Engineering Structures 33 (2011) 154–162

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Probabilistic model for failure initiation of reinforced concrete interior

beam–column connections subjected to seismic loading
Nilanjan Mitra a,∗ , Sudeshna Mitra a , Laura N. Lowes b
a
Department of Civil Engineering, Indian Institute of Technology, Kharagpur, India
b
Department of Civil Engineering, University of Washington, Seattle, WA, USA

article info abstract

Article history: The results of previous experimental tests indicate that reinforced concrete interior beam column joints
Received 21 April 2009 may exhibit significant strength and stiffness loss under earthquake loading, and the results of post-
Received in revised form earthquake reconnaissance indicate that joint failure may result in structural collapse. Thus seismic
26 September 2010
evaluation and design of reinforced concrete frames requires accurate prediction of the potential for
Accepted 30 September 2010
Available online 25 October 2010
joint failure. This paper presents a binomial logit model, developed using data from 110 experimental
tests, which define the probability that a reinforced concrete interior beam–column building connection,
Keywords:
with a specific set of design parameters, will exhibit either a non-ductile joint shear failure prior to
Probabilistic model beam yielding or a ductile failure that initiates with beam yielding. The calibrated model identifies the
Reinforced concrete relative importance of various design parameters in determining the connection’s response mechanism.
Beam-column connections The model can be used by an engineer designing a new connection, constructed of normal or high-strength
Logistic regression materials, to estimate the likelihood of joint failure initiation. The model can also be used by an engineer
Failure initiation evaluating an existing structure to estimate the likelihood of joint failure, determine the factors that most
Joint shear failure significantly affect this likelihood, and, thereby, develop a suitable and efficient retrofit strategy.
Beam yielding © 2010 Elsevier Ltd. All rights reserved.

