Engineering Structures 29 (2007) 1422–1441

www.elsevier.com/locate/engstruct

The scaled nonlinear dynamic procedure

Mark Aschheim a,∗ , Tjen Tjhin b , Craig Comartin c , Ron Hamburger d , Mehmet Inel e
a Civil Engineering Department, Santa Clara University, Santa Clara, CA, USA
b Buckland & Taylor Ltd., North Vancouver, British Columbia, Canada
c CDComartin Inc., Stockton, CA, USA
d Simpson Gumpertz & Heger, San Francisco, CA, USA
e Department of Civil Engineering, Pamukkale University, Kinikli Kampusu, Denizli, Turkey

Received 17 June 2004; received in revised form 14 July 2006; accepted 24 July 2006
Available online 13 October 2006

Abstract

Although nonlinear static procedures (NSPs) have become widely accepted for use in seismic design and evaluation in recent years, their
accuracy is poor for response quantities that are significantly affected by the vibration of multiple degrees of freedom (termed MDOF effects).
In recent work performed for the ATC-55 project, a design-oriented approach, called the scaled nonlinear dynamic procedure (Scaled NDP), was
identified for determining such response quantities for nonlinear systems. The Scaled NDP provides an alternative to current code approaches for
scaling ground motions for dynamic analyses and is readily used in performance-based seismic design and evaluation. The Scaled NDP appears
to provide a valid basis for establishing force quantities at stated levels of confidence and provides an indication of deformation demands for use
in design and evaluation. The results can be used to determine the strengths required of members in order to ensure that ductile behavior develops,
and to evaluate the deformation performance of a given design.
c 2006 Elsevier Ltd. All rights reserved.

Keywords: Dynamic analysis; Earthquake engineering; Nonlinear response

1. Introduction The procedures contained in all three documents use similar

Nonlinear static procedures (NSPs), or pushover methods of
analysis, have become popular means of predicting earthquake
demands in performance-based seismic design approaches.
These methods of analysis are most commonly used for the
evaluation of existing buildings and are generally believed
to be superior to elastic methods of analysis for predicting
seismic performance. In use for many years in the petroleum
industry, for the evaluation of offshore structure response
to extreme wave loading, their use was first introduced
to seismic design with the publication, in the mid-1980s,
of the Tri-Services Seismic Design Guidelines for Essential
Facilities [14]. Use of these procedures was popularized for
seismic design applications with the publication of two seismic
retrofit criteria documents, ATC-40 [3] and FEMA 273 [16].
∗ Corresponding address: Civil Engineering Department, Santa Clara
University, 500 El Camino Real, Santa Clara, CA 95053, USA. Tel.: +1 408
554 6871; fax: +1 408 554 5474.
E-mail address: maschheim@scu.edu (M. Aschheim).
methods of applying patterns of lateral forces to a mathematical
model of the building. The pattern of forces is typically
selected to match the pattern of inertial forces that would be
produced in the structure when responding in its fundamental
mode, or as an approximation of multi-modal response. The
forces are scaled to a level of “first yielding” in which the
weakest element is predicted to reach its elastic limit. The
mathematical model is then modified to represent the post-
elastic stiffness of the “yielding” element, and the forces
are re-applied to the model and scaled until additional
elements are determined to reach their elastic limits. At each
loading increment, the acceptability of element behavior is
evaluated by examination of inelastic deformation demands
and other response parameters. This process is repeated until
structural instability or other unacceptable structural behavior is
predicted. The results of the analysis are commonly portrayed
in the form of a plot of total applied lateral force against lateral
roof displacement, termed a “pushover curve”. This curve is
then used, together with elastic response spectra, to estimate the

c 2006 Elsevier Ltd. All rights reserved.

