replicating and extending the good-enough level model of change: considering session frequency
TRANSCRIPT
This article was downloaded by: [University of California, San Francisco]On: 16 December 2014, At: 01:59Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK
Psychotherapy ResearchPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/tpsr20
Replicating and extending the good-enough levelmodel of change: Considering session frequencyRobert J. Reese a , Michael D. Toland a & Nathaniel B. Hopkins ba University of Kentucky, Educational, School, and Counseling Psychology , Lexington, USAb University of Kentucky, University Counseling Center , Lexington, USAPublished online: 21 Jul 2011.
To cite this article: Robert J. Reese , Michael D. Toland & Nathaniel B. Hopkins (2011) Replicating and extending thegood-enough level model of change: Considering session frequency, Psychotherapy Research, 21:5, 608-619, DOI:10.1080/10503307.2011.598580
To link to this article: http://dx.doi.org/10.1080/10503307.2011.598580
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
Replicating and extending the good-enough level model of change:Considering session frequency
ROBERT J. REESE1, MICHAEL D. TOLAND1, & NATHANIEL B. HOPKINS2
1University of Kentucky, Educational, School, and Counseling Psychology, Lexington, USA & 2University of Kentucky,
University Counseling Center, Lexington, USA
(Received 26 November 2010; revised 11 June 2011; accepted 13 June 2010)
AbstractThe good-enough level (GEL) model posits that the rate of change in psychotherapy is related to the total dose of therapy. Thepsychotherapy dose-response literature has typically measured dose as number of sessions attended without considering thenumber of days or weeks it takes to complete the sessions (session frequency). The current study sought to replicate the GELmodel and explore if session frequency moderates the influence that the number of sessions has on the rate of change inpsychotherapy. An archived naturalistic data set with a US university counseling center sample (n�1,207), with treatmentprogress measured using the Outcome Questionnaire-45 (Lambert et al., 1996), was used. Our results are consistent with theGEL model (i.e., clients who attended fewer sessions evidenced faster rates of change), but extended it by showing that therate of change was also influenced by session frequency (i.e., clients who attended more sessions on average per weekdemonstrated more rapid improvement). Evidence suggests that clinicians and researchers should give consideration tosession frequency, both in their work with clients and how ‘‘dose’’ is operationalized in psychotherapy research.
Keywords: psychotherapy; dose response; client progress and outcome; session frequency
The psychotherapy outcome literature (e.g.,
Anderson & Lambert, 2001; Hansen & Lambert,
2003; Lutz, Lowry, Kopta, Einstein, & Howard,
2001; Okiishi et al., 2006) has generally found that
the dose-response relationship is marked by a
negatively accelerated curve. Specifically, clients
generally tend to improve early in treatment and
the benefit of therapy diminishes as clients attend
more sessions. This finding is so prevalent in the
outcome literature that many psychotherapy out-
come researchers assume that this outcome trajec-
tory is a fact of psychotherapy (e.g., Lutz,
Martinovich, & Howard, 1999).
Some researchers have questioned this assump-
tion, however, because the studies that find a
negatively accelerating curve rest on another as-
sumption that all clients attend a similar number of
sessions (e.g., Baldwin, Berkeljon, Atkins, Olsen, &
Nielsen, 2009; Barkham et al., 1996; Barkham et al.,
2008). Barkham et al. (1996) proffered that if the
number of sessions are taken into account, the nega-
tively accelerating curve may disappear because such
consideration may show that clients respond to
treatment at different rates. They referred to this as
the ‘‘good-enough level’’ (GEL) model, and hy-
pothesized that individuals who attend fewer sessions
do so because they respond more quickly to treat-
ment than those who attend more sessions: Thus,
improving to a just good enough level where therapy
is no longer needed. They further suggest that if
separate outcome trajectories were organized and
computed by the number of sessions attended, the
respective outcome trajectories would all be linear in
nature and the slopes would be steeper for those who
attended fewer sessions.
Baldwin et al. (2009) conducted a dose-response
study that compared the traditional aggregate model
(i.e., ignoring number of sessions attended by clients)
and a stratified model (i.e., accounting for the number
of sessions attended by clients as consistent with the
GEL model; we will use ‘‘GEL’’ when referring to
the ‘‘stratified model’’). Using multilevel growth
curve modeling, results indicated that the GEL model
more appropriately described the nonlinear growth in
the sample data, suggesting that the amount of change
on the Outcome Questionnaire-45 (OQ-45; Lambert
et al., 1996) reported by clients is a function of the
number of sessions attended. They found clients who
Correspondence concerning this article should be addressed to Robert J. Reese, University of Kentucky, Educational, School, &
Counseling Psychology, 235 Dickey Hall, Lexington 40506-0017, USA. Email: [email protected]
Psychotherapy Research, September 2011; 21(5): 608�619
ISSN 1050-3307 print/ISSN 1468-4381 online # 2011 Society for Psychotherapy Research
DOI: 10.1080/10503307.2011.598580
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
attended fewer sessions demonstrated faster rates of
improvement and those who attended more sessions
progressed slower. More specifically, outcome trajec-
tories for clients who attended fewer sessions (i.e.,
four or eight) demonstrated more immediate growth
but began to level out at the end of treatment.
Outcome trajectories for clients who attended several
sessions (i.e., 16 or 20) had slower initial growth,
which leveled out in the middle sessions, and then
became steeper toward the later sessions. Although
this lends support that sessions do not lessen in
potency over time for all clients, it does indicate that
sessions are not equivalent across individuals. More-
over, the Baldwin et al. findings are not entirely
consistent with Barkham et al.’s (1996) predictions
that the respective outcome trajectories would all be
linear in nature, but they did show steeper slopes for
those who attended fewer sessions.
Either way, it appears that both the traditional
aggregate and GEL models directly belie the idea of
sessions as equal in nature with regard to potency of
treatment. The aggregate dose-response model con-
tends that sessions cease to be equivalent given the
effect of diminished returns as treatment progresses,
and the GEL model posits that clients respond to
treatment doses (i.e., sessions) at different rates.
Hansen and Lambert (2003) mentioned this con-
cern when they noted that a limitation of their dose-
response study was the assumption that sessions
were equal in nature. Most, if not all, of the dose-
response literature makes this strict assumption that
one session equals one unit or dose. Psychotherapy’s
use of the word ‘‘dose’’ implies a level of specificity
similar to a dosage of medication given. To continue
this analogy, medication is also typically adminis-
tered at regular intervals, such as 100 mg every
12 hours. There are two potential problems with
applying this analogy to therapy. First, we cannot
assume that all therapy sessions are equal in dosage
given the wide variety of treatments employed by
different clinicians with complex individuals who
present with a myriad of issues that may respond to
any one session differently (Hansen & Lambert,
2003). Second, it is problematic to overlook when
the dosage is delivered. With medication, the interval
between administrations is taken into consideration.