1. Introduction
In a reinforced concrete frame subjected to earthquake loading,
beam–column joints are critical for developing frame action and
ensuring that inertial loads are transferred through the frame
to the foundation. Post-earthquake reconnaissance efforts have
attributed the collapse of many reinforced concrete frames to
the failure of joints [1]. Similarly, analyses of building frames,
using models that simulate joint stiffness and strength loss,
show that nonlinear joint action reduces lateral load resistance
and that joint failure may result in structural collapse [2].
Given the importance of these components, numerous previous
experimental investigations have addressed the seismic behavior
of beam–column joints, the mechanisms that determine behavior,
and the design parameters that affect behavior.
The results of previous experimental investigations show
that joints may exhibit significant stiffness and strength loss
under lateral loading. The results of previous research suggest
also that, in addition to material properties and geometric
configuration, a number of different design parameters may affect
∗
Corresponding author. Tel.: +91 3222 283430.
E-mail addresses: nilanjan@civil.iitkgp.ernet.in (N. Mitra),
sudeshna@civil.iitkgp.ernet.in (S. Mitra), lowes@u.washington.edu (L.N. Lowes).
0141-0296/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2010.09.029
joint response. These design parameters include joint shear stress
demand [3–13], joint transverse reinforcement ratio [3,6,14–17],
bond stress demand for beam longitudinal reinforcement passing
through the joint [3,7,18–23], and column axial load [7,9,14,17,
24–29]. For joints with sufficient strength to develop the yield
strength of the beams framing into the joint, experimental data
indicate also that drift history affects strength deterioration of
the joint [30,31]. Experimental investigations at the University of
Washington [30,31] also indicate that drift has minimal impact on
connection strength.
The ACI Committee 352 [32] defines a beam–column joint as
‘‘that portion of the column within the depth of the deepest beam
that frames into the column’’, and a connection as ‘‘the joint plus
the columns, beams and slabs adjacent to the joint’’. The strength
of a beam–column connection may be determined by the flexural
yield strength of the beams or columns framing into the joint, or
by the joint region. The results of previous research provide a basis
for the current ACI Code [33] requirements that are intended to
ensure that connection response is determined by flexural yielding
of beams and that connection strength is determined by beam
flexural strength. These requirements include a minimum volume
of transverse reinforcement, a minimum anchorage length for
beam longitudinal reinforcement, a minimum column-to-beam
flexural strength ratio, and a limit on the joint shear stress demand.
Joints designed prior to 1967 [18,27,30,34,35,31] typically do not
comply with the current ACI Code [33] requirements and may
 N. Mitra et al. / Engineering Structures 33 (2011) 154–162
result in connections that exhibit joint failure or column yielding
prior to beam yielding.
Very limited guidance is provided by the ACI Code [33]
for design of structures with high-strength materials; however,
high-strength materials have been used in recent construction
project. Joints designed using high-strength materials may result
in connections for which seismic behavior is determined by joint
failure and not beam yielding [12,14,16,19,36–40]. In recent years,
high strength concrete (concrete with compressive strength in
excess of 59 MPa (8.6 ksi), as specified by ACI-363 committee [41])
has been used successfully to reduce member sizes in cast-in-
place concrete buildings and high-rise structures [39]. However,
research studies by Noguchi et al. [37] report that the use of
smaller, high-strength concrete column increases the potential for
joint shear failure prior to beam yielding. High strength reinforcing
steel has been used also in recent construction to reduce the
congestion of reinforcing bars in slender members [16,19,38].
However, research studies by Fujii and Morita [14] concluded
that high strength reinforcing steel in beams may increase beam
flexural strength and joint shear demand such that joint shear
failure precedes beam yielding.
Identification of the mechanism that determines connection
strength and seismic behavior (i.e., beam yielding, column yielding,
joint failure) provides insight into the performance of the
connection. Previous research indicates that connections that
exhibit joint failure prior to flexural yielding of beams and columns
exhibit minimal ductility and minimal drift capacity [5,7,9,12,
14,16,19,38]. Connections for which strength is determined by
column yielding may develop a soft-story mechanism leading
to structural collapse at relatively low drift demands [18,
35]. However, connections in which joints exhibit minimal
stiffness loss and for which strength is determined by beam
yielding typically exhibit a ductile response and significant drift
capacity [3,6,12,16,22,23,40,42,43].
In evaluating and retrofitting existing structures and designing
new structures, it is appropriate to assess the potential for
beam–column connections to exhibit a non-ductile response as
well as the impact of inelastic joint action on frame response.
To address this issue, a number of previous studies have focused
on the development of mechanistic models for simulation of
connection response. Simple mechanistic models [44–47] are
computationally efficient, but require the engineer to make a
number of assumptions about beam, column and joint behavior,
including the mechanism that determines connection strength
and drift capacity. A detailed discussion of examples of relatively
simple models can be found in Mitra [48]. Previous research
has resulted also in more sophisticated models [49–55] that
provide improved prediction of the mechanism that determines
the connection’s behavior. These models require far fewer
assumptions by engineers, but are typically both computationally
intensive as well as time consuming to calibrate for a particular
connection with specific design parameters. The reader is referred
to Mitra and Lowes [53] for an example of one of these
sophisticated models and Mitra [48] for a detailed discussion of the
models referenced above.
The above models enable the engineer to assess the potential for
non-ductile connection response as well as the impact of inelastic
joint action on frame response. However, use of the above models
typically requires a significant investment of an engineer’s time.
The objective of the research present here was to develop a
simple, easily applied, computationally efficient model that the
engineer can use to assess the potential for non-ductile connection
response and thereby the need for further investigation using more
sophisticated and time consuming analysis methods.
2. Research objectives, motivation and activities
The research presented here seeks to fill the need for a simple
and efficient tool for preliminary assessment of connection
155
performance. Specifically, the objectives of the research are
to
1. Use existing experimental data to develop a simple model that
will define the probability of failure initiation of reinforced
concrete interior beam–column connection, either by a non-
ductile joint failure mechanism or a ductile, beam-yielding
mechanism, subjected to seismic loading.
2. Quantify the impact of various design parameters on the
likelihood of the connection exhibiting joint failure.
It is expected that this model will provide
1. An engineer designing new connections with the ability to es-
timate the likelihood of joint failure controlling connection
response. With this model, an engineer can determine the pa-
rameters determining the connection’s response and, thereby,
modify a new design to reduce the likelihood of a joint failure
mechanism for a strong-column-weak-beam building frame
subjected to earthquake loading. While the building code pro-
vides guidance for design of connections using normal-strength
materials, it provides very limited guidance for design using
high strength materials.
2. An engineer evaluating or retrofitting as existing structure with
the ability to assess the likelihood of joint failure controlling
connection response as well as improved understanding of
the design parameters in determining the increased likelihood
of joint failure. This information could be used to develop a
suitable, efficient retrofit scheme.
3. A researcher with an estimate of the relative importance of
various design parameters to the likelihood of a connection
exhibiting a brittle joint failure as well as an example of the
application of logistic models for prediction of the behavior of
structural components and systems.
To accomplish the above objectives,
1. A large experimental data set was assembled that includes
data characterizing the response of building frame connections,
with a wide range of design parameters, subjected to simulated
earthquake loading.
2. These experimental data were used to calibrate a binomial
logit model which determines the probability of initiation of
connection failure initiation due to either joint failure or beam
yielding.
3. Statistical goodness-of-fit tests were employed to validate the
logit model.
4. The calibrated logit model was used in example applications
to assess the relative importance of various connection design
parameters on the likelihood of joint failure.
3. Experimental data set
An extensive experimental data set was used to support
development of the statistical model. The data set comprises
110 laboratory tests of two-dimensional interior beam–column
connections conducted by 20 research teams from around the
world during the last 40 years [3,5,7,9,12–14,16,19,21–23,30,34,
31,36,38,40,42,43,56]. Mitra [48] provide a detailed presentation of
the data set, including material, geometric and design parameters
for each test specimen. The specimens in the data set span a
wide range of joint design parameters. However, the data set is
limited to two-dimensional connections in which a continuous
column intersects a continuous beam and specimen response
is determined by flexural yielding of beams at the beam-joint
interface and/or joint failure. Too few tests were found in the
literature in which connection failure was determined by column
hinging to enable use of these data in the analysis. To improve the
accuracy of the model, connections with slabs, eccentric beams
156 N. Mitra et al. / Engineering Structures 33 (2011) 154–162
(axis of the beam and column are not aligned), or out-of-plane
beams were not included in the data set. Test specimens with plain
round (smooth) reinforcing steel bars were also eliminated from
the data set. Since too few tests were found in literature, the dataset
does not include specimens in which the beam-hinging effect has
been shifted from the beam–column joint interface to a distance
away from the interface and also experimental investigations in
which the column’s axial load was varied during the test. All
specimens were subjected to quasi-static, cyclic loading to develop
load distributions that are representative of those that develop in
a frame under earthquake loading.
3.1. Design parameters
Based on the results of previous research, a number of poten-
tial design parameters were identified to determine the response
of building connections. These include concrete compressive
strength, joint shear stress demand, bond stress demand, ratio of
anchorage length to longitudinal bar diameter, column axial load,
aspect ratio of the joint, transverse horizontal steel reinforcement
ratio within the joint, interior column longitudinal reinforcement
ratio within the joint, yield strength of longitudinal reinforcing
steel, and the ratio of the total strength of the beam top longitu-
dinal steel to that of the bottom. Drift demand and drift history
were not considered as experimental data since drift has a mini-
mal impact on connection strength, though drift demand and his-
tory may affect strength deterioration once a failure mechanism
forms [30,31]. A preliminary calibration of a probabilistic model
(explained in detail in Section 4), using all of the considered pa-
rameters, resulted in identification of a reduced set of parameters
that have a significant impact on connection response. These pa-
rameters, as well as previous experimental research confirming the
importance of these parameters in determining the seismic perfor-
mance of connections, are discussed in the following subsections.
Table 1 provides statistics (e.g. minimum, maximum, mean) for
these parameters for the entire experimental data set; computed
values for individual test specimens are provided in Mitra [48].
3.1.1. Nominal joint shear stress demand
To ensure that joint failure does not determine connection
response, ACI 318–05 [33] and ACI 352–02 [32] limit joint shear
stress demand for special moment resisting frames (SMRFs), which
are expected to experience multiple post-yield cycles of loading
 3.1.3. Joint transverse steel contribution to nominal joint shear force
during an earthquake. This limit is defined as 15 fc′ psi where fc′
is the concrete compressive strength in psi (1.2 fc′ MPa for fc′ in

MPa). ACI 352–02 [32] provides recommendations for calculating
this quantity. The results of a number of previous experimental
investigations indicate that high joint shear stress demand can
result in significant joint damage, loss of joint stiffness and loss of
joint strength prior to flexural yielding of the beams that frame into
the joint (i.e. joint failure) [3–13].
For the current study, nominal joint shear stress, τ , is defined as
the shear stress in the joint when beams reach nominal flexural
strength on either side of the joint, normalized by the square root
of the concrete compressive stress:
 