0141-0296/$ - see front matter

doi:10.1016/j.engstruct.2006.07.020
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
global displacement demand on the structure under particular
ground shaking intensities, and the acceptability of behavior of
the various elements of the structure at these demand levels is
assessed.
Each of the design guidelines that employ NSPs to predict
seismic performance uses different procedures to estimate the
displacement demand for a given ground motion spectrum. The
ATC-40 procedures employ a “substitute structure” method in
which the demand displacement is determined based on the
response of a linear single-degree-of-freedom (SDOF) structure
having effective stiffness and damping properties obtained
from the pushover curve. The FEMA 273 procedures use
empirical coefficients to modify elastic displacement demands
for the effects of nonlinearity considering the structure’s period,
ductility and hysteretic behavior. For a given structure, the
two methods may produce different estimates of displacement
demand, a problematic situation for designers. Of perhaps
greater concern, the basic analytical process seems incapable
of correctly accounting for the effects of higher modes of
response (termed “MDOF effects” for nonlinear systems).
Finally, because each of the NSPs predicts a single pattern and
magnitude of deformation demand at a given ground motion
intensity level, these procedures mask the inherent variability
in nonlinear structural response at a given ground motion
intensity level and create an inappropriate understanding of
the predictability of nonlinear structural response. Nonlinear
dynamic analyses avoid many of these shortcomings but are
infrequently used for design or evaluation, because variability
in response quantities must be addressed by conducting a
number of separate dynamic analyses using ground motions
that are scaled to match a design response spectrum.
Techniques to account for multi-degree-of-freedom (MDOF)
effects have been proposed to improve the NSP (e.g. [3,7,4,
24]). These techniques introduce approximations, such as con-
sidering responses in each “mode” independently or consider-
ing the response of multiple modes in an incremental fashion,
reflecting a particular pattern of development of nonlinear be-
haviour within the structure. The conditions under which these
approximations provide results of adequate fidelity are unclear;
experience indicates that relatively simple approaches are
not uniformly reliable, while modifications that purportedly
improve fidelity make the methods mathematically com-
plex, while still masking the variability inherent in nonlinear
response.
A comprehensive treatment of variability may be considered
using approximate techniques such as those suggested
in FEMA 351 [17]. Incremental dynamic analysis [26]
emphasizes randomness in potential ground motion excitations
(record-to-record variability). Numerical evaluation of response
over the complete hazard curve is computationally intensive,
and the accuracy and robustness of the component and damping
models is less certain as the severity of damage increases. The
modeling and prediction of collapse is an emphasis of current
research in earthquake engineering.
From a pragmatic point of view, the influence of MDOF
effects on response should be considered where these effects
may have significant consequences on performance. Aleatory
1423
randomness in ground motions (at a given spectral intensity)
causes the intensity and timing of higher mode peaks to
vary. As a result, ground motion variability can introduce
dispersion in those response quantities that are significantly
affected by higher modes, while the consideration of multiple
modes can shift the central tendency of a response quantity
relative to a prediction based on the first mode. The Scaled
NDP directly uses the results of nonlinear dynamic analyses of
MDOF models. Variability in response quantities (attributable
to aleatory randomness in ground shaking as reflected in the
excitement and interaction of multiple degrees of freedom)
is made apparent to the user, and the effects of sample
size are considered in estimating response quantity statistics.
The Scaled NDP is intended for use in design at one or
several discrete hazard levels, for levels of inelastic response
that do not involve substantial degradation and are far short
of collapse. Because the method relies upon the computed
dynamic response of the inelastic structure, approximations
that may restrict the generality of other methods are absent.
Extensions to consider additional sources of variability, such as
uncertainties in material properties, member dimensions, and
modelling idealizations could be introduced, but are not the
object of the current paper.
Although NSPs were originally presented for use in
evaluation, they may also be applied to design, as suggested by
Fajfar [15], Chopra and Goel [6], and Aschheim and Black [1].
These methods generally rely on the use of an equivalent SDOF
(ESDOF) system for preliminary proportioning to address
peak displacement and system ductility responses. The ESDOF
estimates of roof displacement provide reasonably accurate
estimates of peak displacement response at levels of inelasticity
far short of collapse. Techniques such as the Scaled NDP may
then be used to establish the expected performance of the
preliminary design and to determine design values for those
components of the structure that are intended to remain elastic
as inelastic response develops elsewhere within the structure.
2. Higher mode contributions to response quantities
The pushover curve determined in a NSP reflects the
displacement response of a reference point in the structure,
usually, the center of mass of the roof. In some forms of
application, modal participation factors are used to convert
the domain of the plot from base shear vs. reference
displacement to spectral response acceleration vs. spectral
response displacement, with the latter plot characterizing the
response of the ESDOF oscillator. Values of various response
quantities (e.g. story shears and plastic hinge rotations) are
determined as the values computed in the nonlinear static
(pushover) analysis at the instant in the analysis at which the
reference displacement is equal to the estimated displacement
demand. Modifications developed in the ATC-55 project to
improve the accuracy of the demand displacement estimates are
reported in the FEMA 440 [18] document.
The load pattern used in pushover analyses generally is
similar to a first mode pattern. Many researchers (e.g. [22,
10,13,8]) have shown that such approaches can lead to good
estimates of peak displacements for MDOF structures, typically
1424 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
overestimating the roof displacement to a small degree. The
success of these first and quasi-first mode pushover approaches
can be attributed to the relatively small contribution of higher
modes to displacements. This can be understood for elastic
response by noting that the vector of peak displacements due
to the ith mode, xi , is given by
xi = Γi Sd (Ti )φ i (1)
where Sd (Ti ) is the spectral displacement associated with the
period Ti , Γi is the modal participation factor, and φ i is the
mode shape for the ith mode. Higher mode contributions to
displacements typically are minor because both Γi and Sd
typically are much smaller for the 2nd and higher modes,
relative to their 1st mode values.
Higher modes can contribute more significantly to other
quantities, such as story shears and interstory drifts. This can
be appreciated by noting for elastic response that the vector
of lateral forces, Fi , associated with developing the peak
displacements, xi , can be expressed as
Fi = Γi Sa (Ti )Mφ i (2)
where M is the mass matrix and Sa (Ti ) is the spectral pseudo-
acceleration associated with the period Ti . Because the shape of
the response spectrum will often result in higher mode spectral
pseudo-accelerations that are similar to or larger than those
of the first mode, the lateral forces typically have significant
contributions from the higher modes.
These observations were extended to structures responding
inelastically by applying Principal Components Analysis to
dynamic response data [2]. In general, displacements often may
be approximated by a quasi-first mode pattern, while interstory
drifts, story shears, and other response quantities can have
significant contributions from two or more modes.
3. ATC-55 MDOF studies
The Scaled NDP was developed from an observation made
in studies of MDOF systems for the ATC-55 project. The
MDOF studies were conducted to illustrate the accuracy of
several static pushover analysis techniques in relation to the
results obtained from nonlinear dynamic analysis. Five building
models were used, consisting of an 8-story reinforced concrete
wall building previously used as an example in ATC-40, 3- and
9-story steel moment-resistant frames previously used in the
SAC program [25], and two variants of these frames in which
weak stories were introduced at the lowest story. Only a sample
of the results obtained for these five buildings is presented here.
The 3- and 9-story steel moment frames were designed and
modeled as part of the SAC Joint Venture [25]. The “pre-
Northridge” designs developed for Los Angeles were modeled
using a lumped plasticity beam–column element (Element 02)
in Drain-2DX [23]. These elements extended along the beam
and column centerlines as shown in Fig. 1 for the 9-story
frame. Beam–column joints were not modeled and rigid-end
offsets were set to zero, as was done for the SAC “M1” models.
P–Delta effects were considered for all building models, using
dead loads in combination with 40% of the design live loads.
Tributary gravity load was applied to the frames and the
remainder was applied to a “dummy” column that was used to
obtain a truss-bar approximation of P–Delta effects. Rayleigh
damping was set equal to 2% of critical viscous damping at
the first mode period and at a period of 0.2 s, as was done in
the SAC program. The fundamental period of the 9-story frame
was 2.34 s for the model with P–Delta effects.
The 8-story reinforced concrete shear wall model is based
on the Escondido Village building that is described in ATC-40.
0 00
Typical floor heights are 2.77 m (9 -1 ), as shown in Fig. 2.
A two-dimensional model of the wall was developed using
a beam–column fiber element (Element 15) in Drain-2DX.
Rayleigh damping (proportional to mass and stiffness) was
set equal to 5% of critical viscous damping in the first and
fourth modes of vibration. The fundamental period was 0.71 s,
determined according to the tangent stiffness properties of the
cracked wall at a base shear equal to 60% of the effective yield
strength.
The recorded ground motions (Table 1) were selected to
represent motions that potentially could occur at a given
site, characterized by NEHRP Site Class C soil conditions
(as defined in BSSC [5]), with magnitudes (Ms ) between
6.6 and 7.6, and epicentral distances of between 8 and 20
km. To investigate accuracy as a function of drift level, the
records were each scaled to achieve a peak roof displacement
of 0.5%, 2%, or 4% of the height for the steel frames and
0.1%, 1%, and 2% of the height for the concrete wall. These
drift levels represent values expected for many steel frame
and concrete wall buildings and represent a moderate degree
of inelastic response; one that can be represented reasonably
well with traditional nonlinear models of component response.
The lowest drift for each building type corresponds to elastic
response.
The pushover techniques consisted of quasi-first mode
load vectors (consisting of first mode, inverted triangular,
rectangular or uniform, and “code” load vectors, an “SRSS”
load vector, and an adaptive first mode vector), as well as a
modified multi-mode pushover analysis (MPA) method. The
“code” load vector consists of lateral loads in proportion
to the seismic weight and height of each floor raised to a
power k, where the exponent k depends on the period of
the structure, in the manner described in the International
Building Code [20]. The SRSS technique applies the lateral
forces required to generate a pattern of story shears; the story
shears are determined as an SRSS combination of the story
shears obtained from elastic modal responses. Three modes
were included in the SRSS combinations to represent at least
90% of the mass, and modal shears were determined based
on an elastic response spectrum. The modified MPA method
uses an SRSS combination of the 2nd and 3rd modes assuming
elastic response and potentially inelastic contributions of the 1st
mode. This modification to the MPA method is described in the
FEMA 440 document and by Chopra et al. [9]. Priestley and
Amaris [24] also use this approach to estimate wall shears, but
introduced a modification for estimating wall moments.
The quasi-first mode load vectors were applied until the
roof displacement was equal to the pre-determined demand
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441 1425

Fig. 1. Elevation view of 9-story frame.

