Inelastic spectra for displacement-based seismic design B. Borzi a,1 , G.M. Calvi b , A.S. Elnashai a, * , E. Faccioli c , J.J. Bommer a a Department of Civil and Environmental Engineering, Imperial College, London SW7 2BU, UK b Dipartimento di Meccanica Strutturale, Universita ´ degli Studi di Pavia, Via Ferrata 1, 27100 Pavia, Italy c Dipartimento di Ingegneria Strutturale, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milan, Italy Accepted 6 September 2000 Abstract In recognition of the emergence of displacement-based seismic design as a potentially more rational approach than force-based techniques, this paper addresses derivation of inelastic displacement spectra and associated topics. A well-constrained earthquake strong-motion dataset is used to derive inelastic displacement spectra, displacement reduction factors and ductility–damping relationships. These are in a format amenable for use in design and assessment of structures with a wide range of response characteristics. q 2001 Elsevier Science Ltd. All rights reserved. Keywords: Attenuation relationship; Displacement spectrum; Displacement-based approach; Ductility; Damping; Displacement modification factor 1. Introduction Force-based seismic design remains, in spite of its short- comings, the method widely used in codes. However, defor- mation-based (Calvi and Kingsley [1], Kowalsky et al. [2], Priestley et al. [3]) or deformation-controlled (Panagiotakos and Fardis [4]) procedures have recently been proposed. The new approach utilises displacement spectra expressed as spectral ordinates versus effective period (period at maxi- mum displacement) to quantify the demand imposed on structural systems. The origins of displacement-based design may be traced to work published as early as the 1960s, where comments on the displacements of inelastic systems and their relationship to their elastic counterparts were made (e.g. Muto et al., 1960, as reported by Moehle [5]). However, it was the work of Sozen and his associates (Gulkan and Sozen [6], Shibata and Sozen [7]) that developed the concept of a substitute structure. The substitute structure is a single degree of freedom elastic system, the characteristics of which represent the inelastic system. This concept enables the use of an elastic displacement spectrum in design, while using the displacement capacity of an inelastic system. Various contributions were made towards the develop- ment of displacement-based seismic design since the early work mentioned above. However, it was in the 1990s that formal proposals were made to implement the emerging ideas into a design procedure. The earliest is that by Moehle and his co-workers (Moehle [5], Qi and Moehle [8]). There- after, a complete and workable procedure for seismic design of structures that sets aside forces and relies on displace- ment as a primary design quantity was proposed by Kowalsky et al. [2] for single degree of freedom systems (such as bridge piers). A parallel paper on multi-degree of freedom systems is due to Calvi and Kingsley [1]. The steps comprising the design process for SDOF systems are given below for simplicity: a. A target displacement for the structure is selected, based on the type of structure and the governing limit state. b. Knowing the yield displacement, and the material and structural system, a value of equivalent damping is deter- mined. c. Displacement spectra representative of the seismo- tectonic environment are used. The inputs are the target displacement and the equivalent damping. The output is an effective period of vibration. d. The structure is dimensioned to give an effective period, taking into account reduced stiffness consistent with the level of deformation, equal to that obtained from the displa- cement spectra. e.If the effective period is not sufficiently close to the required period, go to step b. above and repeat until convergence. Soil Dynamics and Earthquake Engineering 21 (2001) 47–61 0267-7261/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved. PII: S0267-7261(00)00075-0 www.elsevier.com/locate/soildyn * Corresponding author. Tel.: 144-20-7594-6058; fax: 144-20-7225- 2716. E-mail address: [email protected] (A.S. Elnashai). 1 Present address: EQE International Ltd, Warrington WA3 6WJ, UK.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Inelastic spectra for displacement-based seismic design
B. Borzia,1, G.M. Calvib, A.S. Elnashaia,*, E. Facciolic, J.J. Bommera
aDepartment of Civil and Environmental Engineering, Imperial College, London SW7 2BU, UKbDipartimento di Meccanica Strutturale, Universita degli Studi di Pavia, Via Ferrata 1, 27100 Pavia, Italy
cDipartimento di Ingegneria Strutturale, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milan, Italy
Accepted 6 September 2000
Abstract
In recognition of the emergence of displacement-based seismic design as a potentially more rational approach than force-based techniques,
this paper addresses derivation of inelastic displacement spectra and associated topics. A well-constrained earthquake strong-motion dataset
is used to derive inelastic displacement spectra, displacement reduction factors and ductility±damping relationships. These are in a format
amenable for use in design and assessment of structures with a wide range of response characteristics. q 2001 Elsevier Science Ltd. All
It is clear from the above that whereas displacement-
based design is the logical framework for seismic
design, since the primary source of seismic energy
dissipation is inelastic deformation, it imposes new
requirements for veri®able design. Primarily, accurate,
representative and parametrically described displace-
ment spectra are essential ingredients that have only
very recently become available (Bommer et al. [9],
Bommer and Elnashai [10], Tolis and Faccioli [11]).