1 ML + MR
τ= √ − Vc (1)
f c hc b j hb
where hc is the height of the column, bj is the maximum out-of-
plane dimension of the beam or column, hb is the height of the
beam, fc is the concrete’s compressive strength, ML and MR are the
flexural strengths of the beam on the right and left of the joint
computed in accordance with ACI 318–05, and Vc is the lateral
load applied to the top of the column at the nominal strength of
the beams. This definition is similar to that provided by ACI Com.
352 with the exception that it (1) employs a slightly larger joint
volume with the result that horizontal and vertical shear stresses
are equal and (2) defines demand on the basis of frame member
flexural strengths rather than longitudinal steel areas with the
result that the determination of the frame member moments and
column shear is consistent.
It should be noted that because the nominal joint shear stress
demand defined by Eq. (1) is computed using the nominal flexural
strengths of the beams framing into the joint, it is neither the
maximum shear stress demand carried by nor the true joint shear
strength of the experimental test specimens The nominal shear
stress is used in the model because it is a value that is computed
easily for an existing or newly designed connection, and the
maximum strength of existing or newly designed connections is
not readily available.
3.1.2. Nominal bond stress demand within the joint
ACI 318–05 [33] requires that for joints in SMRF’s the width
of the joint should exceed 20 times the maximum diameter
of the beam’s longitudinal reinforcement. This requirement is
intended to limit the bond stress demand for beam reinforcement,
which is expected to yield under seismic loading. The results of
previous research indicate that short anchorage lengths and high
average bond stress demand can result in joint damage, reduced
joint strength, joint failure preceding beam yielding, and reduced
drift capacity when connection strength is determined by beam
yielding [3,6,7,12,16,19–23,40,42,43].
For evaluation of joints of variable design, the ratio of column
dimension to bar diameter is of limited value as it does not account
for the impact of concrete strength on bond strength or steel
strength on bond demand. Thus, for the current study, bond stress
demand is characterized using the bond index, µ, proposed by
Kitayama et al. [7]:
fy db
µ= √ (2)
2hc fc
where fy is the actual yield strength of the beam reinforcement,
db is the beam bar’s diameter, and hc and fc are as defined
previously. The bond index is the maximum bond stress demand
within the joint, normalized by the square root of the concrete’s
compressive strength, assuming that beam steel yields in tension
and compression on opposite sides of the joint.
The results of previous research show that the volume of
transverse reinforcement within the joint can determine joint
strength and deformation capacity under seismic loading. Restrepo
and Lin [17] show that the volume of transverse reinforcement
within the joint determines the flow of forces in the joint core.
Joh et al. [6] conclude that joint transverse reinforcement reduces
slippage of the beam bars in the joint, and thereby enhances joint
stiffness after cracking. These conclusions have been supported
by the experimental investigations [3,14–16]. ACI 318–05, on
the basis of the above research, requires a minimum transverse
reinforcement ratio within the joint.
Thus, for the current study, the transverse reinforcement ratio
is included in the model as the ratio of the total force carried by the
joint’s transverse reinforcement, assuming it yields, to the nominal
joint shear force demand:
Ast_T fyt
Φ= √ (3)
τ fc hc bj
where Ast_T is the total area of joint transverse reinforcement
passing through a plane normal to the beam axis, fyt is the actual
yield strength of the joint’s transverse reinforcement, and τ , fc , hc
and bj are as defined previously.
 N. Mitra et al. / Engineering Structures 33 (2011) 154–162
Table 1
Summary statistics for joint design parameters for data set presented in Mitra [48].
Parameter Units Minimum value
√
τ √MPa 0.23
µ MPa 0.92
Φ Unitless 0.00
p Unitless 0.02
ϖ Unitless 0.99
3.1.4. Column axial load ratio
There is no consensus within the research community as to the
impact of column axial load on the seismic response of joints. It has
been argued that axial load improves the shear resistance of the
beam–column joint by confining the joint core or by equilibrating
part of the load transferred by an inclined compressive strut that
forms in the joint as a result of joint shear action [17,27,29].
Agababian et al. [24] conclude that column axial load affects both
strength and deformation capacity, stating that an increase in
column axial load induces an increase in joint shear capacity of
approximately 30% and an increase in the deformation capacity
of approximately 50%. Kitayama et al. [7], however, conclude
that while a column axial stress less than 0.5f ′ c does not affect
joint strength, high axial stress accelerates strength loss in the
diagonal compression strut that forms in the joint core. Similarly,
Fu et al. [26] conclude that while an increase in column axial load
prior to the formation of a diagonal strut in the joint inhibits yield
penetration for beam longitudinal reinforcement, an increase in
column axial load after formation of a diagonal strut increases the
compression force in the strut and the likelihood of crushing of
joint core concrete. Based on experimental observations, it has also
been concluded by others that column axial load affects only joint
deformation and damage but not strength [14,25,28]. Meinheit and
Jirsa [9] conclude merely that changes in column axial load result
in changes in the pattern of cracking in the joint as well as the shear
at which joint cracking initiates.
Given the above, the impact of column axial load on connection
response was investigated by including the column axial load ratio,
p, in the model:
P in either positive or negative bending. Beam yield strength was
p=
Ag fc
where P is the column axial load, Ag is the gross cross-sectional
area of the column, and fc is as defined previously.
It should be noted that the column’s axial load was held
approximately constant for all of the experimental tests included
in the data set. A few previous studies have addressed the impact
on structural response of variation in column axial load due to the
overturning effect and vertical excitation of earthquake loading.
Agababian et al. [24] and Higazy et al. [57] conclude that variations
in column axial force due to earthquake can result in a significant
deterioration in the structural response. However, too few tests
were found in the literature in which variable column axial loading
was applied to the specimen to enable calibration of the statistical
model for this effect.
3.1.5. Ratio of top to bottom beam longitudinal steel strength
Research by Ichinose [58] indicates that the difference in the
total strength provided by the top and bottom beam reinforcement
influences connection behavior. If there is a significant difference
in the strength of top and bottom steel, it is more likely that
the beam will reach the lower flexural yield strength (associated
with yielding of the smaller volume of longitudinal steel) prior to
joint failure. Once beam yielding occurs, if there is a significant
difference in the strength of top and bottom reinforcement, cyclic
loading will result in significantly different steel and concrete
157
Maximum value Mean Coefficient of variation.
2.32 0.86 0.49
4.28 1.92 0.40
2.76 0.63 1.04
0.48 0.14 0.62
2.50 1.25 0.32
stress–strain histories on the top and bottom of the beam and,
thus, significantly different joint boundary and loading conditions.
On the side of the beam with lower steel strength, cyclic
loading will result in yielding of steel in compression as well
as tension, minimal accumulated plastic strain, and likely the
closing of concrete cracks. This could be expected to result in
premature deterioration of beam flexural strength. On the side
of the beam with greater steel strength, cyclic loading will result
in accumulated tensile strain in the steel, progressive widening
of concrete cracks, and no closing of cracks under compressive
loading. This could be expected to impact the formation of a
concrete compression strut within the joint as well as increase
yield penetration into the joint, and thereby reduce drift capacity.
Thus, the beam top to bottom steel strength ratio, ϖ is included
in the model:
(n.fy .As )bt
ϖ = (5)
(n.fy .As )bb
where the subscript (.)bt refers to the top longitudinal beam bars
and subscript (.)bb refers to the bottom, n is the number of bars, fy
is the actual yield strength of the steel, and As is the cross-sectional
area of a single bar.
3.2. Response parameters
For the current study, specimens are considered to exhibit
either ‘‘joint failure’’, in which the maximum strength of the
connection is less than that required to develop the yield strength
of the beams, or ‘‘beam yielding’’, in which the joint strength is
sufficient to enable beams to develop yield strength. Beam yield
strength was defined by first yield of the beam reinforcement,
(4) computed by performing a moment curvature analysis of a fiber-
type discretization of the beam section in which concrete fibers
were modeled based on a modified Kent–Park model and steel
fibers were modeled by a bilinear steel hardening response. The
classification of each specimen in the data set is provided in
Mitra [48].
For the data set used in the study, 72 specimens exhibited beam
yielding prior to joint failure and 38 specimens exhibited joint
shear failure prior to beam yielding.
3.3. The relationship between design and response parameters
The results of previous experimental research suggest that the
joint design parameters defined above determine the likelihood of
a connection response being determined by joint failure or beam
yielding. However, the results of previous experimental research
do not provide models that (1) directly link design parameters with
a response mechanism, (2) quantify the likelihood of connection
response being determined by a particular mechanism, and (3)
quantify the relative importance of individual design parameters
in determining the response. To provide preliminary information
about the relationship between connection design and response
parameters, these data were plotted for the specimens in the data
set (Fig. 1). The data in Fig. 1 support the following conclusions.
• The data in Fig. 1(a,b), on average, support the commonly held
understanding [e.g., 3, 16, 38, 43] that high joint shear stress
demand and high bond stress demand increase the likelihood of
158 N. Mitra et al. / Engineering Structures 33 (2011) 154–162