Table 1
Details of the ground motions

# Identifier Earthquake Date Magnitude Station location Component PGA PGV Char. Source
(g) (cm/s) Period (s)
Ordinary
1 ICC000 Superstitn November 24, 1987 Ms = 6.6 01335 El Centro Imp. Co. Cent 000 0.358 46.4 0.60 CDMG
2 LOS000 Northridge January 17, 1994 Ms = 6.7 90057 Canyon Country–W Lost 000 0.41 43 0.59 USC
Cany
3 G02090
Loma October 18, 1989 Ms = 7.1 47380 Gilroy Array #2 090 0.322 39.1 0.69 CDMG
Prieta
4 TCU122N Chi-Chi, September 20, Ms = 7.6 TCU122 N 0.261 34 0.85 CWB
Taiwan 1999
5 G03090 Loma October 18, 1989 Ms = 7.1 47381 Gilroy Array #3 090 0.367 44.7 0.40 CDMG
Prieta
6 CNP196 Northridge January 17, 1994 Ms = 6.7 90053 Canoga Park–Topanga Can 196 0.42 60.8 0.61 USC
7 CHY101W Chi-Chi, September 20, Ms = 7.6 CHY101 W 0.353 70.6 1.27 CWB
Taiwan 1999
8 ICC090 Superstitn November 24, 1987 Ms = 6.6 01335 El Centro Imp. Co. Cent 090 0.258 40.9 1.03 CDMG
9 CNP106 Northridge January 17, 1994 Ms = 6.7 90053 Canoga Park–Topanga Can 106 0.356 32.1 0.45 USC
10 E02140 Imperial October 15, 1979 Ms = 6.9 5115 El Centro Array #2 140 0.315 31.5 0.29 USGS
Valley
11 E11230 Imperial October 15, 1979 Ms = 6.9 5058 El Centro Array #11 230 0.38 42.1 0.27 USGS
Valley
1426 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441

(a) Wall elevation. (b) Wall cross section: basement to 1st floor.

(c) Wall cross section: 1st floor to 3rd floor. (d) Wall cross section: 3rd floor to 5th floor.

(e) Wall cross section: 5th floor to 7th floor. (f) Wall cross section: 7th floor to roof.

Fig. 2. Elevation view and sections of 8-story wall.

displacement, in order to identify differences in response
quantities associated with the choice of load vector. The
response quantities (floor displacement, interstory drift, story
shear, and overturning moment at each level of the building)
were determined at this displacement for the quasi-first mode
load vectors. The ground motions were scaled individually to
obtain the same predetermined peak roof displacement in the
nonlinear dynamic analyses of each building. Higher mode
contributions in the modified MPA were determined based on
the mean of the elastic spectra associated with the ground
motions after scaling to achieve each predetermined drift level.
Some results are illustrated in the following.
Fig. 3 compares estimates of the interstory drifts (Fig. 3(a))
and story shears (Fig. 3(b)) made using various pushover
methods with the range of values computed by nonlinear
dynamic analysis, for the SAC 9-story steel frame at a roof
drift of 4% of the height of the building. The bar symbol
at each floor (or story) indicates the minimum, maximum,
mean, and mean plus and minus one standard deviation results
obtained from the 11 dynamic analyses; the “+” indicates the
median value. Multiple-degree-of-freedom effects are reflected
in the results obtained in the nonlinear dynamic analyses of
the yielding structure, but are absent from the quasi-first mode
pushover results. The more complex multiple mode calculation
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441 1427

Fig. 3. Comparison of NSP estimates and values computed by nonlinear dynamic analysis using 11 ground motion records scaled to achieve a roof drift of 4%, for
the 9-story steel frame building: (a) interstory drifts, and (b) story shears. (Note: 1 kip = 4.448 kN.)

is often an improvement over the quasi-first mode estimates, but
the estimates obtained by this approach were not consistently
reliable, with significant errors developing for some cases.
Fig. 4 compares estimates of story shears (Fig. 4(a)) and
overturning moments (Fig. 4(b)) made using various pushover
methods with the range of values computed by nonlinear
dynamic analysis, for the 8-story reinforced concrete wall
building at a roof displacement equal to 2% of the height of
the building. Again, MDOF effects cause the dynamic peaks
to be systematically higher than the quasi-first mode pushover
estimates.
Also shown in Figs. 3 and 4 (with circles and diamonds)
are peak values obtained using two additional ground motions,
each scaled to achieve the same predetermined roof drift in
the nonlinear dynamic analyses. The results of these individual
ground motion records are consistent with those obtained for
the 11 ground motions. This finding was repeated for each
of the five frame and wall building models considered in the
ATC-55 studies. That a single nonlinear dynamic analysis could
provide results of higher fidelity than a static pushover analysis,
or the equivalent lateral force analyses specified by building
codes, for structures in which the interaction of multiple
degrees-of-freedom is important, should not be surprising
because MDOF effects are not adequately represented in these
static analysis methods. This observation forms the basis of the
Scaled NDP described below.
4. The scaled NDP
4.1. Description of the method
Step 1. Given a smoothed elastic response spectrum
representative of the site hazard of interest, estimate the
demand displacement of the roof (or more generally, a “control
point”) using the displacement coefficient or capacity spectrum
approach. Improved methods developed by the ATC-55 project
may be used for this purpose, as described in FEMA 440 [18].
Step 2. Select n ground motion records that reflect the
characteristics of the hazard (e.g. magnitude, distance, site
class) and for each record, conduct a nonlinear dynamic
analysis, with the record scaled iteratively until the peak
displacement of the control point is equal to the estimate
determined in Step 1. Extract peak values of the response
quantities of interest from the results of each analysis and
compute the sample mean, x̄n , of each peak quantity of interest.
At this preliminary stage, at least three analyses (n ≥ 3) are
suggested.
Step 3. Although the sample mean, x̄n , is the best estimate
of the population mean, sampling error may be present due
to the limited number of observations (n) of each quantity.
Furthermore, estimates of response quantities may be desired at
the mean plus κ standard deviation level and at a particular level
of confidence. Thus, the estimate at the mean plus κ standard
1428 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441

Fig. 4. Comparison of NSP estimates and values computed by nonlinear dynamic analysis using 11 ground motion records scaled to achieve a roof drift of 2%, for
the 8-story reinforced concrete wall building: (a) story shears, and (b) overturning moments. (Note: 1 kip = 4.448 kN and 1 kip ft = 1.356 kN m.)