In the study by Bommer et al. [9], it was shown that
a well-balanced catalogue for acceleration studies may
be unsuitable for displacement purposes. The ordinates
of acceleration response spectra are insensitive to the
processing applied to the accelerograms, whereas the
in¯uence of the ®lter parameters on the displacement
spectra is pronounced.
In this paper attenuation coef®cients for inelastic displa-
cement constant ductility spectra are calculated. Attenuation
coef®cients for residual displacements were also taken into
account. This is of great importance in the assessment of
displacement-based design situations, hence the spectra are
presented as ratio of maximum response displacement. By
means of a comparison between elastic and inelastic displa-
cement spectral ordinates a reduction coef®cient taking into
account the effects of real inelastic behaviour in a simpli®ed
method of design based on displacement is calculated. For
this coef®cient, herein denoted h -factor, regression coef®-
cients were evaluated (Borzi et al. [12], Borzi [13]) for all
magnitudes, distances and site conditions. Further, average
values obtained constitute a valid reference for new code
implementations for simpli®ed methods based on displace-
ments.
2. Input motion
2.1. Strong-motion dataset selection and processing
The dataset employed for the de®nition of inelastic
constant ductility spectra was presented by Bommer et al.
[9] and used for the derivation of frequency-dependent
attenuation equations for ordinates of displacement
response spectra. All records were ®ltered individually at
Imperial College (London, UK) and used to derive displa-
cement spectra for different levels of damping, from 5 to
30%. Each record was individually processed using an ellip-
tical ®lter and a long-period cut-off that resulted in a physi-
cally reasonable displacement time-history (Bommer and
Elnashai [10]). This ®ltering process removes permanent
displacements, caused by ground deformation or fault slip,
in order not to mix these phenomena with transitory ground
vibrations. The dataset has been adapted from that
employed by Ambraseys et al. [14,15] to derive attenuation
relationships for ordinates of acceleration response spectra.
This is a high-quality dataset in terms of both accelerograms
that have been individually corrected, and information
regarding the recording stations and earthquake character-
istics. The original dataset of Ambraseys et al. [14,15] has
been modi®ed in the following ways:
1. records for small magnitude earthquakes have been
removed;
2. some site classi®cations have been corrected in the light
of new information;
3. two new strong-motion records have been added;
4. all of the original records have been ®ltered with indivi-
dually selected cut-offs.
The necessity of eliminating all the records generated by
earthquakes of magnitude less than Ms 5.5 is due to the
particular interest in long-period radiation. In Fig. 1 the
distribution of records comprising the dataset with regard
to magnitude, distance and site classi®cation are
B. Borzi et al. / Soil Dynamics and Earthquake Engineering 21 (2001) 47±6148
Fig. 1. Distribution of records for: (a) rock; (b) stiff; and (c) soft soil.
demonstrated to emphasise credentials of this set of natural
records.
While the source distance and the surface-wave magni-
tude are available for all the accelerograms, for three
records the local site geology is unknown. For the remaining
180, the percentages of distribution in the three site group-
ings of rock, stiff and soft soil are 25.0, 51.1 and 23.9%,
respectively. These percentages are close to the distribution
of the original dataset, being 25.5, 54.3 and 20.2%, respec-
tively. For two records only one component of the motion is
known. Therefore a total of 364 accelerograms were
processed.
It is acknowledged that site intensity could be used for the
selection of strong-motion datasets. However, this has its
own problems, and is not considered to be superior to the
selection based on magnitude, distance and site condition.