a brittle joint shear failure prior to beam yielding. However, the Table 2
data in these figures show that there are exceptions, suggesting Logit model results.

that other design parameters must also be considered to Covariate Estimated β t-stat p-value Odds ratio
accurately predict the likelihood of joint failure prior to beam τ /τmean 1.63 1.86 0.0624 5.10
yielding. µ /µmean 3.90 3.44 0.0060 49.4
• The data in Fig. 1(c) show that connections in which joints have ϕ/ϕmean −1.40 −1.88 0.0603 0.25
a high ratio of transverse steel strength to total shear demand p/pmean −1.64 −2.59 0.0095 0.19
ϖ /ϖ mean −3.04 −2.24 0.0249 0.05
Eq. (3) are more likely to exhibit beam yielding prior to joint
constant −0.66
failure. Similar observations have been made by others using
smaller data sets [6,15,16,25].
• The results of previous research suggest that column axial 4.1. The binomial logit model
load [7,9,14,17,24–29] and the ratio of beam top to bottom
longitudinal reinforcement ratio [14,16,58] also influence In this study, the dependent variable Y in Eq. (6) represents
connection response. However, the data in Fig. 1(d,e) do not the likelihood of observing a brittle joint failure described as ‘‘joint
suggest a strong relationship between failure mode and these failure prior to beam yielding’’ versus a more ductile joint response
parameters individually. described as ‘‘beam yielding prior to joint failure’’. To facilitate
presentation of the calculations, the case of ‘‘joint failure prior to
4. Methodological approach beam yielding’’ is referred to as Event 1 and the case of ‘‘beam
yielding prior to joint failure’’ is referred to as Event 0. It should be
Experimental observations (Section 3.1) provide a qualitative noted that the values 0 and 1 do not have any physical significance,
measure of the impact of various design parameters on connection and other values could have been used, such as 200 and 150.
response. For the current study, a statistical model linking quan- Additionally, only event values of 0 and 1 are acceptable and event
titative design parameters and qualitative connection response values between 0 and 1 are meaningless.
was desired. Linear and/or nonlinear regression is one possible ap- In Eq. (6), the likelihood of observing a discrete event of brittle
proach for developing such a model; however, this is not ideal joint failure is defined by the log of the odds ratio for that event. The
because it requires assigning a quantitative measure to the qual- odds ratio for Event 1 is the ratio of the probability of occurrence
itative connection response parameter [59]. Logistic regression, of Event 1, PE =1 , to the probability of occurrence of Event 0, PE =0 .
however, is ideally suited for developing this type of model. This Thus,
method allows for quantification of the conditional probability of     K
a qualitative measure occurring based on quantitative data. Lo- P E =1 P E =1 −
gistic regression is an established technique, in fields other than Y = log = log = β0 + β k Xk (7)
1 − P E =1 P E =0 k=1
structural engineering, for developing relationships between qual-
itative response variables and a set of independent quantitative where βi are logistic regression parameters, Xi are the joint design
parameters [60–65]. Additionally, logistic regression was used by parameters, normalized by the mean value for the specimens
Liao et al. [66] to define the probability of soil liquefying under in the data set, and K is total number of design parameters
earthquake excitation. For the current study, logistic regression considered. As described in Ben-Akiva and Lerman [60], Eq. (7) may
was used to develop a relationship between the mechanism that be manipulated to define the probability of occurrence of Events 1
determines the connection response described in Section 3.2 (a and 0:
qualitative measure) and the set of independent design parame-
K
ters described in Section 3.1 (quantitative measures). β0 + βk Xk
∑

In comparison with linear/nonlinear regression, the logistic e k=1

P E =1 = (8a)
regression model has less stringent requirements since it does K
β0 + βk Xk
∑
not assume linearity (or nonlinearity) of relationship between 1+e k=1
the independent variables and the dependent variable, does not
require normally distributed variables, and does not assume 1
P E =0 = K
. (8b)
homoscedasticity [62]. Both the linear and the logistic models β0 +
∑
βk Xk
employ the regression relationship: 1+e k=1

Y = β0 + β1 X1 + β2 X2 + β3 X3 + βk Xk .
However, while an ordinary linear regression model defines a
relationship between continuous variables, Xi , and a continuous
dependent variable, Y , a logistic regression model defines a
relationship between continuous variables, Xi , and the likelihood
of occurrence of a discrete event, Y . For a linear regression
model, model parameters, βi in Eq. (6), are computed to minimize
an error function, typically the sum of the squares of the
difference between the measured and computed (Eq. (6)) response
parameter Y . In a logistic regression model, the discrete nature
of the variable Y precludes this approach. Thus, the method of
maximum likelihood [64], which provides a means of choosing an
asymptotically efficient estimator for a set of parameters, typically
is used to compute logistic regression parameters, βi . A logistic
model may be used to predict the likelihood of multiple discrete
outcomes. For the current study, only two outcomes of joint failure
prior to beam yielding and/or beam yielding prior to joint failure
were considered. Thus, a binomial logit model was developed for
the study.
(6)
5. Analysis results and the logit model
The STATA econometric software package [67] was used to
obtain the regression parameters, βi in Eq. (7), using the method
of maximum likelihood. For each of the independent variables,
Table 2 shows the computed regression parameters, βi , as well
as statistical parameters for use in model evaluation. Having
determined the logistic regression parameters βi, one can use
Eq. (8) to evaluate the probability of occurrence of Event 1 and
Event 0.
5.1. Impact of design parameters on connection failure
The computed regression parameters provide insight into the
impact of individual design parameters on connection response.
Given the definition of Y in Eq. (7), the sign of a regression
parameter indicates whether an increase in the associated design
parameter increases or decreases the likelihood of joint failure
 N. Mitra et al. / Engineering Structures 33 (2011) 154–162 159

(a) Variation of nominal joint shear stress demand with (b) Variation of bond stress demand with failure
failure initiation mode. initiation mode.