Table 2
Values of c at the 90% confidence level
−1
n Φt,n−1 (α) Coefficient of variation
0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
3 1.886 1.06 1.12 1.20 1.28 1.37 1.48 1.62 1.77 1.96 2.19
5 1.533 1.04 1.07 1.11 1.16 1.21 1.26 1.32 1.38 1.45 1.52
7 1.440 1.03 1.06 1.09 1.12 1.16 1.20 1.24 1.28 1.32 1.37
10 1.372 1.02 1.05 1.07 1.10 1.12 1.15 1.18 1.21 1.24 1.28
20 1.328 1.02 1.03 1.05 1.06 1.08 1.10 1.12 1.13 1.15 1.17
50 1.299 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08 1.09 1.10
100 1.290 1.01 1.01 1.02 1.03 1.03 1.04 1.05 1.05 1.06 1.07

deviation level is given by the quantity c(1 + κCOV)x̄n . As
shown in Appendix A, this quantity exceeds the population
mean plus κ standard deviation level with confidence level
α if the response quantities are normally distributed. In the
preceding, c is given by
1 population at a confidence level of 90%. The term κ assumes
c=
−1
1 − Φt,n−1 (α) COV
√
n sought. The coefficient of variation may be estimated as the
where COV is the coefficient of variation of the sample of
n observations of response quantity x, and Φt,n−1−1
(α) is the
variate of the Student’s t-distribution with n − 1 degrees of
freedom at a confidence level α. Eq. (3) simplifies to c = 1 for
a confidence level, α, of 50%. For a confidence level of 90%,
values of c determined using Eq. (3) are given as a function
of sample size and COV in Table 2. Table 2 indicates that
the benefit of running additional analyses (indicated by the
reduction in c) is a function of the dispersion (or COV) of the
response quantity.
Thus, using an appropriate value of c from Table 2, the
quantity c(1 + κCOV)x̄n would be expected to exceed the
population mean plus κ standard deviation value for the
(3) a value of zero where estimates of the population mean are
sample COV for large samples (perhaps for n ≥ 7). For
smaller sample sizes, it is tentatively suggested that the COV
be assumed equal to a baseline value of perhaps 0.30, based
on the results of the multi-degree-of-freedom studies. Aleatory
randomness in material properties or member dimensions,
and epistemic uncertainties in the modeling and response
computation could have been considered in determining the
sample of response quantities.
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
The nonlinear static analysis of Step 1 typically requires the
greatest effort, primarily for developing a nonlinear model of
the structure. The dynamic analyses required in Step 2 are fairly
straightforward. Experience suggests that perhaps five to ten
runs are needed to converge on the target roof displacement,
with subsequent ground motions scale factors estimated on the
basis of the results of previous iterations. The analyses are much
like the “stripe” analyses done in the Incremental Dynamic
Analysis (IDA) method [26], where demands are determined
for a given intensity measure (the intensity measure in this
case is the target roof drift). Software such as Drain-2DX,
OpenSees, RAM-Perform, RUAUMOKO and SAP-2000 may
be used. Much or all of the analytical work can be automated.
As noted by Vamvatsikos and Cornell [27] even for the much
larger number of analyses required for IDAs, preparation of
the structural model requires substantially more time than
computation of the responses.
4.2. Illustration of the method
It is anticipated that the NSP will be used in preliminary
design to determine the strength and stiffness required for the
structure to satisfy global performance criteria, or to evaluate
the demand displacement of an existing structure. Once the
proportions of the structural members are established and the
demand displacement has been determined, the Scaled NDP
may be used to assess or characterize the performance of the
structure, or to determine some quantities required for design of
those portions of the structure that are intended to remain elastic
while inelastic behaviour develops elsewhere at prescribed
locations. An example from the ATC-55 MDOF studies is used
to illustrate estimates of interstory drift and story shear.
4.2.1. Interstory drift estimate
The sample mean of the peak values of interstory drift at the
lowest story of the 9-story frame at a predetermined roof drift
of 4% is x̄n = 6.5% (Fig. 3(a)). The COV is estimated from
the 11 peak dynamic responses to be 0.16. For this COV and
n, Eq. (3) results in c = 1.07. The mean peak interstory drift
is estimated to not exceed c x̄n = 1.07 (6.5%) = 7.0% at the
90% confidence level. That is, there is a 90% probability that
the mean peak interstory drift at the lowest story is less than
7.0% for the hazard level that produces a roof drift of 4%.
4.2.2. Story shear estimate
The sample mean of the peak story shears (Fig. 4(a)) at
the lowest story of the 8-story wall at a predetermined roof
drift of 2% is x̄n = 4.76 MN (1070 kips). To guard against
the potential for shear failure, a reasonable upper bound shear
demand is needed for the design of the wall. Based on the
11 analyses, the COV of the story shears for the population
is estimated to be 0.22. Using Eq. (3), c = 1.10. Therefore,
there is a 90% probability that the mean plus one standard
deviation peak story shear is less than (1 + κCOV)c x̄n =
(1 + 0.22)(1.10)(4.76 MN) = 6.39 MN (1436 kips), for the
hazard that produces a roof drift of 2%.
1429
4.3. Observed coefficients of variation
The coefficients of variation (COVs) of the response
quantities determined in the MDOF studies were examined for
each response quantity at each floor or story for each of the
five building models, at each of the three predetermined drift
levels. In general, the COVs differ for each response quantity
and for these quantities the COVs are highest at the upper
stories and near the base of each model. The COVs of several
response quantities determined for two of the building models
are plotted in Figs. 5 and 6, at each of the three drift levels.
The COVs for floor displacements diminish to zero at the top,
due to the methodology employed in the study. Approximate
upper bounds to the COVs are summarized in Table 3, where
“approximate” indicates that the limit was exceeded by a small
amount at a limited number of locations. Until more detailed
information is available, it is suggested that a COV of perhaps
0.30 be used for all quantities where the number of analyses (n)
is insufficient to establish a better estimate of the population
COV. Larger COVs would be expected if additional sources
of variability were included, such as those associated with
uncertainties in material properties, member dimensions, and
modelling idealizations.
4.4. Dependence of sample mean and COV on sample size
Data generated in the ATC-55 studies was re-interpreted
in order to observe the influence of n on the sample mean
and sample COV. Three sequences of the eleven ground
motions used in the original analyses were randomly selected
and statistics on the peak response quantities (displacement,
interstory drift, story shear, and overturning moment) were
computed for the first n records of each sequence, for 2 ≤ n ≤
11. Results are presented in Fig. 7 for selected locations in the
9-story steel frame and in Fig. 8 for selected locations in the 8-
story reinforced concrete wall. The figures illustrate a reduction
in scatter as n increases, although the presence of sampling
error must be presumed even for n = 11. One may interpret
the figures as supporting the use of n ≥ 3 for determination of
x̄n and n ≥ 7 for determination of the COV.
4.5. Discussion
Most analysis methods used for design produce single-
valued (deterministic or point) estimates of design quantities,
even though substantial variability due to randomness and
uncertainty is widely acknowledged in earthquake engineering.
While nonlinear dynamic analysis generally is considered
to be the most accurate of the available analysis methods,
the substantial variability in computed results has been
cumbersome to address. The simplifying assumptions in other
analytical techniques (such as the neglect of higher modes,
assumptions of linearity, or assumptions of independence of
each potentially nonlinear “modal” response), may produce
design quantities that bear little relation to the demands
expected on the basis of nonlinear dynamic analysis. The
Scaled NDP inherently reflects the interaction of multiple
1430 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441