2.2. Attenuation model and regression analysis
Studies concerning the evaluation of seismic hazard
related to earthquakes utilise predictive models commonly
referred to as attenuation relationships. These models gener-
ally express values of strong-motion parameters as a func-
tion of source characteristics, propagation path and local site
geology.
In order to de®ne elastic spectra, a common approach to
perform a hazard analysis is to de®ne the hazard in terms of
the peak ground acceleration (PGA), which anchors the
zero-period ordinate for a standard spectral shape. A criti-
cism of this approach is that the resulting spectra do not
correspond to the same seismic hazard for all periods (i.e.
non-uniform hazard spectra). To obtain spectra charac-
terised according to the same seismic hazard for every spec-
tral ordinate (uniform hazard spectra) the de®nition of
period-dependent attenuation relationships are proposed.
The attenuation model used in this work is that of Ambra-
seys et al. [14,15] employed to de®ne elastic acceleration
spectra. The formulation of the attenuation relationship is:
log�y� � C1 1 C2Ms 1 C4 log�r�1 CaSa 1 CsSs 1 sP �1�where y is the strong-motion parameter to be predicted, Ms is
the surface wave magnitude and:
r �����������d 2 1 h2
0
q�2�
in which d is the shortest distance from the station to the
surface projection of the fault rupture in km and h0 is a
regression constant. The h0 coef®cient takes into account
that the fault projection is not necessarily the source of
the peak motion, but it does not represent explicitly the
effect of the focal depth on the motion. The coef®cients
C1, C2, C4, Ca, Cs and h0 are determined by regression. In
this attenuation model three soil conditions are distin-
guished by the average shear wave velocity. When the
shear wave velocity is higher than 750 m/s the soil is clas-
si®ed as rock. Soft soil has a shear wave velocity is less than
360 m/s, whilst stiff soil is assumed in the intermediate
range of shear wave velocity. In the regression model Sa
takes the value of 1 for soft soil conditions and 0 otherwise,
while Ss takes the value of 1 for stiff soil conditions and 0
otherwise. Finally, P is a parameter that is multiplied by the
standard deviation s and takes the value of 0 when the mean
value of log(y) is calculated and 1 for the 84-percentile
value of log(y).
3. Structural models
3.1. Elastic perfectly-plastic model
In order to determine the in¯uence of magnitude, distance
and soil condition on inelastic response spectra, attenuation
relationships have been de®ned using an elastic perfectly-
plastic (EPP) response model. The EPP model was
employed since it is the simplest form of inelastic force-
resistance as well as being the basis for early relationships
between seismic motion and response modi®cation factors.
Moreover, by virtue of its two parameters de®nition: level of
force-resistance and stiffness, few structural characteristics
are included, hence the in¯uence of strong-motion records
may be better visualised. The stiffness corresponds to the
period of vibration for which the spectral ordinate has to be
calculated and the resistance is derived iteratively. In this
work inelastic constant ductility spectra were obtained.
Therefore the resistance of the system corresponds to the
resistance for which the system has a required ductility
equal to the target ductility. The ensuing inelastic spectra
would re¯ect solely the characteristics of the input motion.
3.2. Hysteretic hardening±softening model
In order to investigate the in¯uence of the response char-
acteristics of structures on inelastic displacement spectra, a
hysteretic hardening-softening model (HHS) was used
(Ozcebe and Saatcioglu [16]). The structural model is char-
acterised by the de®nition of a primary curve, unloading and
reloading rules. The primary curve for a hysteretic force±
displacement relationship is de®ned as the envelope curve
under cyclic load reversals. For non-degrading models the
primary curve is taken as the response curve under mono-
tonic loading. In this model the primary curve is used to
de®ne the limits for member strength. Two points on the
primary curve have to be de®ned. It is essential to de®ne
cracking and yield loads (Vcr and Vy) and the corresponding
displacements (D cr and D y), as shown in Fig. 2. If, for exam-
ple, this model was used to describe the hysteretic behaviour
of reinforced concrete members, the cracking load would
correspond to the spreading of cracks in the concrete and the
yielding load would be related with the load at which the
strain in bars is equal to the yield strain of steel or some
other criterion can be selected by the user. Unloading and
reloading branches of the HHS model have been established
through a statistical analysis of experimental data
B. Borzi et al. / Soil Dynamics and Earthquake Engineering 21 (2001) 47±61 49
(Saatcioglu and co-workers [17,18]). The load reversal rules
are brie¯y described below.