An assumption the psychotherapy outcome literature
often seems to make is that psychotherapy follows
the traditional approach of one 50-minute session
conducted once per week and thus the interval
between the sessions is equal. There is some research
indicating that clients do not cleanly conform to this
attendance pattern (Kraft, Puschner, & Kurdy,
2006), but this variable is rarely mentioned as a
potential confound.
The GEL model has attempted to address the first
concern (i.e., sessions not being equal in potency
across all clients) by attempting to more fully
consider that clients may respond differently to
sessions. The model considers the context of the
client and therapist more fully, positing that clients
improve at different rates and that therapy is
continued until the client is improved enough to no
longer need treatment. However, the second concern
has received less attention. Session limits are becom-
ing the norm rather than the exception. Inspection of
the dose response literature might lead one to
conclude that it would be equally beneficial to space
sessions out, for instance attending therapy once
every 2 weeks in order to extend the time in therapy
or meet the demand by being able to see more
clients. Conversely, one might also conclude that
having more than one session per week may accel-
erate growth with clients. Is either approach plau-
sible? Is the growth rate for those attending eight
sessions in 4 months equivalent to eight sessions in 2
months? The current dose-response literature does
not provide a clear answer.
The reality is that many clients are not seen for
therapy on a strict week-to-week basis for a variety of
reasons: intentionally spacing sessions to meet
agency or insurance policy limits, generally good
functioning, missed and rescheduled appointments,
lack of motivation, poor time management, or even
holidays. As a result, dose-response studies do not
take into account the passage of time between
sessions during which clients may be making sig-
nificant progress, or lack thereof, on their therapeu-
tic concerns. Of course, there may also be differences
between a client-therapist dyad that choose to meet
every other week versus one that is intentionally
more flexible, or a client who periodically misses a
scheduled weekly session. In this study, we investi-
gate how psychotherapy outcome is impacted by the
interaction of the number of sessions attended and
frequency at which sessions are attended, providing a
more detailed picture of how clients respond to
therapy.
Most of the psychotherapy research investigating
the influence of session frequency has been con-
ducted in Europe (Freedman, Hoffenberg, Vores, &
Frosch, 1999; Kordy, Rad, & Senf, 1988; Kraft
et al., 2006; Sandell, Blomberg, & Lazar, 2002) and
is from the psychoanalytic tradition, where it is
common to have more frequent and more numerous
sessions. Freedman et al. (1999) found that clients
attending two or three sessions per week benefitted
more than clients who attended one session per
week. However, Kordy et al. (1988) did not find that
session frequency influenced outcome.
Psychotherapy dose-response 609
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
A more recent study (Kraft et al., 2006) found that
session frequency does matter with psychodynamic
and psychoanalytic therapy but not cognitive-
behavioral therapy. They found that the more weeks
clients went without treatment attenuated the ther-
apeutic alliance for those who received psychody-
namic treatment. Clients in psychoanalytic-based
treatment who missed more sessions early in treat-
ment improved more slowly across their treatment.
Interestingly, they also found that the clients in the
study attended less than one session per week on
average, and that the level of initial impairment did
not impact the total number of sessions attended
regardless of theoretical orientation employed.
We could only identify two studies using a US
sample that focused on session frequency (Reardon,
Cukrowicz, Reeves, & Joiner, 2002; Takuya et al.,
2009). Reardon et al. (2002) found that the longer it
took a client to attend 11 or fewer sessions the
less likely the client was to have a positive outcome.
In contrast, Takuya and colleagues (2009) found a
statistically significant negative relationship
(r � �.04) between session frequency and outcome
scores, but, given the small effect, concluded that the
practical implications were negligible. However,
neither study considered the influence session fre-
quency has on growth rates.
Thus, further delineation of ‘‘dose’’ is needed with
regard to the psychotherapy outcome literature,
which provides an unclear picture on how session
frequency influences the dose-response relationship.
Our exploratory study proposes to extend the GEL
model by including session frequency. We believe
that such inclusion in the GEL model may better
account for patterns of change during treatment than
the traditional aggregate or GEL models. The two
research questions addressed in this study are: (1)
Does the GEL model explain growth trajectories
better than the traditional growth model (i.e.,
aggregate model); and (2) Does the inclusion of
session frequency to the GEL model explain growth
trajectories better than the GEL model?
Method
Participants
The data used in this study were collected from
archived client records from a university counseling
center at a major Southeastern university in the
United States. The data reflect roughly 6 years of
clinical services at the center between 2001 and
2007. The campus enrollment grew from 24,000 to
27,000 undergraduate and graduate students during
the time period of the data sample. The information
gathered was available due to the implementation of
computerized record storage using the Protege
program and the clinical use of the OQ-45 (Lambert
et al., 1996). The data set represents a naturalistic
view of both client outcomes as well as the measure-
ment procedures of those outcomes.
The clinical staff of the counseling center con-
sisted of doctoral-level psychologists and doctoral
practicum students. The center offered a variety of
free services to enrolled students, including under-
graduate, graduate, and professional; however, only
those in individual therapy were administered the
OQ-45. Thus, all of the data in this study are from
clients in individual therapy working on a full range
of mental health concerns (n�3270). The data set
was reduced by excluding those who: attended only
one session (n�837), had a significant gap in treat-
ment (i.e., 90 or more days between two sessions;
n�737), received a second course of treatment
(n�368), did not have a date stamp necessary for
computing length of therapy were removed (in days or
weeks; n�365). Next we removed those who came in
for career or couple counseling (n�62). Finally,
those who had fewer than two administrations of
the OQ-45 (n�59) were removed. The resulting
data set includes 1207 clients who presented with
primary complaints that consisted of depression
(25.3%), anxiety (11.9%), academic issues (14.3%),
interpersonal relationship concerns (19.3%), and other
issues (e.g., stress, obsessive thoughts, alcohol, eating).
The representation of presenting issues appears consis-
tent with other university counseling centers in the
United States (e.g., Minami et al., 2009).
According to available data, 28.3% (n�342) of
clients were male and 69.7% (n�785) female; their
ages at the start of therapy ranged from 17.64 to 63.69
years (Mage�23.72, Mdnage�21.78, SDage�5.78).
The majority of clients were Caucasian (75%;
n�905), followed by African Americans (6.4%,
n�77), international clients (5.5%, n�66), Asian
American (2.2%, n�26), multiracial (1.5%, n�18),
Latino/a American (1.2%, n�15), and Middle
Eastern American (0.2%, n�3). The majority
of clients identified themselves as underg
raduate students (63.7%, n �769) with classifica-
tion as follows: freshmen (15.8%, n�191), sopho-
mores (16.6%, n�200), juniors (13.8%, n�166),
seniors (17.6%, n�212), graduate/professional
(24.3%, n�294), and other (1.1%, n�13). With
regard to enrollment, most clients indicated full time
status (82.1%, n�991), while 3.4% indicated part
time status (n�41). Most clients were single 78.5%
(n�948), followed by married (9.0%, n�5109),
divorced or separated (n�45, 3.7%), and partnered
or other (1.5%, n�18).