Fig. 5. COVs (in percent) of (a) interstory drifts and (b) story shears determined by nonlinear dynamic analysis using 11 ground motion records scaled to achieve
roof drift of 0.5%, 2%, and 4%, for the 9-story steel frame building.

Table 3
Approximate upper bounds to the COVs over the height of each building model

Building model Interstory drift Story shear Overturning moment

3-story steel frame 0.15 0.15 0.15
3-story steel frame (weak story) 0.20 0.15 0.15
8-story reinforced concrete wall 0.10 0.20 0.15
9-story steel frame 0.20 0.20 0.20
9-story steel frame (weak story) 0.30 0.25 0.25

degrees of freedom in the nonlinear dynamic response and
accounts for (1) capacity limits on demands that are associated
with mechanism development, (2) the influence of gravity
loads on response quantities, (3) the dispersion associated with
aleatory randomness in the excitation of multiple degrees of
freedom (which varies with the response quantity of interest,
the location within the structure, and the degree of inelasticity
(or drift) that develops), and (4) the influence of sample size on
estimates of population statistics.
Higher modes (or MDOF effects) generally have a small
contribution to the roof displacement relative to that associated
with the first mode equivalent SDOF system. In effect,
the Scaled NDP removes the dispersion associated with the
inelastic response of a SDOF oscillator, while preserving
the dispersion associated with higher mode response (MDOF
effects) on the nonlinear response of the structure, as described
in the following section.
5. Comparison with other scaling procedures
Cornell and his coworkers describe the search for
intensity measures that are efficient, sufficient, and computable
(e.g. [19]). An efficient intensity measure results in relatively
small variability in the response quantity and therefore allows
statistically meaningful estimates to be made using a reduced
number of analyses. Sufficiency refers to the absence of a
conditional dependence on other ground motion characteristics,
such as magnitude, duration, and source-to-site distance.
Computability refers to the effort required to define the site-
specific hazard curve in terms of the intensity measure.
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
Fig. 6. COVs (in percent) of (a) story shears and (b) overturning moments determined by nonlinear dynamic analysis using 11 ground motion records scaled to
achieve roof drift of 0.2%, 1%, and 2%, for the 8-story reinforced concrete wall building.
The peak roof displacement or its estimate is one of many
possible intensity measures. Cordova et al. [12] describe an
intensity measure based on elastic response that accounts for
period elongation. Cornell and Luco [11] describe intensity
measures that are functions of the first and second mode
responses, wherein the first mode response may be elastic
or inelastic, and the second mode response is elastic where
included. In the categorization described by Mackie and
Stojadinovic [21], the peak roof displacement would be a
Class III intensity measure. However, the roof displacement is
predominantly associated with the first mode response and thus
is closely correlated to the peak displacement of the (nonlinear)
equivalent SDOF system, which is a Class II intensity measure.
Mackie and Stojadinovic find that the peak displacement of the
equivalent SDOF system results in reduced dispersion relative
to Sa (T1 ), i.e. the spectral acceleration associated with the first
mode period, when estimating the peak displacement of bridge
structures.
Determination of the actual roof displacement hazard
curve is a cumbersome proposition, as it would require a
computationally intensive nonlinear seismic hazard analysis
to be done for any building of interest. Using the target
roof displacement as an intensity measure is a convenient
simplification, and relies on approximate relationships to
estimate the peak roof displacement on the basis of the
response of an inelastic ESDOF system in conjunction with
conventional elastic response spectra to characterize the hazard.
(As an alternative, the response of the ESDOF system could
1431
be estimated using inelastic spectral descriptions of the hazard,
derived directly from attenuation relations developed for
inelastic response.) One may consider that the contribution
of higher modes to roof displacement response introduces
dispersion that is not reflected in ESDOF estimates of the target
roof displacement. Alternatively, one may view the target roof
displacement as a good estimate of the displacement expected
for a given hazard level.
To better understand the significance of approximations
inherent in using the target roof displacement as an intensity
measure, results obtained using the Scaled NDP are compared
with those obtained using two other intensity measures. These
are (a) the spectral acceleration associated with the first mode
period, denoted Sa (T1 ), and (b) the peak displacement of the
equivalent SDOF system. In order to obtain a consistent set
of ground motion records for use with the three intensity
measures, scaling procedures were adopted as follows:
Scaling Method I is the Scaled NDP, wherein the amplitude
of each ground motion record is scaled such that the peak roof
displacement in the nonlinear dynamic response is equal to
the target roof drift (or demand displacement), which would
be estimated using an equivalent SDOF (ESDOF) system in
conjunction with a nonlinear static analysis. This method is
referred to as scaling to match the target roof drift in subsequent
text and figures.
For Scaling Method II, peak displacements of the ESDOF
systems are determined by nonlinear dynamic analysis for
the ground motions as originally scaled according to Scaling
1432 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441

Fig. 7. Means and COVs as a function of n for the 9-story frame at 2% drift.

Method I. Then, the mean of these peak displacements is
determined, and each ground motion record is rescaled such
that the peak displacement of the ESDOF system determined by
nonlinear dynamic analysis matches the mean. In this way, the
dispersion in the peak displacements of the inelastic ESDOF
system is reduced to zero, while the mean is identical to that
associated with the peak dynamic response of inelastic ESDOF
systems to the ground motion records used in Scaling Method I.
This method is referred to as scaling to match the inelastic
displacement in subsequent text and figures.
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441 1433

Fig. 8. Means and COVs as a function of n for the 8-story wall building at 1% drift.