Structural members exhibit stiffness degradation under
cyclic loading. When the number of cycles or the amplitude
of inelastic deformation increases, the system becomes
softer. Furthermore, the hysteretic behaviour is affected by
pinching. The axial load is an important parameter in
predicting pinching effects (due to the onset of crack
closure). The slope of reloading branches increases beyond
the crack load.
The slopes of the lines connecting the origin to the crack-
ing point (K1 in Fig. 2) and the yield point to the cracking
point in the opposite quadrant (K2 in Fig. 2) are used to
de®ne the unloading branches under cyclic loads. The latter
slope depends on deformation and force levels attained at
the beginning of unloading. Experimental results indicate
that if unloading starts between the cracking and the yield
load, and the yield load has not been exceeded in the rele-
vant quadrant, then unloading stiffness is bounded by K1 and
K2. In this model a linear variation between these limits was
proposed as a function of displacement ductility. If the
unloading load exceeds the yield load, the unloading
curve changes the slope to a value close to the cracking load.
4. Procedural considerations
In order to de®ne inelastic displacement spectra and
displacement modi®cation (h?, displacement ductilities of
2, 3, 4 and 6 are considered. Ductility levels higher than 6
are not included because they constitute global displace-
ment ductility and structures very rarely have a local ducti-
lity supply commensurate with global ductility above 6. The
inelastic spectra have been de®ned between 0.05 and 3 s.
For the HHS model the initial elastic period can be consid-
ered either as the stiffness before Vcr or the secant stiffness.
In the current work period corresponding to the secant stiff-
ness was considered. This is because the stiffness before Vcr
is not representative of the structural behaviour. A low
damping value of 1% was included. This viscous damping
is representative of the non-hysteretic dissipation, since
hysteretic damping is already included. Sensitivity analyses
were undertaken and have indicated that further precision of
this assumption is unwarranted.
The input parameters for the HHS model described above
are the monotonic curve and the relationship between axial
compressive force and nominal concentric axial capacity. In
order to de®ne the inelastic constant ductility spectra the
magnitude of the monotonic curve is not an input parameter.
It is de®ned in an iterative way forcing the relationship
between maximum and yield displacements to satisfy the
target ductility. To obtain the inelastic spectra and displace-
ment modi®cation factor (h ) an approximation of the
primary curve with three linear branches has been assumed
(Fig. 3). Consequently, the input parameters de®ning the
shape of the primary curve are:
1. the relationship between the cracking and the yielding
load (Vcr/Vy);
2. the relationship between the stiffness before the cracking
load and the secant stiffness (Kcr/Ky);
3. the slope of the post yield branch.
B. Borzi et al. / Soil Dynamics and Earthquake Engineering 21 (2001) 47±6150
Fig. 3. Shape of primary curve used in this work.
Fig. 2. HHS model for structural members.
To select the values of parameters to be employed, exten-
sive analysis of the in¯uence of each parameter on the
inelastic spectra was undertaken. The results of a parametric
investigation indicate that the parameter with the strongest
in¯uence on inelastic spectra is the slope of the post yield
branch. Hence ®xed ratios between Vcr and Vy and between
Kcr and Ky were considered. Although the aforementioned
ratios have a large variability, constant values are assumed,
since the results of preliminary parametric investigations
using a ®ne mesh of variation show that they do not have
a signi®cant in¯uence in terms of the inelastic displacement
spectra. From the experimental results of Paulay and Priest-
ley [19], Priestley et al. [20], Calvi and Pinto [21] and Pinto
[22], it is reasonable to consider the secant stiffness at the
yield point equal to 50% of the stiffness before Vcr; The
latter is taken equal to 30% of Vy. The ratio between the
cracking and yield loads in¯uences the pinching behaviour
that does not occur often for structures with loads higher
than approximately 30% of the yielding load Vy. The consid-
ered representative slopes of the structural behaviour are:
² K3 � 0 : (elastic perfect plastic behaviour)
² K3 � 10%Ky : (hardening behaviour)
² K3 � 220%Ky : (softening behaviour)
² K3 � 230%Ky : (softening behaviour)
An axial load equal to 10% of the nominal axial load is
assumed, since the model does not account for second-order
effects. The above characteristics have been veri®ed to
cover both new structures with seismic detailing, and exist-
ing poorly detailed structures (Borzi et al. [12], Borzi [13]).