The center typically limits sessions to a maximum
of 15 sessions per academic year, but this limit is not
610 R. J. Reese et al.
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
always adhered to based on clinical need. In reality
the number of clients reaching that number of
sessions is a small percentage of the total sample
(i.e., 90.1% of the sample had 15 or fewer sessions).
The median number of sessions attended was
five (M�7.82, Mode�2 [16.9%], SD�7.87,
Skewness�3.77, Kurtosis�21.7), while 43.25% were
seen for more than six sessions. Moreover the median
length in therapy was 8 weeks (M�12.75, Mode�1
[3.6%], SD�16.05, Skewness�3.88, Kurtosis�23.68)
with 49.1% continuing therapy for more than 8 weeks
and 41.5% continuing therapy for more than 10 weeks.
Also, the median number of sessions attended per
week was 0.62 (M�0.67, SD�0.35), which means
each client is attending a session roughly every 10 to
12 days.
The clinical staff (n�49; 65.4% female) were
primarily Caucasian (70.9%), followed by African
American (9.8%), multiracial (18.6%), and Asian
American (0.3%), while 0.3% did not identify their
race/ethnicity. The number of clients per staff ranged
from 1 to 204 with an average of 24.6 (Mdn�8).
Theoretical orientations of staff were diverse, includ-
ing cognitive-behavioral, solution-focused, psycho-
dynamic, interpersonal process, and narrative; most
staff indicated using an integrative/eclectic approach
to therapy drawing upon multiple theoretical orien-
tations.
Measures
All clients receiving individual therapy at the coun-
seling center completed the OQ-45. The OQ-45 is a
45-item self-response inventory that measures sub-
jective distress. Each item is answered on a 5-point
Likert-type response scale ranging from 1 (never) to
5 (almost always) with instructions to respond based
on the client’s experience of the previous week.
Higher scores indicate greater distress. The inven-
tory consists of a total scale score, as well as three
subscales: Symptom Distress, Interpersonal Rela-
tionships and Social Role. The average total scale
score for a clinical population in a counseling center
is 75, while in the normal student population it is 46
(Lambert et al., 1996). The OQ-45 is commonly
used in mental health settings as a brief assessment
tool. It has been shown to generate scores that are
internally consistent (a�.93) and possess short-
term temporal stability (r � .84) with a university
counseling center sample (Lambert et al., 1996).
The measure has demonstrated evidence of concur-
rent validity when compared to other commonly
used brief assessments such as the Symptom Check-
list � 90 � Revised (r �.78 with a counseling center
sample; Umphress, Lambert, Smart, Barlow, &
Clouse, 1997). Internal consistency data could not
be provided for the current sample because client
scores were not recorded and not available at the
individual item level.
Procedure
The OQ-45 was introduced as a standard procedure
of the previously described counseling center several
years before computerized records allowed data
collection. Initially, clients were to be administered
the OQ-45 at intake, again at their first follow-up
session and then at every third session thereafter.
This method of data collection is common to other
counseling centers in the United States and observed
in other research studies (see e.g., Brown, Burlingame,
Lambert, Jones, & Vaccaro, 2001; Spielmans,
Masters, & Lambert, 2006). It was determined by
the center that administering the OQ-45 in the first
session after intake was redundant and that require-
ment was removed part-way through data collection.
Additionally, clinicians could administer the OQ-45
at any session they felt was clinically relevant, though
this was uncommon. As a result, some of the records
in the data set do not follow the proscribed schedule
of OQ-45 administrations. The OQ-45 was adminis-
tered for clients where termination was planned
(n�742). Each clinician was responsible for identi-
fying on the schedule when a particular client was to
be given the OQ-45, with the receptionist staff then
giving the client the measure when they arrived for a
session. Clinicians would then hand-score the mea-
sure and enter the results into a client’s record.
Data Analysis
To evaluate the primary research questions we
applied multilevel growth models (Hox, 2010), also
known as mixed-effects models and hierarchical
linear models (Raudenbush & Bryk, 2002; Singer
& Willett, 2003). Although our data are inherently
nested within therapists at the highest level, initial
analyses indicated that therapist random effects (i.e.,
both intercepts and slopes) were either near zero or
non-estimable for more complex models. In our case
this finding is probably attributed to the fluctuating
number of observations within therapists (i.e., the
number of clients per staff ranged from 1 to 204 with
an average of 24.6 and median of 8). According to
Hox (2010) a solution to this problem is to remove
random effects, which simplifies the model. In
addition, initial analyses comparing the uncondi-
tional (i.e., no predictors or covariates) three-level
growth model (repeated observations within clients
within therapists) to the two-level growth model
(repeated observations within clients) showed no
improvement in fit to the data. As a result we were
Psychotherapy dose-response 611
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
limited to using a two-level growth model to analyze
the nested structure of the data.
The specific models examined in this study con-
sisted of the aggregate model, GEL model, and
modified GEL model using session-by-session as the
metric of time. The aggregate model is the dose-
effect response model or basic two-level uncondi-
tional growth model which estimates the average rate
of change across all clients. The GEL model extends
the aggregate model by specifying that improvement
is a function of the total number of sessions a client
attended. As such, the modified GEL model extends
the GEL model by specifying that improvement is
also conditional on the session frequency (fre-
quency), which can be conceptualized as the average
number of sessions per week:
frequency¼ 1
n� 1
Xn�1
i¼1
Si � Siþ1¼1
n� 1
Xweeks; (1)
where n �total number of sessions attended; Si �Si�1�number of weeks between two consecutive
sessions; weeks � total number of weeks in therapy.
For instance, a client who attended four sessions
(i.e., intake, session 2, session 3, and session 4) in 3
weeks received a session frequency of 1 because the
average amount of time between each session was a
week. To test these models a series of hierarchical
growth models were examined.
The aggregate model is more formally defined as:
Yti ¼ c00þc10ðsessionÞtiþr0iþr1iðsessionÞtiþeti; (2)
where Yti is the OQ-45 score at session t for client i;
g00 is the initial client mean score on the OQ-45
(centered at the start of therapy or the first session);
g10 is the average linear rate of change between
adjacent session points, respectively; r0i, and r1i are
the random client effects or the deviation of client i’s
OQ-45 score from the average initial status and
linear rate of change, respectively; and eti is the
random client effect or client i’s error at time t. Also,
note that the intercept-linear covariance (s01) com-
ponent was estimated.