(c) Variation of column axial load ratio with failure (d) Variation of transverse steel contribution to nominal
initiation mode. joint shear stress with failure initiation mode.

(e) Variation of ratio of the total force of the top and the
bottom longitudinal beam bars with failure initiation
mode.

Fig. 1. Variation of demand parameters with failure initiation mode.

prior to beam yielding. A positive (negative) regression parameter
indicates that increasing the associated design parameter increases
(decreases) the likelihood of a joint failure prior to beam yielding.
Similarly, a negative (positive) regression parameter indicates that
increasing the associated design parameter increases (decreases)
the likelihood of beam yielding prior to joint failure. The relative
magnitude of a coefficient indicates the relative importance of
the parameters in determining connection response. The following
paragraphs compare the analysis results with those of previous
experimental investigations of joint behavior.
5.1.1. Influence of nominal joint shear stress demand
The positive sign of the regression coefficient associated with
nominal joint shear stress demand (Table 2) indicates that higher
shear stress results in an increased probability of brittle joint
failure prior to beam yielding. As discussed in Section 3.1.1, a
similar conclusion has been reached by others [4–10,13,40]. The
relative magnitude of the regression parameter indicates that
nominal joint shear stress is not the most critical design parameter
in determining the likelihood of joint failure prior to beam yielding
but is instead significantly less important than either bond stress
demand or beam reinforcement ratio. While no previous studies
have sought to identify and quantify the relative importance of
the design parameters identified in Section 3.1 in determining
connection response, this result contradicts general understanding
within the community, as incorporated in the ACI Code [33] that
shear stress demand is the most critical parameter in determining
connection response.
5.1.2. Influence of beam bar bond stress demand
Analysis results in Table 2 indicate also that increased
beam bar bond stress demand results in increased likelihood
of joint failure determining connection response. As discussed
in Section 3.1.2, this conclusion is supported by previous
160 N. Mitra et al. / Engineering Structures 33 (2011) 154–162
experimental research [3,20,21] The relative magnitude of the
regression coefficient indicates that bond stress demand is the
most influential parameter in predicting the connection response
mechanism. The importance of bond in determining connection
response is also supported by previous research [68–70] as well
as the ACI Code [33]. However this study represents the first
attempt to quantify the relative importance of this parameter
in comparison with other parameters that influence connection
response.
5.1.3. Influence of transverse steel contribution to nominal joint shear
force
Statistical analysis results (Table 2) indicate that increasing
the ratio of joint transverse steel strength to nominal joint
shear demand reduces the likelihood of joint failure prior to
beam yielding. As discussed in Section 3.1.3, this conclusion is
supported by the results of previous experimental studies. The
experimental investigations [3,14–16] conclude that increasing
the volume of joint transverse steel inhibits joint failure prior to
beam yielding, and the investigations [6,16] conclude that reducing
shear stress demand reduces the likelihood of joint failure. The
relative magnitude of the regression parameter indicates that the
transverse steel ratio defined by Eq. (3) is the least important of
the design parameters considered in determining the likelihood of
joint failure determining connection response.
5.1.4. Influence of column axial load ratio
The data in Table 2 show that increasing column axial load
ratio reduces the likelihood joint failure prior to beam yielding.
As discussed in Section 3.1.4, there is no consensus within the
community as to the impact of column axial load on connection
response. Thus, this result supports the conclusions of some
previous studies [17,24,26].
5.1.5. Influence of the ratio of the top to bottom beam longitudinal
steel strength
The analysis results presented in Table 2 support the conclusion
of Ichinose [58] that an increase in the ratio of the strength of
beam top to bottom reinforcement will increase the likelihood
of beam yielding. The relative magnitude of the regression
coefficients indicates that the ratio of beam steel strengths
is one of the most important parameters in determining if
beam yielding will precede joint failure. The paucity of previous
experimental studies investigating this parameter, suggests that
few in the community consider this parameter to be important
in determining connection behavior. Thus, this result indicating
the importance of this parameter, both contradicts general
understanding of the parameters that affect connection response
as well as suggests an important direction for future experimental
testing.
5.2. The odds ratio for evaluation of the relative importance of model
variables
The relative importance of independent variables in determin-
ing the likelihood of an Event occurring can also be assessed using
the odds ratio. When quantitative or continuous independent vari-
ables are used in binomial logistic estimation, the natural logarith-
mic exponential of the logistic coefficient equals a multiplicative
factor by which the predicted odds changes given a unit increase
in the predictor variable and holding constant all other predictor
variables in the equation. The odds ratio is estimated as follows:
K
β0 +
∑
βk Xk
PY =1
odds ratio = =e . k=1 (9)
PY =0
The odds ratio ranges in value from zero to positive infinity. If
the natural logarithmic exponential of the coefficient is equal to
1.0 (i.e., βi , = 0), then changes in the predictor have no effect on
the predicted odds. If the natural logarithmic exponential of the
coefficient is greater than 1.0 (i.e. βi , > 0), then an increase in
the predictor will yield an increase in the predicted odds. If the
natural logarithmic exponential of the coefficient is less than 1.0
(i.e. βi , < 0), then an increase in the value of the predictor will
yield a decrease in the predicted odds. The relative magnitudes of
the odds ratios for each independent variable in the logit model can
be compared to assess the relative importance of each variable in
determining the likelihood of occurrence of either of the events.
To determine the relative importance of all of the independent
variables, odds ratios greater than 1.0 must be compared with the
reciprocal of odds ratios less than 1.0. It should be noted that this
evaluation of odds ratio is identical to evaluation of the relative
magnitudes of the model coefficients.