An iterative procedure was utilised for the de®nition of
spectral ordinates corresponding to target ductility. In rare
cases it has not been possible to obtain a convergent solution
with the HHS model. The percentage of spectral ordinates
that have not been considered in the regression analyses are
reported in Table 1.
It is observed that the number of spectral ordinates to be
excluded from parametric analysis for the slope of the third
branch equal to 230%Ky and ductility equal to 4, is very
high. The attenuation relationship for this combination of
parameters was therefore not considered. The above obser-
vations (non-convergence) are fully justi®ed by noting that
highly degrading systems are inherently of low ductility.
This was further investigated by comparison with a different
program for inelastic spectra available to the authors
(program inspect). The lack of convergence was observed
for the same range of degrading stiffness and high ductility,
thus con®rming that these situations correspond to structural
collapse.
5. Inelastic displacement spectra
The cut-off period of every accelerogram is associated
with an initial elastic period TI, which corresponds to the
secant stiffness at yield. This is less than the cut-off periods
employed in order to ®lter the records. For inelastic systems
there is doubt as to whether spectral ordinates should corre-
spond to the equivalent elastic period TE rather than the
initial elastic period TI, both of which are less than the
cut-off period. However, it was observed that even if
equivalent elastic periods are somewhat higher than the
cut-off period of 3 s, the inelastic spectral ordinates are
still valid. This was con®rmed by comparison to selected
digital records Borzi [13].
In order to study the in¯uence of input motion para-
meters on inelastic displacement spectra, attenuation coef-
®cients were calculated for the EPP model. Fig. 4a±c
show the in¯uence of magnitude, distance and soil condi-
tions on inelastic displacement spectra. They con®rm the
strong in¯uence of input motion parameters on inelastic
displacement spectra, as already demonstrated in previous
studies for elastic displacement spectra by Bommer et al.
[9] (Fig. 5).
To compare the elastic and inelastic spectra, attenuation
coef®cients were de®ned for elastic spectra. Fig. 6 shows the
in¯uence of ductility on inelastic displacement spectra. In
this representation the initial elastic period of vibration
corresponding to the secant stiffness at the yield point is
assumed. Thus the elastic and inelastic systems with differ-
ent ductility requirements are characterised by the same
initial stiffness. The results con®rm the established observa-
tion that elastic and inelastic systems with the same initial
stiffness reach similar maximum displacements. However,
the difference between elastic and inelastic spectral ordi-
nates calculated in this work is larger than in previous
ones. This is due to the fact that the damping value used
herein is small (1% of critical). Therefore, the elastic spec-
tral ordinates for periods corresponding to the soil frequency
tend to be higher than the inelastic ones. As an example, in
the work of Miranda [23] a damping of 5% of critical was
employed. In comparing elastic and inelastic displacements,
small damping values should be employed, because the
damping must represent only the dissipation of energy not
related with inelastic behaviour. However the ratio between
inelastic and elastic displacements tend to be equal to 1 only
in the long period range, as already observed in previous
studies (Miranda [23], Gupta and Sashi [24], Whittaker et al.
[25], Rahnama and Krawinkler [26]). In the short period
range the inelastic demand exceeds the elastic one as
shown in Fig. 6b. In this work the ratio between inelastic
and elastic displacement demand of a system characterised
B. Borzi et al. / Soil Dynamics and Earthquake Engineering 21 (2001) 47±61 51
Table 1
Percentage of ordinates excluded from regression analysis
K3 m � 2 (%) m � 3 (%) m � 4 (%) m � 6 (%)
0 0.16 0.58 1.23 3.16
10%Ky 0.03 0.13 0.27 0.80
220%Ky 1.67 5.12 12.00 ±
230%Ky 2.45 8.16 30.98 ±
by the same initial elastic period of vibration TI is not further
investigated, because in the direct displacement-based
approach an equivalent elastic period of vibration TE is
assumed. The substitute elastic systems used in previous
work and in the current one are shown in Fig. 7a and b,
respectively. Attention is drawn to the comparison of displa-
cement reached by the system shown in Fig. 7b, as discussed
hereafter.