When we extend the aggregate model to include
the total number of sessions a client attended
(#sessions) we arrive at the GEL model, which is
defined as:
Yti¼ c00þc01ð#sessionsÞiþc10ðsessionÞtiþ c11ðsessionÞtið#sessionsÞiþr0iþr1iðsessionÞtiþeti;
(3)
where the random client (r0i, r1i, and eti) and fixed
effects (g00 and g10) are the same as in the aggregate
model, but because total number of sessions was
centered (see below) these fixed effects and all others
in this model refer to clients with an average of 5.72
attended sessions; g01 is the average difference in
initial client OQ-45 as a function of total number of
sessions attended; g11 is the average difference in
linear growth rates as a function of total number of
sessions a client attended. Note that the natural log
transformation for total number of sessions a client
attended was used because it had a positively skewed
distribution. Then, this variable was grand mean
centered (i.e., ln #sessions �1.7442, which is equal
to 5.72 sessions).
When we extend the GEL model to include the
frequency at which clients attended therapy per week
we arrive at the modified GEL model, which is
defined as:
Yti¼ c00þc01ð#sessionsÞiþc02ðfrequencyÞiþc03ð#sessionsÞiðfrequencyÞiþc10ðsessionÞtiþc11ðsessionÞtið#sessionsÞiþc12ðsessionÞtiðfrequencyÞiþc13ðsessionÞtið#sessionsÞiðfrequencyÞiþ r0i
þr1iðsessionÞtiþeti;(4)
where the random client (r0i, r1i, and eti) and fixed
effects (g00, g10, and g11) of the modified GEL model
are the same as in the GEL model, but because
session frequency was centered (see below) these
fixed effects and all others in this model refer to
clients attending one session per every 11 to 12 days
with an average of 5.72 attended sessions; g02 is the
average difference in initial client OQ-45 as a
function of session frequency; g03 is the average
difference in initial client OQ-45 as a function of
both total number of sessions a client attended and
session frequency; g12 is the average difference in
linear growth rates as a function of session fre-
quency; g13 is the average difference in linear growth
rates as function of both total number of sessions a
client attended and session frequency. Note that the
natural log transformation for session frequency was
used because it had a positively skewed distribution.
Then, this variable was grand mean centered (i.e., ln
frequency�[�.5145]), which is equal to about one
session per every 11 to 12 days).
To compare the improvement in fit between
nested models we used the likelihood ratio test
(LRT) which tests the �2 log likelihood (�2LL)
or deviance statistic between models. Specifically,
the difference in the �2LL values was used to test
the null hypothesis that two nested models do not
have statistically significant different model fits from
each other. The difference in �2LL from two
nested models is Chi-square distributed (Dx2) with
degrees of freedom equal to the difference in
parameters estimated by the two models. We also
(3)
612 R. J. Reese et al.
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
used the Bayesian information criterion (BIC) and
Akaike information criterion (AIC) to examine fit,
where smaller values for both indicate better model
to data fit. All growth models were estimated using
proc mixed in SAS version 9.2 using a full informa-
tion maximum likelihood estimation method, which
utilizes all available data and treats any missing data
as missing at random. Kenward-Roger degrees of
freedom were also estimated. Then, a global
pseudo-R2 effect size statistic for each longitudinal
model was computed by taking the predicted OQ-45
scores and correlating it with observed OQ-45 scores
(see Peugh, 2010). This value was then squared to
get a statistic similar to R2 in traditional multiple
regression, which measures the amount of variation
in OQ-45 scores that can be explained by the model
under consideration. These pseudo-R2 effect size
statistics were estimated in Mplus version 6.1 using
maximum likelihood estimation which also utilizes
all available data and treats any missing data as
missing at random.
Results
Growth Analyses
For our data, a linear, quadratic, and cubic compo-
nents model was fitted first per recommendations by
Nagin (2005) as a sensible starting point. A cubic
polynomial was selected as the starting point for the
aggregate model because the general shape of OQ-45
growth across sessions has been found to be cubic
(Baldwin et al., 2009). A simpler growth model was
interpreted if the predicted trajectories showed little
practical utility. Results for the aggregate model
consisting of linear, quadratic, and cubic components
showed each piece to be statistically significant.
A comparison of this cubic model (BIC �30,816.
6, AIC �30,775.8) to a quadratic model
(BIC �30,840.2, AIC �30,804.6) showed the cubic
model had better fit to the data, Dx2(1) �30,790.6�30,759.8 �30.8, pB.001. Also, a comparison of the
cubic model to a linear model (BIC �30,834.2,
AIC �30,803.6) showed the cubic model had better
fit to the data, Dx2(2) �30,791.6�30,759.8 �31.8,
pB.001. However, a plot of the cubic model and
inspection of the components (intercept �70.6993,
linear �-1.733, quadratic �0.03843, cubic��0.00046) indicated the negatively accelerating
rate of change in predicted OQ-45 scores was mostly
linear with subtle to nearly nonvisible nonlinear
growth rates. Thus, from a practical utility and
parsimonious perspective the data appear to be best
described by a linear pattern. The results for the linear
aggregate model are shown in the first column of
Table I. Fixed effects estimates (top) and variance-covariance estimates (bottom) for models of the prediction of growth in OQ-45 scores
across sessions
Parameter Aggregate model GEL model Modified GEL model
Fixed effects
Mean initial OQ-45 (intercept; g00) 70.34** 71.27** 71.45**
Mean linear growth rate (session; g10) �1.36** �2.40** �2.41**
#Sessionsa (g01) 1.66* 2.10**
#Sessions�session (g11) 1.10** 1.09**
Frequencya (g02) 6.35**
Frequency�#sessions (g03) 3.88*
Frequency�session (g12) �0.80*
Frequency�#sessions�session (g13) 0.30
Random effects
Residual variance (eti) 121.81** 122.31** 122.53**
Intercept variance (r0i) 377.45** 374.61** 368.73**
Linear variance (r1i) 1.40** 0.84** 0.79**
Covariance (intercept, linear; s01) �4.18* �4.18** �3.80*
Pseudo-R2 .005 .035 .0524
Deviance (�2LL) 30,791.6 30,693.4 30,675.3
Dx2(df) 98.2 (2)** 18.1 (4)**
BIC 30,834.2 30,750.1 30,760.5
AIC 30,803.6 30,709.4 30,699.3
Note. #Sessions � total number of sessions a client attended therapy; Frequency � frequency at which clients attended therapy per week;
�2LL � �2*log likelihood.aThe natural log transformation for number of sessions and frequency was used because each variable had a positively skewed distribution.
#Sessions was grand mean centered (i.e., ln #sessions �1.7442, which is equal to 5.72 sessions, while frequency was grand mean centered
(i.e., ln frequency-[�.5145]), which is equal to about one session per every 11 to 12 days). The zero-order correlation between the natural
log of the centered predictors frequency and number of sessions was r��.11, which precludes the results as being an artifact of
multicollinearity between these two variables.
**p B .01; *p B .05.
Psychotherapy dose-response 613
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
Table I and a plot of the aggregate model is provided
in Figure 1.