For Scaling Method III, the mean elastic spectrum is equivalently, peak spectral displacement) at the fundamental
determined for the ground motions as originally scaled elastic period matches the mean. In this way, the dispersion of
according to Scaling Method I. Each ground motion is then the elastic spectral amplitudes (at T1 ) is reduced to zero, while
rescaled such that the peak pseudo-spectral acceleration (or the mean is identical to that associated with the response of
1434 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
Fig. 9. Response spectra for ground motions used for analysis of the 9-story steel moment-resistant frame, scaled according to Scaling Methods I, II, and III.
elastic ESDOF systems to the ground motions used in Scaling
Method I. This method is referred to as scaling to match the
elastic displacement in subsequent text and figures.
For the analyses reported in this section, P–Delta
effects were excluded and the reinforced concrete wall
was modeled with beam–column elements having a bilinear
moment–rotation behavior, in order to make the interpretation
of the results unambiguous. The predetermined drifts of 4%
for the steel frame and 2% for the reinforced concrete wall
represent a relatively large demand for these structures, but
one that is considered within the range of applicability of the
modeling assumptions. This is desirable for the purposes of
this study, as it amplifies differences in the scaling techniques
that might not be as readily apparent at smaller (or elastic) drift
levels.
Figs. 9 and 10 present the scaled elastic response spectra for
the three scaling methods, along with the peak displacement
response of the corresponding ESDOF systems for Scaling
Method II, for the 9-story frame (at 4% drift) and the
8-story wall (at 2% drift), respectively. Relatively large
spectral amplitudes are required to obtain the predetermined
drifts. Differences in the elastic response spectra suggest that
the scaled ground motion suites do not represent identical
intensities and thus may reflect different hazard levels.
However, the differences in the central tendencies of response
quantities reported subsequently are small; these differences
may reflect the normalization technique rather than being an
intrinsic property of the intensity measure.
Figs. 11 and 12 present the statistical distributions on
response quantities determined for the 9-story steel frame and
the 8-story wall, respectively, according to the three scaling
methods. Dispersions are plotted in Figs. 13 and 14.
The three scaling methods generally produced similar
distributions of force quantities (story shears and overturning
moments) for both buildings. The elastic displacement method
often produced slightly larger mean shears and moments in the
lower stories of the wall building, as well as larger dispersions
in the moments in the lower stories of the wall building.
Larger differences were observed in the distributions of
deformation quantities (floor displacements and interstory
drifts). The elastic displacement method produced the largest
dispersions in floor displacements and interstory drifts for
both buildings. The distributions of floor displacements and
interstory drifts obtained by elastic displacement scaling
tended to be asymmetric, with mean values significantly larger
than medians. This is consistent with the well-documented
assumption of log-normality (e.g. of the conditional distribution
of θmax given Sa (T1 ; 5%)) referred to by Vamvatsikos and
Cornell [26]. (However, the applicability of a log-normal
distribution was not uniformly supported for the distributions
of force quantities determined by elastic displacement scaling.)
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
Fig. 10. Response spectra for ground motions used for analysis of the 8-story reinforced concrete wall, scaled according to Scaling Methods I, II, and III.
The distributions of deformation quantities obtained with
the target roof displacement scaling method tended to be
symmetric, with means approximately equal to the medians
of the distributions obtained using the inelastic displacement
method. The dispersions of the deformation quantities obtained
with the target roof displacement method were the smallest,
ranging from substantially less than to approximately equal to
those obtained with the other scaling methods.
The distributions of deformation quantities obtained with the
inelastic displacement method often were asymmetric (perhaps
log-normal) for the frame building. The distributions for the
wall building, obtained with both the inelastic displacement
and target roof drift methods, had such small dispersions that
distinctions between log-normality and normality are of little
significance.
Dispersions in floor displacement obtained with the target
roof drift method diminish to zero as one approaches the
roof of each building as a result of the scaling procedure.
Response statistics obtained with this scaling method are
conditioned on the roof drift being equal to the target roof
drift. The distributions obtained with this method appear to
be reasonably symmetric (suggestive of a normal distribution)
and the dispersions were less than or approximately equal to
those obtained with the other scaling methods. The apparent
symmetry of the distributions obtained with the target roof drift
method is the basis for the assumption of normality in the
derivation of Eq. (3).
1435
The target roof drift and inelastic displacement scaling
methods differ in that the dispersion in roof displacement
associated with MDOF effects is represented in the response
data obtained with the latter method, whereas the scale factors
applied to the ground motions used in the former method are
adjusted to reflect the effect of the interaction of multiple
degrees of freedom on the peak roof displacement. The close
correspondence of the mean deformation obtained with the
target roof drift method and the median deformations obtained
with the inelastic displacement method indicates that MDOF
(or higher mode) effects primarily affect the dispersions in
the deformation quantities while having little effect on central
tendencies. The substantially greater dispersion observed for
deformation quantities obtained by elastic displacement scaling
and the tendency for the distributions to be asymmetric (perhaps
log-normal) are largely attributable to variability associated
with inelastic response in the first mode. This variability
apparently does not carry through to the force quantities.
Clearly, a reduction in dispersion improves the efficiency of
the intensity measure, as indicated by Eq. (3) and as observed
by Giovenale et al. [19] and Mackie and Stojadinovic [21].
Both inelastic displacement and target roof drift scaling result
in substantially smaller dispersions than elastic displacement
scaling. Use of inelastic displacement scaling would produce
design values that are conditioned on the estimate of inelastic
displacement and thus represent variability in MDOF response
relative to SDOF response. Use of target roof drift scaling
1436 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441

Fig. 11. Scaling Methods I, II, and III applied to the 9-story frame.

produces design values that are conditioned on the target roof but apparently produces smaller estimates of deformation
drift level; uncertainty in the estimated target roof drift level quantities when the estimates are made at confidence levels
has been neglected. This appears to have a negligible effect greater than 50% and/or at some number of standard deviations
on the force quantities that would be determined in design, above the mean (Figs. 11 and 12).
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441 1437

Fig. 12. Scaling Methods I, II, and III applied to the 8-story wall.

