faulting mechanism. Also tried various other constraints on faulting coefficients but find similar median predictions. - Compare predicted and observed PGAs for Molise 2002, Friuli 1976, Irpinia 1980 and five M
_{w}4.6 earthquakes. Find generally good agreement. - For T < 0.2s the inter-station component of variability is larger than the inter-event component while for T > 0.4s the two are similar. By comparing variabilities to a previous model (Bindi et al., 2010) using a different site classification conclude that EC8 classification improves prediction of long-period motion for soft and very soft sites but may not be suitable for short-period site response of Italian sites.
- Find that inter- and intra-event residuals (shown for PGA, 0.1s and 1.0s) well behaved.

^{32}This value is given in the text (p. 1903) but they probably mean 199 records since otherwise the total is 669 (also see their
Figure 4).