The results for the GEL model (refer to the second
model in Table I) had a statistically significant better
model to data fit than the traditional aggregate model,
Dx2(2) �30,791.6�30,693.4 �98.2, pB.001. A com-
parison of the BIC and AIC statistics (where smaller is
better; see the bottom two rows of Table I) show the
superiority of the GEL model (BIC �30,750.1,
AIC �30,709.4) to the traditional aggregate model
(BIC �30,834.2, AIC �30,803.6). Also, the pseudo-
R2 improved from .005 (0.5%) to .035 (3.5%). This
means the GEL model explained about 3% more
variability in OQ-45 scores than the traditional
aggregate model. When inspecting the GEL model
results, we find clients who attended a different
number of sessions improved at different linear rates.
Descriptively, those who attended more than 5.72
sessions tended to have less steep negative linear
slopes, while those attending fewer than 5.72 sessions
tended to have steeper negative linear slopes. This
provides evidence that the aggregate model does not
accurately capture many clients’ rate of change during
therapy because it assumes rate of change is the same
regardless of number of sessions attended. However,
the GEL model assumes the frequency at which
clients attend therapy is non-consequential to growth
rates.
As a result, the modified GEL model which
included session frequency was compared to the
GEL model and found to have a statistically signifi-
cant better fit to the data, Dx2(4) �30,693.4�30,675.3 �18.1, pB.001 (refer to third model in
Table I). A comparison of the AIC statistic shows
the superiority of the modified GEL model
(BIC �30,760.5, AIC �30,699.3) to the GEL mod-
el (BIC �30,750.1, AIC �30,709.4), but the BIC
did not. Considering, in combination the AIC, BIC,
and likelihood ratio test (Dx2), the modified GEL
model is the better fitting model. The pseudo-R2
improved from .035 (3.5%) to .0524 (5.24%). This
means the modified GEL model explained 1.74%
more variability in OQ-45 scores than the GEL
model. The modified GEL model shows the linear
rate of change differed for those clients who attended
a different number of sessions and for those differing
in session frequency, but no evidence was found for
linear rates of change varying by the combination of
session frequency and number of sessions attended,
giving evidence that the GEL model is not entirely
complete in reflecting client rate of change in therapy.
A plot of the number of sessions attended by
session and session frequency by session interactions
found in the modified GEL model for select total
number of sessions attended and session frequency is
provided in Figure 1. Figure 1 shows that the rate of
linear change for the average client is a function of
total number of sessions attended or session fre-
quency. Specifically, Figure 1 shows that as clients
attend sessions less frequently (i.e., have larger gaps
in weeks between sessions*one session per two
weeks to one session per week) the amount of
change between adjacent sessions tends to diminish
across sessions, regardless of the number of sessions
attended. Also, the modified GEL model in Figure 1
shows that clients who attended fewer sessions
tended to have larger changes between adjacent
sessions, regardless of session frequency.
It is also important to point out that results from
the modified GEL model show that the relationship
between initial OQ-45 scores and session frequency
is moderated by number of sessions attended. The
positive slope for this interaction (g03�3.88) sug-
gests that as the number of sessions attended
increases, the session frequency by number of
sessions attended slope becomes more positive (or
less negative) in direction. This means that those
clients with higher number of sessions attended did
40
45
50
55
60
65
70
75
80
0 2 4 6 8 10 12 14 16 18 20
Pre
dic
ted
OQ
-45
Sco
re
Sessions
5 Sessions Once Per Wk 10 Sessions Once Per Wk 20 Sessions Once Per Wk5 Sessions Once Per 2 Wks 10 Sessions Once Per 2 Wks 20 Sessions Once Per 2 WksAggregate Model
Figure 1. Modified GEL model trajectories for select number of sessions attended by frequency and trajectory for the aggregate model.
614 R. J. Reese et al.
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
not always start out with higher initial distress.
Specifically, those clients with a higher number of
sessions attended (e.g.,�20) coupled with a lower
session frequency (e.g., one session about every 5
weeks) tended to have lower initial distress levels
than those with fewer attended sessions (e.g.,B3)
and low session frequency. The difference in initial
distress between clients attending many (e.g.,�20)
vs. few (e.g.,B3) sessions was reversed at higher
levels of session frequency. However, the differences
in initial distress levels between high and low number
of sessions attended becomes negligible when session
frequency approached about once every 2�3 weeks.
Interestingly, the correlation between the intercept
and linear variance components in the modified
GEL model was statistically significant, r��.22,
p�.01, indicating that those clients with lower initial
OQ-45 scores tended to have steeper linear slopes,
while clients with higher initial OQ-45 scores tended
not to have as steep linear rates of change across
sessions, controlling for or partialing out all other
predictors in the model. Clients in less distress
tended to improve more rapidly, and clients in
more distress improved more slowly. This finding is
consistent with outcome research that generally finds
individuals in more distress take more time to
improve in treatment, they simply respond more
slowly. A similar correlation was found between
these two components in the GEL and aggregate
models, r��.23, p�.006, and r��.18, p�.01,
respectively.
Given the possibility that missing OQ-45 scores at
the last session could influence estimated parameters
in the three models (aggregate, GEL, and modified
GEL) we statistically addressed this concern by
contrasting completers vs. dropouts (i.e., clients
with or without data at the last session; com-
pleter �1, dropout �0). To examine the longitudi-
nal dropout seen in our study we focused on
Hedeker and Gibbons’ (1997) approach to ran-
dom-effects pattern-mixture modeling by applying
their technique to each of the four longitudinal
models (see also Little, 1995). Specifically, ran-
dom-effects pattern-mixture modeling involved add-
ing the variable dropout to each parameter estimated
in each model which resulted in two, four, and
eight additional parameters being estimated for
the aggregate, GEL, and modified GEL models,
respectively (labeled pattern-mixture).
A comparison of the aggregate model BIC
(30,834.2) and AIC (30,803.6) statistics to the
pattern-mixture aggregate model BIC (30,841.1)
and AIC (30,800.3) statistics showed no improve-
ment in model fit when using the BIC statistic or a
subtle improvement in model fit when using the AIC
statistic with the inclusion of the variable completers
and interaction with completers. Moreover, a like-
lihood ratio test indicated that the pattern-mixture
aggregate model had a statistically significant better fit
to these data than the aggregate model,
Dx2(2) �30,791.6.8�30,784.3 �7.3, p�.03. How-
ever, an inspection of the two additional parameters
included in the pattern-mixture aggregate model
showed both were not statistically significant
(completers � �1.96, p�.11; completers by
session � �0.27, p�.10). Given these findings, we
determined that the aggregate model provided an
accurate representation of the data for completers and
dropouts and that initial status and linear rate of
change did not vary by completer status. A compar-
ison of the pattern-mixture GEL model
(BIC �30,776.7, AIC �30,715.6) to the GEL
model (BIC �30,750.1, AIC �30,709.4) showed
no improvement in fit, Dx2(4) �30,693.