6. Evaluation of the model
6.1. Statistical verification of the significance of the model parameters
To identify the significance of the independent variables t-
statistic and p-values were computed. The t-statistic determines
whether the coefficient of a particular parameter is significantly
different from zero, i.e. whether the null hypothesis H0 : βi = 0
can be rejected or not, and the p-value provides the probability
at which the null hypothesis cannot be rejected [61,65]. The
computed t-statistics and p-values, as shown in Table 2, indicate
that except variables τ /τmean and ϕ/ϕmean , all parameters are
significant at least at the 5% level. Variables τ /τmean and ϕ/ϕmean
are significant at about the 6% level, which indicates that all the
parameters in this study are significant at least at the 10% level.
Hence, the null hypothesis that these variables have no influence
on the failure mechanism could be rejected and the relationship
between the dependent variable and a set of independent variables
cannot be attributed by chance.
6.2. Goodness-of-fit of the model
To further evaluate the model, two goodness-of-fit tests are
performed, since no single test can be considered comprehensive.
The first test is the Hosmer and Lemeshow test [62] where the
test statistic follows a chi-square distribution as shown by Hosmer
and Lemeshow and the second one is the log-likelihood ratio
test [64,65].
To compute the Hosmer and Lemeshow statistic [62], the
total number of specimens is partitioned into 10 equal-sized
groups on the basis of the predicted probability of joint failure
preceding beam yielding. The statistic, which follows a chi-square
distribution with 8 degrees of freedom, was computed as 7.67
with p-value of 0.47. Since this p-value is more than 0.05 it is not
considered to be significant and we fail to reject the null hypothesis
that there is no difference between observed and model-predicted
values. Thus, the test implies that the model’s estimates fit the data.
The likelihood ratio test [65] statistic is
X 2 = 2[LL(βR ) − LL(βU )] (10)
where LL(βR ) is the log likelihood at convergence of the ‘‘restricted’’
model (i.e. model in which all the parameters except constant term
is equal to 0), and LL(βU ) is the log likelihood at convergence of
the ‘‘full’’ or ‘‘unrestricted’’ model (i.e. model under investigation
defined by β values listed in Table 2). The X 2 statistic is chi-square
distributed with the degrees of freedom equal to the difference in
the number of parameters in the ‘‘restricted’’ and ‘‘unrestricted’’
model. For the proposed logit model, the value of the obtained
log-likelihood for the ‘‘unrestricted’’ or ‘‘full’’ model is −44.9
whereas for the ‘‘restricted’’ model is −70.91. The chi-squared
 N. Mitra et al. / Engineering Structures 33 (2011) 154–162
Fig. 2. Probability of occurrence of Event 1 with different failure initiation modes.
value obtained is 52.63 with 5 degrees of freedom, resulting in
a p-value of less than 0.001. Thus, inclusion of the independent
parameters in the predictive model significantly improves the
goodness-of-fit.
6.3. Predictive efficiency of the model
To assess the predictive efficiency of the statistical model,
the likelihood of joint failure prior to beam yielding (Event 1),
computed using Eq. (8a) with βi from Table 2, and was plotted
versus the observed event. Specimens from the data set exhibiting
beam yielding prior to joint failure (Event 0) are plotted as circles
and specimens exhibiting joint failure prior to beam yielding
(Event 1) are plotted as squares. If the model were perfect, all
specimens exhibiting Event 0 would have a computed probability
of occurrence of Event 1 of 0.0; while all specimens exhibiting
Event 1 would have a computed probability of occurrence of 1.0.
The data in Fig. 2 show that the model is not perfect. However,
using a probability of 50% as indicative of response, the model
correctly predicts the connection response mechanism for 82% of
the specimens.
7. Application of the logit model
The previous sections describe the model’s calibration as
well as the statistical significance, goodness-of-fit, and predictive
efficiency of the model. Following is the recommended process for
using the model to predict the mechanism that determines the
seismic response of a particular connection:
1. Compute the nominal flexural strength of the beams framing
into the joint per the ACI Code [33].
2. Compute the critical design parameters included in the logit
model using Eqs. (1)–(5).
3. Normalize the design parameters by the mean values computed
for the data set used in the model calibration (Table 1).
4. Use Eqs. (8a) and (8b) with the β values listed in Table 2
to obtain, respectively, the probabilities of joint shear failure
preceding beam yielding and beam yielding preceding joint
failure.
To further evaluate the model, the above process was used to
predict the response mechanism for two connections tested by
Goto and Joh [50] but not included in the data set used for model
calibration. The first of these connection, identified by Goto and
Joh as ‘‘LO’’ in the ‘‘J-Series’’, had a high joint shear stress demand,
high bond stress demand, low transverse reinforcement ratio, high
column axial load and equal top and bottom beam reinforcement
strength (τ = 1.56; µ = 3.15; Φ = 0.0067; p = 0.3 and ϖ = 1).
The computed probability of joint failure prior to beam yielding
for this specimen is 94% in comparison to the probability of beam
yielding followed by joint failure, which was obtained as 6%.
161
The second connection, identified by Goto and Joh as ‘‘PH’’ in
the ‘‘BJ-Series’’, was observed under laboratory loading to exhibit
beam yielding prior to joint failure. This specimen had an average
joint shear stress demand, low bond stress demand, high joint
transverse reinforcement ratio, average column axial load, and a
significantly greater top beam reinforcement strength than the
bottom (τ = 0.89; µ = 1.54; Φ = 0.08; p = 0.17 and ϖ = 1.4).
The computed probability of joint failure prior to beam yielding
for this specimen is 19% in comparison to the probability of beam
yielding followed by joint failure, which was obtained as 81%.
Thus, the statistical model, expressed by Eq. (8) and Table 2
predicts the observed connection response mechanism given the
material and geometric properties of the specimen.
8. Summary, conclusions and future research
A probabilistic model, predicting the mechanism (joint failure
or beam yielding) that will determine the connection strength
and seismic response, was developed using the results of 110
laboratory tests of interior beam–column connections. The model
predicts the likelihood that the connection response will be
determined by joint failure versus beam yielding on the basis of
five, easily computed, joint design parameters: nominal joint shear
stress demand, average bond stress demand, the ratio of joint