Consequently, the Scaled NDP can be recommended for 6. Conclusions

estimating force and deformation quantities, conditioned on
the target roof drift. Based on the preceding observations, the
estimates of force quantities appear to be credible within one
or several standard deviations of the mean and at varied levels
of confidence, while the estimates of deformation quantities,
while useful, do not fully reflect uncertainty in the roof
displacement.
Results reported in FEMA 440 [18] and elsewhere illustrate
that substantial errors can occur when estimating response
quantities such as interstory drift, story shear, and overturning
moments using various load vectors that have been proposed for
the NSP. These errors are attributed to the vibration of multiple
degrees of freedom, or the response of higher modes in the case
of elastic systems.
1438 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
Fig. 13. COVs for the 9-story frame.
The scaling method proposed herein, in which the ground
motion excitations are scaled such that the peak roof drift
obtained in a nonlinear dynamic analysis matches the estimated
or target roof drift, may be considered to be an application of a
dynamic load vector (rather than a static load vector) to reach
a demand displacement. Each ground motion thus represents
a different dynamic load vector. The variability in response
quantities, associated with aleatory effects on the vibration of
multiple degrees of freedom, the influence of capacity limits on
demands, and the effect of sample size on statistical estimates
of population characteristics are explicitly represented in the
Scaled NDP.
Response quantities were obtained for models of a 9-story
steel frame and an 8-story reinforced concrete wall using the
Scaled NDP (termed target roof drift scaling) and two other
intensity measures (termed elastic displacement and inelastic
displacement scaling). A comparison indicated that:
(1) The three intensity measures generally produced similar
central tendencies and dispersions of force quantities (story
shears and overturning moments).
(2) The elastic displacement scaling method produced substan-
tially larger dispersions in floor displacements and inter-
story drifts, with the distributions in these quantities possi-
bly following a log-normal distribution. The dispersions in
these deformation quantities are largely attributed to vari-
ability in the inelastic response of the first mode ESDOF
system.
(3) The distributions of deformation quantities obtained with
the target roof drift scaling method lacked the asymmetry
observed with elastic displacement scaling. A normal
distribution may be appropriate for deformation quantities
obtained using this intensity measure.
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
Fig. 14. COVs for the 8-story wall.
(4) Means of the distributions of deformation quantities
obtained with target roof drift scaling were approximately
equal to the medians of the distributions obtained
using inelastic displacement scaling. Dispersions in the
deformation quantities obtained with roof drift scaling were
less than or approximately equal to those obtained with
inelastic displacement scaling.
(5) Distributions obtained with target roof drift scaling are
conditioned on the roof drift estimate, and thus, do not
reflect uncertainty in the actual roof drift. This appears to be
inconsequential for the distributions of force quantities. The
dispersions in deformation quantities were, in some cases,
severely reduced as a result of this scaling method.
Estimates of response quantities made with the Scaled NDP
are conditioned on the target roof drift (that is, that the actual
roof drift matches the estimated value). The similarity in the
distributions of force quantities obtained for the three scaling
methods suggests that this is a good assumption for estimating
forces and moments at various levels above or below the mean
1439
and at various levels of confidence. Dispersions in deformation
demands obtained with the Scaled NDP were often smaller than
those obtained with other scaling methods. Thus, estimates of
deformation quantities made with the Scaled NDP, while useful,
are more sensitive to uncertainties in roof drift.
Because elastic spectra may be used as the basis for the peak
displacement estimates of the NSP (e.g. using the displacement
modification or modified acceleration–displacement response
spectrum methods of FEMA 440), the Scaled NDP makes
use of the substantial effort that already has gone into the
development of site-specific spectral descriptions of hazard in
many countries. The simplicity of the method and the reliance
on spectral descriptions of hazard makes the Scaled NDP
amenable to specification in codes for the design of buildings.
For design in the context of performance-based earthquake
engineering, one may use methods based on NSPs for
preliminary determination of the strength and stiffness required
for the structure to satisfy global seismic performance
objectives. These preliminary estimates may be used to develop
the detailed design of the structure. The Scaled NDP may then
1440 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441

be used to characterize the expected performance of the design If the standard deviation σ is not known a priori, but rather,
(e.g. mean interstory drifts) and to establish the values of forces must be estimated by the sample standard deviation s, then the
that must be resisted elastically (e.g. wall shears or collector sample mean has the Student’s t-distribution with n − 1 degrees
forces) at a desired level of confidence, to ensure that the of freedom:
structure develops the intended inelastic mechanism. Most of  
s
the required effort is in the development of the analytical model x̄n ∼ tn−1 µ, √ . (A.2)
of the structure. Response computations and iterative scaling n
of ground motions are fairly simple and may be automated. This can be expressed as
As with other analysis procedures, the accuracy of the results
x̄n − µ
depends on the fidelity of the structural model. Unlike linear √ ∼ tn−1 (0, 1). (A.3)
procedures and nonlinear static procedures, the Scaled NDP s/ n
requires that the cyclic behavior of the components be defined We seek to establish c0 such that
to an acceptable degree of accuracy.
Refinements and improvements to the Scaled NDP P(c0 x̄n > µ + κσ ) = α (A.4)
potentially may be made in (1) characterization and selection
which can be restated as
of site specific ground motions, (2) determination of the
confidence levels (α) and numbers of standard deviations µ + κσ
 
P x̄n < = 1 − α. (A.5)
above the mean (κ) that should be used for various response c0
quantities, (3) establishment of minimum numbers of analyses
Given Eq. (A.2), this probability can be re-expressed as
required for estimation of the mean and COV, (4) precision
of the NSP estimates of peak roof displacement, (5) methods µ+κσ
!
c0 − µ
used to address P–Delta effects and cracking in reinforced Φt,n−1 √ =1−α (A.6)
concrete structures, and (6) treatment of additional sources s/ n
of variability in response quantities, such as uncertainties where Φt,n−1 is the cumulative distribution function for the
in material properties, member dimensions, and structural Student’s t-distribution with n − 1 degrees of freedom. Thus,
modeling.
µ+κσ
!
c0 − µ
Acknowledgments Φt,n−1 (1 − α) =
−1
√ . (A.7)
s/ n

This paper describes work performed under the auspices of Algebraic manipulation allows c0 to be expressed as
the ATC-55 project of the Applied Technology Council, funded 1 + κσ/µ
by the Federal Emergency Management Agency. Guidance c0 = (A.8)
provided by the Project Management Committee is appreciated.
−1
1 + Φt,n−1 (1 − α) s/µ
√
n
In addition, the assistance of Professor Y.K. Wen, and the
or equivalently as
thoughtful consideration given by Professor Helmut Krawinkler
are gratefully acknowledged. The views expressed do not 1 + κCOV
necessarily represent those of the above individuals and c0 = (A.9)
−1
1 − Φt,n−1 (α) COV
√
n
organizations.
where σ/µ is estimated by the sample coefficient of variation,
Appendix A. Statistical derivation COV. For convenience, we may express c0 = c(1 + κCOV)
where
The sampling problem is formulated as follows: a response 1
quantity X has peak values x1 , x2 , . . . , xn in n dynamic c= (A.10)
−1
1 − Φt,n−1 (α) COV
√
n
analyses of a structure. The mean of the n responses is x̄n . The
responses are assumed to be normally distributed, in part based −1
and Φt,n−1 (α) is the value of the variate of the Student’s t-
on results shown in Figs. 11 and 12, with a standard deviation distribution with n − 1 degrees of freedom at a confidence level
σx . We are interested in determining the scale factor c0 such of α.
that c0 x̄n exceeds the population mean plus κ standard deviation
level (µ + κσx ) with a specified level of confidence α. The Appendix B. Notation
derivation is as follows:
X is normally distributed with mean µ and standard The following symbols are used in this paper:
deviation σ . That is, X ∼ N (µ, σ ).
The sample mean of X is given by x̄n where c scaling parameter
c0 scaling parameter
σ
 