4.3�30,691.6 �1.8, p�.77. Similarly, a comparison
of the pattern-mixture modified GEL model
(BIC �30,811.4, AIC �30,709.5) to the modified
GEL model (BIC �30,760.5, AIC �30,699.3)
showed a reduction in fit, Dx2(8) �30,699.5�30,675.3 �24.2, p�.002. Overall, these results
show that study completion and interactions with
study completion are not statistically significantly
related to OQ-45 scores, over and above the influ-
ences of the predictors included in the original
aggregate, GEL, and modified GEL models.
Discussion
We had two major findings for the current study. The
first major finding is that our study demonstrated
results consistent with the GEL model. The GEL
model better fit the data than the aggregate model,
indicating that the number of sessions attended at
the individual level needs to be considered in out-
come research and that growth in therapy is linear
when this variable is considered and may not con-
form to the negatively accelerating curve as has been
found using an aggregate dose-response model (e.g.,
Howard, Kopta, Krause, & Orlinsky, 1986; Howard,
Moras, Brill, Martinovich, & Lutz, 1996; Kordy
et al., 1988). This appears to be consistent with
Barkham et al.’s (1996) claim that clients who attend
fewer sessions often improve to a ‘‘good enough
level’’ and then discontinue therapy. Clients who
attended fewer sessions responded more quickly and
clients who attended more sessions responded more
slowly. Research with the aggregate model has often
led to the conclusion that treatment sessions yield
diminished returns, that is the ‘‘dose’’ weakens after
a certain number of sessions. Our results suggest that
this was not the case when the data were stratified
by sessions attended or not stratified. Clients in
Psychotherapy dose-response 615
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
our study appeared to have different rates of
improvement based on either number of sessions or
session frequency, but not the interaction of these
two variables. This indicates session frequency is not
the only important component necessary for describ-
ing growth in therapy.
One finding that we did not anticipate was that the
aggregate model best fit a linear rate of change; there
was not a negatively accelerating curve. Rather,
client improvement was found to be similar through-
out the course of therapy when the number of
sessions a client attended was not considered.
Although this result has been found in other studies
(Barkham et al., 1996; Shapiro et al., 2003), most
dose-response studies have reported an effect of
diminished returns as therapy progresses. It is
possible this may have been revealed with more
sessions plotted, but the session limit policy (ap-
proximately 90% attended 15 of fewer sessions)
prevented us from extending the number of sessions.
Although the aggregate model in our study was
linear, it still did not portray the dose-response
relationship as accurately as the GEL model. The
conclusion, however, that the number of sessions a
client attends is the variable that best represents
differences in rate of improvement is premature.
Baldwin et al. (2009) note that it is unknown
whether the number of sessions attended and
resulting rates of improvement are the results of
different client subgroups or populations. If so, there
could be a dose-response but simply specific to
different groups. For example, it is important to
note that in the current study we found a client’s
initial OQ-45 score was related to the interaction
between session frequency and number of sessions
attended. In other words, clients’ initial level of
distress, number of sessions attended, and session
frequency could be one of several variables that
could be related to possible subgroupings if growth
mixture modeling was used.
The second major finding suggests that when we
consider the number of sessions in treatment with
session frequency, the rate of change presents a
different and perhaps more accurate picture. The
modified GEL model provided a better model fit
than the GEL model and showed that having more
frequent sessions (i.e., more sessions in fewer days or
weeks) resulted in steeper growth curves. Specifi-
cally, clients who attended more sessions within a
period of time (e.g., attended eight sessions rather
than four sessions in a span of 8 weeks) had faster
rates of change. Moreover, session frequency re-
sulted in faster rates of improvement independent of
the number of sessions attended (i.e., when the
number of sessions attended was controlled for).
This finding is consistent with studies conducted by
others (Kraft et al., 2006; Reardon et al., 2002) that
were focused on the amount of time (measured in
days or weeks) that it took to complete a certain
number of sessions. Session frequency appears to
matter*all therapy sessions are not equal. The
results suggest clients who had shorter intervals
between therapy sessions improved more quickly.
Although the major findings are tentative
and warrant replication, they do encourage further
attention be given to how we conceptualize and
operationalize the term ‘‘dose.’’ Both the aggregate
dose-response model (e.g., Hansen & Lambert,
2003; Lutz et al., 1999) and the GEL model
(Baldwin et al., 2009; Barkham et al., 1996) strongly
suggest that sessions are not equivalent in nature.
The GEL model, however, illustrates more clearly
that the effectiveness of any given session may vary
greatly across individuals and a host of other con-
textual factors. Our findings indicate that considera-
tion should be given to the of number of days/weeks
in treatment, or session frequency, and the notion
that sessions can be assumed to be a metric
composed of equal units and equal intervals.
The psychotherapy outcome literature, as can be
observed in its nomenclature, has tethered itself to
the medical/pharmacological paradigm. The results
of our study suggest that more thought be given to
the underlying metric(s) by which we measure
treatment progress. We offer two possibilities. First,
as consistent with a pharmacological model, time
between doses (sessions) needs to be considered
more fully as is done when considering dosage and
efficacy of medications. Prescriptions for medication
provide intervals along with dosage information in
order to maximize effectiveness and to minimize
potential adverse effects. Second, the singular term
of ‘‘dose’’ should be reconsidered. Borrowing from
the logic that reporting null hypothesis results as
‘‘significant’’ rather than the more precise ‘‘statisti-
cally significant’’ is semantically sloppy and leads to
faulty inferences (Thompson, 1996), perhaps ex-
tending the term ‘‘dose’’ to ‘‘weekly session’’ or
simply reporting the mean interval across all sessions
would provide a more precise metric for under-
standing the relationship of the number of sessions
attended in relation to an actual time metric, be it
days or weeks. For example, it may be useful to
report the median and average number of sessions
attended per week. The median number of sessions
attended per week in our study was 0.62 (M�0.67,
SD�0.35), indicating that our data set better
reflected a week-and-a-half between adjacent ses-
sions.
It would be nice to be able to compare our results
with other dose-response studies, particularly studies
that use large data sets from university counseling
616 R. J. Reese et al.
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
centers that consider time in treatment from both
session-to-session and session frequency perspec-
tives. Inferences about the relationship between
sessions attended and session frequency could be
made. Considering the time metric more fully would
also allow researchers to be more confident about the
similarity of the data sets. As a result, assumptions
about session intervals may be inaccurate. For
example, Kraft et al. (2006) found that even for
clients who attended psychoanalytic therapy, where
more than one session per week is the norm, they
attended less than one session per week on average,
which is consistent with what we found from our
sample.