transverse reinforcement strength to joint shear stress demand,
column axial load ratio, and the ratio of beam top to bottom
longitudinal reinforcement strength. Model calibration factors
indicate the relative importance of these design parameters in
determining the connection’s response mechanism.
The primary conclusions of the research presented in the
manuscript are:
• The binomial logit probabilistic model, developed in this
research, is able to identify, with reasonable accuracy, the
mechanism that will determine connection response.
• The proposed probabilistic model provides a relative quantita-
tive estimate of the design parameters found to determine con-
nection response. The beam bar bond stress demand parameter
was found to be the most important parameter which deter-
mines connection response.
From the perspective of probabilistic determination of failure
of structural components, this research presents a new approach
for establishing the relationship between a qualitative parameter,
such as failure mechanism, and quantitative design parameters.
It is recommended that similar models be developed for other
structural components. It is also recommended that a similar
logistic regression strategy be developed for other qualitative
parameters, such as the damage state of the structure or
component; this type of model could be used to advance
performance-based design. Finally, the predictive efficiency (as
discussed in Section 6.3) of the proposed probabilistic model
could be improved by employing a Bayesian logistic regression
framework in which the prior information about the effect of
the demand parameter on the type of failure initiation of RCBC
connection is incorporated. With such prior information exact
shape of posterior distribution of coefficients can be assessed from
repeated sampling from the posterior, thus improving the overall
explanatory power of the model.
References
[1] Hall J. Northridge earthquake preliminary reconnaissance report. Tech. rep.,
EERI, January 1994.
[2] Theiss AG. Modeling the earthquake response of older reinforced concrete
beam–column building joints. Master’s thesis, Dept. Civil Eng., Univ. of
Washington (Seattle, WA): 2006.
[3] Durrani AJ, Wight JK. Experimental and analytical study of beam to column
connections subjected to reverse cyclic loading. Tech. Rep. UMEE 82 R3, Dept
Civil Eng, Univ. Michigan: 1982.
162 N. Mitra et al. / Engineering Structures 33 (2011) 154–162
[4] Goto Y, Joh O. An experimental study on shear failure mechanism of reinforced
concrete interior beam–column joints. In: Proc. 11th world conf. earthquake
engineering. 1996. Paper no. 1194.
[5] Higashi Y, Ohwada Y. Failing behaviors of reinforced concrete beam column
connections subjected to lateral load. Mem Fac Technol Tokyo Metrop Univ
Japan 1969;19:91–101.
[6] Joh O, Goto Y, Shibata T. Influence of transverse joint, beam reinforcement
and relocation of plastic hinge region on beam–column joint stiffness
determination. In: ACI SP 123–12: design of beam–column joints for seismic
resistance. (Farmington Hills, MI): ACI; 1991. p. 187–223.
[7] Kitayama K, Otani S, Aoyama H. Earthquake resistant design criteria for
reinforced concrete interior beam–column joints. In: Pacific conference on
earthquake engineering. 1987, p. 315–26.
[8] Leon RT. Shear strength and hysteretic behavior of interior beam–column
joints. ACI Struct J 1990;87(1):3–11.
[9] Meinheit DF, Jirsa JO. The shear strength of reinforced concrete beam–column
joints. In: Tech. Rep. CESRL Report no. 77-1, Univ. Texas at Austin. 1977.
[10] Morita S, Kitayama K, Kishida S, Nishikawa T. Shear force and capacity in
reinforced concrete beam–column joints with good bond along beam and
column bars. In: Proc. 13th world conf. earthquake engineering. 2004. Paper
no. 1761.
[11] Murakami H, Fujii S, Ishiwata Y, Morita S. Shear strength and deformation
capacity of interior reinforce concrete beam–column joint sub-assemblage In:
Proc. 12th world conf. earthquake engineering. 2000. Paper no. 679.
[12] Teraoka M, Kanoh Y, Tanaka K, Hayashi K. Strength and deformation
behavior of RC interior beam–column using high-strength concrete. In:
Proc. 2nd US–Japan–New-Zealand–Canada multilateral meeting on structural
performance of high strength concrete in seismic regions. 1994.
[13] Zaid SSS. Behavior of reinforced concrete beam–column connections under
earthquake loading. Ph.D. thesis, Department of Architecture, Univ. of Tokyo;
Japan. 2001.
[14] Fujii S, Morita S. Comparison between interior and exterior RC beam–column
joint behavior. In: ACI SP 123–12: design of beam–column joints for seismic
resistance. (Farmington Hills, MI): ACI; 1991. p. 145–65.
[15] Kamimura T, Takeda S, Tochio M. Influence of joint reinforcement on strength
and deformation of interior beam–column sub-assemblages. In: Proc. 12th
world conf. earthquake engineering. 2000. Paper no. 2267.
[16] Oka K, Shiohara H. Tests on high-strength concrete interior beam–column
joint sub-assemblages. In: Proc. 10th. world conf. earthquake engineering.
1992. p. 3211–7.
[17] Restrepo JI, Lin CM. Evaluation of the shear strength of beam–column joints
of reinforced concrete frames subjected to earthquake loading. In: Proc. 12th
world conf. earthquake engineering. 2000. Paper no. 522.
[18] Hakuto S, Park R, Tanaka H. Effect of deterioration of bond on beam
bars passing through interior beam–column joints on flexural strength and
ductility. ACI Struct J 1999;96(5):858–64.
[19] Hayashi K, Teraoka M, Mollick AA, Kanoh Y. Bond properties of main rein-
forcing bars and restoring force characteristics in RC interior beam–column
sub-assemblage using high strength materials. In: Proc. 2nd US–Japan–New-
Zealand–Canada multilateral meeting on structural performance of high
strength concrete in seismic regions. 1994.
[20] Leon RT. Interior joints with variable anchorage lengths. J Struct Eng ASCE
1989;115(9):2261–75.
[21] Otani S, Kitayama K, Aoyama H. Beam bar bond requirements for interior
beam–column connections. In: Proc. int. symposium on fundamental theory
of reinforced and prestressed concrete, Nanjing Univ. Tech. China. 1986.
[22] Park R, Milburn JR. Comparison of recent New Zealand and United States
seismic design provisions for reinforced concrete beam–column joints and test
results from four units designed according to the New Zealand code. Bull New
Zealand National Society Earthquake Eng 1983;16(1):3–24.
[23] Park R, Ruitong D. A comparison of the behavior of reinforced concrete beam
column joints designed for ductility and limited ductility. Bull New-Zealand
National Society Earthquake Eng 1988;21(4):255–78.
[24] Agbabian MS, Higazy EM, Abdel-Ghaffar AM. Experimental observations on
the seismic shear performance of RC beam-to-column connections subjected
to varying column axial force. Earthq Eng Struct Dyn 1994;23:859–76.
[25] Bonacci JF, Pantazopoulou SJ. Parametric investigation of joint mechanics. ACI