x̄n ∼ N µ, √ . (A.1) COV coefficient of variation of sample
n Fi vector of floor forces associated with the ith mode
 M. Aschheim et al. / Engineering Structures 29 (2007) 1422–1441
k coefficient in load vector function per ICBO (2000)
M mass matrix
n number of analyses or observations
s sample standard deviation
Sa (T1 ; 5%) spectral acceleration associated with T1 for
viscous damping equal to 5% of critical damping
Sd spectral displacement
Ti period associated with the ith mode
xi vector of floor displacements associated with the ith
mode
xi ith value of response quantity
x̄n mean of n observations of a response quantity
Φt,n−1 cumulative distribution function for the Student’s
t-distribution for n − 1 degrees of freedom
α confidence level
φi mode shape of the ith mode
κ standard deviation level
Γi modal participation factor for the ith mode
µ population mean
θmax peak interstory drift
σ population standard deviation
Subscripts
i index value
References
[1] Aschheim M, Black E. Yield point spectra for seismic design and
rehabilitation. Earthquake Spectra 2000;16(2):317–35.
[2] Aschheim MA, Black EF, Cuesta I. Theory of principal components
analysis and applications to multistory frame buildings responding to
seismic excitation. Engineering Structures 2002;24(8):1091–103.
[3] ATC-40. Seismic evaluation and retrofit of concrete buildings. Report no.
SSC 96-01. Redwood City (CA): Applied Technology Council; 1996.
[4] Aydinoglu N. An incremental response spectrum analysis procedure based
on inelastic spectral displacements for multi-mode seismic performance
evaluation. Bulletin of Earthquake Engineering 2003;1:3–36.
[5] Building Seismic Safety Council (BSSC). NEHRP recommended
provisions for seismic regulations for new buildings and other structures.
FEMA 368. Washington (DC): Federal Emergency Management Agency;
2000.
[6] Chopra AK, Goel RK. Capacity-demand-diagram methods based on
inelastic design spectrum. Earthquake Spectra 1999;15(4):637–56.
[7] Chopra AK, Goel RK. A modal pushover analysis procedure for
estimating seismic demands for buildings. Earthquake Engineering and
Structural Dynamics 2002;31:561–82.
[8] Chopra AK, Goel RK, Chintanapakdee C. Statistics of single-degree-
of-freedom estimate of displacement for pushover analysis of buildings.
Journal of Structural Engineering, ASCE 2003;129(4):459–69.
[9] Chopra AK, Goel RK, Chintanapakdee C. Evaluation of a modified MPA
procedure assuming higher modes as elastic to estimate seismic demands.
Earthquake Spectra 2004;20(3):757–78.
1441
[10] Collins KR, Wen YK, Foutch DA. An alternative seismic design
procedure for standard buildings. In: Eleventh world conference on
earthquake engineering. Pergamon: Elsevier Science; 1996.
[11] Cornell CA, Luco N. Ground motion intensity measures for structural
performance assessment at near-fault sites. In: Proceedings US–Japan
joint workshop and third grantees meeting. US–Japan Cooperative
Research on Urban Earthquake Disaster Mitigation. Seattle (Washington):
University of Washington. 2001. p. 30–8.
[12] Cordova PP, Deierlein GG, Mehanny SS, Cornell CA. Development
of a two-parameter seismic intensity measure and probabilistic
assessment procedure. In: Proceedings of the 2nd US–Japan workshop
on performance-based earthquake engineering for reinforced concrete
building structures. 2000.
[13] Cuesta I, Aschheim M. Using pulse R-Factors to estimate the structural
response to earthquake ground motions. CD Release 01-03. Mid-America
Earthquake Center; March 2001.
[14] Department of Defense. Seismic design guidelines for essential facilities.
Technical manual ARMY TM 5-809-10-1, Navy NAVFAC P-355.1, Air
Force AFM 88-3 Chapter 13 Section A. Departments of the Army, the
Navy, and the Air Force; 27 February 1986.
[15] Fajfar P. Capacity-spectrum method based on inelastic demand spectra.
Earthquake Engineering and Structural Dynamics 1999;28:979–93.
[16] FEMA 273. NEHRP guidelines for the seismic rehabilitation of
buildings. Report no. FEMA 273. Washington (DC): Federal Emergency
Management Agency; 1997.
[17] FEMA 351. Recommended seismic evaluation and upgrade criteria for
existing welded steel moment-frame buildings. Report no. FEMA 351.
Washington (DC): Federal Emergency Management Agency; 2000.
[18] FEMA 440. Improvement of inelastic seismic analysis procedures. Report
no. FEMA 440. Washington (DC): Federal Emergency Management
Agency; 2005.
[19] Giovenale P, Cornell CA, Esteva L. Comparing the adequacy of
alternative ground motion intensity measures for the estimation of
structural responses. Earthquake Engineering and Structural Dynamics
2004;33:951–79.
[20] International Conference of Building Officials (ICBO). Whittier (CA):
International Building Code; 2000.
[21] Mackie K, Stojadinovic B. Improving probabilistic seismic demand
models through refined intensity measures. In: 13th world conference on
earthquake engineering. 2004.
[22] Miranda E. Seismic evaluation and upgrading of existing buildings. Ph.D.
thesis. Berkeley (CA): University of California; 1991.
[23] Powell GH. DRAIN-2DX element description and user guide for element
Type01, Type02, Type04, Type06, Type09, and Type15 Version 1.10.
Report no. UCB/SEMM-93/18. Berkeley (CA): University of California;
1993.
[24] Priestley N, Amaris A. Dynamic amplification of seismic moments and
shear forces in cantilever walls. In: Concrete structures in seismic regions:
fib 2003 symposium. 2003.
[25] SAC Joint Venture. State of the art report on systems performance of
steel moment frames subject to earthquake ground shaking. FEMA 355C.
Washington (DC): Federal Emergency Management Agency; 2000.
[26] Vamvatsikos D, Cornell CA. Incremental dynamic analysis. Earthquake
Engineering and Structural Dynamics 2002;31(3):491–514.
[27] Vamvatsikos D, Cornell CA. Applied incremental dynamic analysis.
Earthquake Spectra 2004;20(2):285–553.