Future Research
Research that further considers the GEL model and
session frequency to evaluate trajectories of change
needs to be conducted. Studies that utilize random
assignment to control for the time between sessions,
time in treatment, and the number of sessions
attended would be ideal. For example, clients could
be randomly assigned to attend therapy every week or
every other week for a specified number of sessions.
Naturalistic studies would also be valuable provided
that studies are structured so variables of importance
could be tracked carefully, including the amount of
time between sessions and reasons for interruptions in
treatment. Such studies would better our under-
standing of how session intervals relate to outcome.
Future research should also address how distress
severity moderates the relationship between outcome
and session intervals. Common sense would seem to
follow that more frequent treatment is necessary when
a client is in more distress (Beutler et al., 2004), but
such a stance does not have a lot of empirical support.
More research should also focus on the trajectory
of change that both replicates and extends recent
work that has challenged the assumption that the
negatively accelerating curve is the most accurate
picture to describe how change occurs in therapy.
Baldwin et al. (2009) and Barkham et al. (2006) have
both provided evidence that many clients do not
necessarily respond to treatment in this manner.
Hayes, Laurenceau, Feldman, Strauss, & Cardaciotto
(2007) take this one step further in their summary by
providing evidence that change does not always
occur in a linear fashion when diagnosis and
theoretical approach are considered. Additionally, it
may be important to consider other treatment out-
comes than just client distress given trajectories of
change may differ for alternate outcomes. Collecting
data across treatment, rather than in a pre-post
fashion, coupled with utilizing multilevel growth
modeling techniques will allow for a more sophisti-
cated understanding of how clients respond to
psychotherapy treatment.
Limitations of Research
Our study has several limitations worth noting. First
are the limitations that are inherent with naturalistic
research. We were not able to account or control for
the variety of treatment interventions provided or the
number of sessions clients attended. Second, using
an archived data set precluded us from being able to
identify the reasons why clients discontinued therapy
or why sessions were missed*were session intervals
intentionally planned or unplanned (no-shows and
cancellations)? A third limitation, and our biggest
concern with the data set, was that the OQ-45 was
not given every session and many clients did not have
OQ-45 data at their last session. Data collected in
this manner may have underestimated treatment
effects. Ultimately, we do not believe this is an
egregious limitation of our study for three reasons.
First, we used multilevel modeling which is designed
to address longitudinal, repeated measure designs
that have missing data. Second, the data collection
format used in this study is not uncommon at
university counseling centers and agencies in the
USA that use outcome assessment (e.g., Brown
et al., 2001; Spielmans et al., 2006). This is certainly
a limitation, but it may also be a strength in terms of
reflecting data collection procedures in real-world
treatment settings where completing measures every
session is not feasible. Third, we compared those
who had OQ-45 data at termination to those without
it to evaluate whether there were differences in any of
the models’ parameter estimates. In general, data for
clients with and without OQ-45 data at termination
were statistically equivalent given support for the
data collection process at the last session.
We also attempted to temper the effects of missing
data by eliminating clients who did not have at least
two observed data points (two OQ-45 measures);
however, this led to a fourth limitation*a smaller
sample size which possibly limits generalizability.
The concerns of sample size and missing data are
lessened by the consistency of our results with other
studies that evaluated the GEL model (Baldwin
et al., 2009; Barkham et al., 2006). Specifically, the
GEL model from our study parallel the Baldwin
et al. (2009) study where data were collected every
session and show evidence of the linear trajectories
per number of sessions attended as predicted by
Barkham et al. (1996). Generalizability is also a
concern because the treatment duration for clients
was not lengthy (approximately 40% had between
five and 15 sessions) and the treatment setting was a
US university counseling center and was composed
Psychotherapy dose-response 617
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
of primarily young, White clients. The duration of
treatment, however, is consistent with other univer-
sity counseling centers, other outpatient treatment
agencies in the United States, and other studies
(Baldwin et al., 2009). A fifth limitation is that we
only used one self-report measure to evaluate treat-
ment outcome. There is evidence, however, that the
OQ-45 has found results consistent with other out-
come measures (Baldwin et al., 2009).
Clinical Implications and Conclusions
Session limits are common to practitioners and
agencies due to resource limitations and external
conditions such as managed care. Our study ad-
dresses how session limits are derived and imple-
mented. First, the replication of the GEL model
provides further evidence that reliance on an aggre-
gate model may not be the most appropriate way to
evaluate how individuals respond to treatment. This
is because an aggregate model tends to generally
underestimate trajectories (i.e., the slopes appear less
steep) for clients with fewer sessions and overesti-
mate trajectories (i.e., the slopes appear steeper) for
clients attending more sessions (see Figure 1).
Deriving session limits based on the aggregate model
may not be beneficial given the differential response
to treatment as indicated by the GEL model. One
size does not fit all. Second, the finding that session
frequency is related to the rate of change suggests
that session frequency should be carefully considered
by therapists and clients. The temptation may be to
spread sessions out (e.g., every other week) to
‘‘ration’’ treatment and increase the length of time
in treatment. Future research is certainly needed,
but our modified GEL models indicate that clients
who attended sessions roughly once per week
incurred more change in fewer sessions compared
to those who had less frequent sessions.
Our findings suggest that clients respond and use
therapy differently, reinforcing the importance of
tracking outcome in treatment to understand how
clients are progressing. The benefits of tracking
client outcome (patient-focused research or client
feedback) across treatment are well-documented
(e.g., Anker, Duncan, & Sparks, 2009; Whipple
et al., 2003), having been found to help clinicians
identify clients not progressing as expected and
reduce premature terminations. Using client feedback
has also been found to generate better therapy out-
comes when compared to treatment that does not
utilize client feedback (e.g., Reese, Norsworthy, &
Rowlands, 2009; Shimokawa, Lambet, & Smart,
2010).
In summary, the results of our exploratory study
offer evidence for the GEL model*how long clients
stayed in treatment was dependent on how quickly
they improved. Those who attended more sessions
responded more slowly and those who attended
fewer sessions improved faster. Additionally, we
found that the frequency of session attendance
influenced the rate of change*more sessions in
fewer days/weeks predicted faster improvement.
Inclusion of session frequency improved the predic-
tion of the GEL model. Considering session fre-
quency may provide researchers and clinicians a
more accurate depiction of the rate of change across
treatment. Further research is certainly warranted.
The availability of statistical analyses such as multi-
level modeling and growth mixture models coupled
with the practice of tracking outcomes with clients
across treatment offers the opportunity to further
explore and evaluate the myriad of client, therapist,
and treatment variables that influence psychotherapy
treatment outcome in a more sophisticated manner.
References
Anderson, E.M., & Lambert, M.J. (2001). A survival analysis of
clinically significant change in psychotherapy. Journal of Clinical
Psychology, 57, 875�888. doi:10.1002/jclp.1056.