Struct J 1993;90(1):61–71.
[26] Fu J, Chen T, Wang Z, Bai S. Effect of axial load ratio on seismic behavior of
interior beam–column joints. In: Proc. 12th world conf. earthquake eng, 2000.
Paper no. 2707.
[27] Hanson NW, Conner HW. Seismic resistance of reinforced concrete
beam–column joints. J Struct Divi ASCE 1967;93(5):533–60.
[28] Kurose Y. Recent studies on reinforced concrete beam column joints in Japan.
Tech. Rep. PMFSEL Report no. 87-8, Department of Civil Eng., Univ. of Texas at
Austin. 1987.
[29] Paulay T, Park R, Priestley MJN. Reinforced concrete beam column joints under
seismic action. J American Concrete Institute 1978;75(11):585–93.
[30] Alire DA. Seismic evaluation of existing unconfined reinforced concrete
beam–column joints. Master’s thesis, Dept. Civil Eng., Univ. Washington,
Seattle. 2002.
[31] Walker SG. Seismic performance of existing reinforced concrete beam–column
joints. Master’s thesis, Dept. Civil Eng., Univ. of Washington (Seattle, WA).
2001.
[32] ACI Committee 352. Recommendations for design of beam–column connec-
tions in monolithic reinforced concrete structures (ACI 352R-02). ACI (Farm-
ington Hills, MI). 2002.
[33] ACI Committee 318. Building code requirements for reinforced concrete (ACI
318–05) and commentary (ACI 318–R05). ACI (Farmington Hills, MI). 2005.
[34] Lehman D, Stanton J, Anderson M, Alire D, Walker S. Seismic performance of
older beam–column joints. In: Proc. 13th world conf. earthquake engineering.
2004. Paper no. 1464.
[35] Pessiki SP, Conley C, Gergely P, White RN. Seismic behavior of lightly reinforced
concrete column and beam–column joint details. In: Tech. Report NCEER 90-
0014. National Center for Earthquake Engineering Research, State Univ. of New
York at Buffalo. 1990.
[36] Attaalla SA, Agbabian MS. Performance of interior beam–column joints cast
from high strength concrete under seismic loads. Advances in Structural Eng
2004;7(2):147–57.
[37] Noguchi H, Fujii S, Teraoka M. Beam column joints with high strength
materials. In: Proc. 2nd US–Japan–New-Zealand–Canada multilateral meeting
on structural performance of high strength concrete in seismic regions. 1994.
[38] Noguchi H, Kashiwazaki T. Experimental studies on shear performances
of RC interior column-beam joints. In: Proc. 10th. world conf. earthquake
engineering. 1992, p. 3163–8.
[39] Sanada A, Maruta M. Seismic performance of high rise RC frame structure
using ultra high strength concrete. In: Proc. 13th world conf. earthquake
engineering. 2004. Paper no. 443.
[40] Teraoka M, Kanoh Y, Hayashi K, Sasaki S. Behavior of interior beam-and-
column sub-assemblages in an RC frame. In: Proc. 1st international conference
on high strength concrete. 1997, p. 93–108.
[41] ACI Committee 363. State of art report on high strength concrete (ACI
363R–92). ACI (Farmington Hills, MI) 1992, Reapproved. 1997.
[42] Beckingsale CW. Post elastic behavior of reinforced concrete beam–column
joints. Ph.D. thesis, Dept. Civil Eng., Univ. Canterbury, Christchurch, New
Zealand. 1980.
[43] Otani S, Kobayashi Y, Aoyama H. Reinforced concrete interior beam–column
joints under simulated earthquake loading. In: Proc. US-New Zealand-Japan
seminar on design of reinforced concrete beam–column joints. 1984.
[44] Alath S, Kunnath SK. Modeling inelastic shear deformation in RC beam–column
joints. In: Proc. 10th. conf. eng. mechanics. 1995. p. 822–5.
[45] Biddah A, Ghobarah A. Modeling of shear deformation and bond slip in
reinforced concrete joints. Struct Eng Mech 1999;7(4):413–32.
[46] El-Metwally SE, Chen WF. Moment rotation modeling of reinforced-concrete
beam–column connections. ACI Struct J 1988;85(4):384–94.
[47] Otani S. Inelastic analysis of reinforced-concrete frame structures. J Struct Div
ASCE 1974;100(ST7):1422–49.
[48] Mitra N. An analytical study of reinforced concrete beam–column joint be-
havior under seismic loading. Ph.D. thesis, Dept. Civil Eng., Univ. Washington;
(Seattle, WA) 2007.
[49] Altoontash A, Deierlein GG. A versatile model for beam–column joints. ASCE
structures congress (Seattle, WA) 2003.
[50] Elmorsi M, Kianoush MR, Tso WK. Modeling bond-slip deformations in
reinforced concrete beam–column joints. Canad J Civil Eng 2000;27(3):
490–505.
[51] Fleury F, Reynouard JM, Merabet O. Multi-component model of reinforced
concrete joints for cyclic loading. J Eng Mech ASCE 2000;126(8):804–11.
[52] Lowes LN, Altoontash A. Modeling reinforced concrete beam–column joints
subjected to cyclic loading. J Struct Eng ASCE 2001;129(12):1686–97.
[53] Mitra N, Lowes LN. Evaluation, calibration and verification of a reinforced
concrete beam–column joint model. J Struct Eng ASCE 2007;133(1):105–20.
[54] Tajiri S, Shiohara H, Kusuhara F. A new macroelement of reinforced concrete
beam column joint for elasto-plastic plane frame analysis. In: Proc. 8th.
national conf. earthquake engineering. 2006. Paper no. 674.
[55] Youssef M, Ghobarah A. Modeling of RC beam–column joints and structural
walls. J Earthq Eng 2001;5(1):93–111.
[56] Birss GR. The elastic behavior of earthquake resistant reinforced concrete
interior beam–column joints. In: Tech. Rep. 78-13, Dept. Civil Eng, Univ
Canterbury, New-Zealand. 1978.
[57] Higazy EMM, Elnashai AS, Agbabian MS. Behavior of beam–column connec-
tions under axial column tension. J Struct Eng ASCE 1996;122(5):501–11.
[58] Ichinose T. Shear reinforcement of reinforced concrete inelastic interior
beam–column joints. Bull New-Zealand National Society Earthquake Eng
1987;20(2):116–26.
[59] Zhu L, Elwood KJ, Haukaas T. Assessment of expected failure mode for
reinforced concrete columns. In: Proc. 8th US national conference on
earthquake engineering. 2006. Paper no. 1001.
[60] Ben-Akiva M, Lerman SR. Discrete choice analysis: theory and application to
predict travel demand. Cambridge, MA: MIT Press; 1985.
[61] Greene WH. Econometric analysis. Singapore: Pearson Education; 2000.
[62] Hosmer DWJ, Lemeshow S. Applied logistic regression. Wiley- Interscience;
2000.
[63] Jaccard JJ. Interaction effects in logistic regression. Sage Publications Inc; 2001.
[64] Menard SW. Applied logistic regression analysis. Sage Publications Inc; 2001.
[65] Washington SP, Karlaftis MG, Mannering FL. Statistical and econometric
methods for transportation data analysis. Chapman and Hall, CRC; 2003.
[66] Liao SSC, Veneziano D, Whitman RV. Regression models for evaluating
liquefaction probability. J Geotechnical Eng ASCE 1988;114(4):389–411.
[67] STATA, Release 9. Stata Press College Station. Texas. 2006.
[68] Pantazopoulou SJ, Bonacci JF. Consideration of questions about beam–column
joints. ACI Struct J 1992;89(1):27–36.
[69] Pantazopoulou SJ, Bonacci JF. On earthquake resistant reinforced concrete
frame connections. Canad J Civil Eng 1994;21:307–28.
[70] Shiohara H. New model for shear failure of RC interior beam–column
connections. J Struct Eng ASCE 2001;127(2):152–60.