B. Borzi et al. / Soil Dynamics and Earthquake Engineering 21 (2001) 47±6152
Fig. 5. In¯uence of: (a) magnitude; and (b) distance on elastic displacement spectra obtained by Bommer et al. (1996).
Fig. 4. In¯uence of: (a) magnitude; (b) distance; and (c) site condition on inelastic displacement spectra evaluated for the EPP model.
Inelastic displacement spectra were also obtained for the
HHS model. In the case of softening behaviour high levels
of ductility were not considered. This must be done because
from a practical point of view high ductilities are not signif-
icant for softening systems. In the case of post elastic stiff-
ness equal to 220 and 230% the secant stiffness for the
yield point ductility up to 4 and 3, respectively, were
assumed.
Fig. 8 shows the in¯uence of ductility and hysteretic
behaviour on inelastic displacement spectra. For these
B. Borzi et al. / Soil Dynamics and Earthquake Engineering 21 (2001) 47±61 53
Fig. 7. Elastic and inelastic systems compared in: (a) previous work; and in (b) current work.
Fig. 6. In¯uence ductility on: (a) displacement spectra; and (b) ratio between inelastic and elastic displacement spectra (EPP model, Ms � 4; d � 10 km; rock
damping value and the ductility factor is assumed for the
HHS model too. Therefore, differences in hysteretic beha-
viour are represented by variations in a . The equivalent
damping values jE and the corresponding a coef®cients
are reported in Table 3.
Assessment of the mean values of h lead to the following
observations:
1. Lower h is obtained with the EPP model than with the
HHS when applied to a perfectly plastic case. This is due
to the higher initial stiffness of the HHS model that uses a
secant stiffness of 50% of the initial stiffness. Therefore,
for the same maximum displacement the ductility
demand for the EPP case is twice that of HHS case,
provided the same initial stiffness is used.
2. In the hardening case lower h was obtained than in the
elasto-plastic case, because the equivalent system corre-
sponding to hardening is stiffer than that corresponding
to the elasto-plastic case.
3. In the softening case the average h value is higher than in
both the EPP and the hardening cases. This is due to the
high values of the displacement modi®cation factor for
stiff systems with softening behaviour. In terms of dissi-
pation of energy, high h values correspond to high
energy dissipation. Stiff systems tend to have a large
number of load reversals. Therefore, for softening
systems a large number of cycles of loading and reload-
ing reach the limit curve, leading to a greater dissipation
of energy.
For long period systems the mean values of h-factor
obtained above are unconservative, since they are in¯u-
enced by the high values characteristic of the response
modi®cation factor of stiff systems. In order to improve
the estimation of h -factor for long period systems, mean
values were evaluated comparing the energy of inelastic
and elastic displacement spectra, both considered for an
equivalent period TE. The displacement modi®cation factor
is thus given by the following expression:
h �
ZT MaxE
0SDEL
ZTMaxE
0SDIN
�10�
Long period ordinates become of greater importance than
short period ordinates, since the area under both the elastic
and inelastic spectra in the short period range is only a small
percentage of the total area. The results of this investigation
for the range of ductilities taken into account are given in
Tables 4 and 5.
The mean values of h found from Eq. (10) are lower than
those reported in Table 2. Therefore, the latter values are
B. Borzi et al. / Soil Dynamics and Earthquake Engineering 21 (2001) 47±61 57
Fig. 13. (a) Inelastic displacement spectra for different ductilities; and (b) elastic displacement spectra for equivalent viscous dampings (HHS model for K3 �0; Ms � 6; d � 10 km; soft soil).
Table 3
Equivalent damping jE and a values
m � 2 m � 3 m � 4 m � 6
jE (%) a jE (%) a jE (%) a jE (%) a
EPP 8 0.16 13 0.19 16 0.21 19 0.23
K3 � 0 11 0.22 15 0.23 18 0.23 20 0.24
K3 � 10%Ky 10 0.19 13 0.20 15 0.20 17 0.20
K3 � 220%Ky 13 0.27 20 0.31 28 0.37 ± ±
K3 � 230%Ky 15 0.31 27 0.41 ± ± ± ±
conservative as they are not in¯uenced by the behaviour of