Anker, M.G., Duncan, B.L., & Sparks, J.A. (2009). Using client
feedback to improve couple therapy outcomes: A randomized
clinical trial in a naturalistic setting. Journal of Consulting and
Clinical Psychology, 77, 693�704. doi:10.1037/a0016062.
Baldwin, S.A., Berkeljon, A., Atkins, D.C., Olsen, J.A., & Nielsen,
S.L. (2009). Rates of change in naturalistic psychotherapy:
Contrasting dose-effect and good-enough level models of
change. Journal of Consulting and Clinical Psychology, 77, 203�211. doi:10.1037/a0015235.
Barkham, M., Connell, J., Stiles, W., Miles, J.N., Margison, F.,
Evans, C., & Mellor-Clark, J. (2006). Dose-effect relations and
responsive regulation of treatment duration: The good enough
level. Journal of Consulting and Clinical Psychology, 74, 160�167.
doi:10.1037/0022.006X.1.160.
Barkham, M., Rees, A., Stiles, W.B., Shapiro, D.A., Hardy, G.E.,
& Reynolds, S. (1996). Dose�effect relations in time-limited
psychotherapy for depression. Journal of Consulting and Clinical
Psychology, 64, 927�935. doi:10.1037/0022-006X.64.5.927.
Beutler, L.E., Malik, M., Alimohamed, S., Harwood, T.M.,
Talebi, H., Noble, S., & Wong, E. (2004). Therapist variables.
In M.J. Lambert (Ed.), Bergin and Garfield’s handbook of
psychotherapy and behavior change (pp. 227�306). NewYork,
NY: Wiley.
Brown, G.S., Burlingame, G.M., Lambert, M.J., Jones, E., &
Vaccaro, J. (2001). Pushing the quality envelope: A new
outcomes management system. Psychiatric Services, 52, 925�934. doi:10.1176/appi.ps.52.7.925.
Freedman, N., Hoffenberg, J.D., Vorus, N., & Frosch, A. (1999).
The effectiveness of psychoanalytic psychotherapy: The role of
treatment duration, frequency of sessions, and the therapeutic
relationship. Journal of the American Psychoanalytic Association,
47, 741�772. doi:10.1177/00030651990470031001.
Hansen, N.B., & Lambert, M.J. (2003). An evaluation of the
dose-response relationship in naturalistic treatment settings
using survival analysis. Mental Health Services Research, 5,
1�12. doi:10.1023/A:1021751307358.
Hayes, A.M., Laurenceau, J., Feldman, G., Strauss, J.L., &
Cardaciotto, L. (2007). Change is not always linear: The study
618 R. J. Reese et al.
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014
of nonlinear and discontinuous patterns of change in psychother-
apy. Clinical Psychology Review, 27, 715�723. doi:10.1016/
j.cpr.2007.01.008.
Hedeker, D., & Gibbons, R.D. (1997). Application of random-
effects pattern-mixture models for missing data in longitudinal
studies. Psychological Methods, 2, 64�78. doi:10.1037/1082-
989X.2.1.64.
Howard, K.I., Kopta, S., Krause, M., & Orlinsky, D. (1986). The
dose-effect relationship in psychotherapy. American Psychologist,
41, 159�164. doi:10.1037/0003-066X.41.2.159.
Howard, K.I., Moras, K., Brill, P.L., Martinovich, Z., & Lutz, W.
(1996). Evaluation of psychotherapy: Efficacy, effectiveness,
and patient progress. American Psychologist, 51, 1059�1064.
doi:10.1037/0003-066X.51.10.1059.
Hox, J. (2010). Multilevel analysis: Techniques and applications (2nd
ed.). Mahwah, NJ: Lawrence Erlbaum.
Kordy, H., Rad, M.V., & Senf, W. (1988). Time and its relevance
for a successful psychotherapy. Psychotherapy and Psychosomatics,
49, 212�222. Retrieved from http://content.karger.com/
ProdukteDB/produkte.asp?Aktion�JournalHome&Produkt
Nr�223864.
Kraft, S., Puschner, B., & Kordy, H. (2006). Treatment intensity
and regularity in early outpatient psychotherapy and its relation
to outcome. Clinical Psychology and Psychotherapy, 13, 397�404.
doi:10.1002/cpp.505.
Lambert, M.J., Hansen, N.B., Umphress, V., Lunnen, K., Okiishi,
J., Burlingame, G., . . . Reisinger, C. (1996). Administration and
scoring manual for the OQ 45.2. Stevenson, MD: American
Professional Credentialing Services.
Little, R.J.A. (1995). Modeling the drop-out mechanism in
repeated-measures studies. Journal of the American Statistical
Association, 90, 1112�1121. doi:10.2307/2291350.
Lutz, W., Lowry, J., Kopta, S.M., Einstein, D.A., & Howard, K.I.
(2001). Prediction of dose-response relations on patient
characteristics. Journal of Clinical Psychology, 57, 889�900.
doi:10.1002/jclp.1057.
Lutz, W., Martinovich, Z., & Howard, K.I. (1999). Patient
profiling: An application of random coefficient regression
models to depicting the response of a patient to outpatient
psychotherapy. Journal of Consulting and Clinical Psychology, 67,
571�577. doi:10.1037/0022-006X.67.4.571.
Minami, T., Davies, D.R., Tierney, S.C., Bettman, J.E., McA-
ward, S.M., Averill, L.A., . . . Wampold, B.E. (2009). Prelimin-
ary evidence on the effectiveness of psychological treatments
delivered at a university counseling center. Journal of Counseling
Psychology, 56, 309�320. doi: 10.1037/a0015398
Nagin, D. (2005). Group-based modeling of development. Cam-
bridge, MA: Harvard University Press.
Okiishi, J.C., Lambert, M.J., Eggert, D., Nielsen, L., Dayton,
D.D., & Vermeersch, D.A. (2006). An analysis of therapist
treatment effects: Toward providing feedback to individual
therapists on their clients’ psychotherapy outcome. Journal of
Clinical Psychology, 62, 1157�1172. doi:10.1002/jclp.20272.
Peugh, J.L. (2010). A practical guide to multilevel modeling.
Journal of School Psychology, 48, 85�112. doi:10.1016/
j.jsp.2009.09.002.
Raudenbush, S.W., & Bryk, A.S. (2002). Hierarchical linear
models: Applications and data analysis methods (6th ed.). New-
bury Park, CA: Sage.
Reardon, M.L., Cukrowicz, K.C., Reeves, M.D., & Joiner, T.E.J.
(2002). Duration and regularity of therapy attendance as
predictors of treatment outcome in an adult outpatient popula-
tion. Psychotherapy Research, 12, 273�285. doi:10.1093/ptr/
12.3.273.
Reese, R.J., Norsworthy, L.A., & Rowlands, S.R. (2009). Does a
continuous feedback system improve psychotherapy outcome?
Psychotherapy: Theory, Research, Practice, Training, 46, 418�431.
doi:10.1037/a0017901.
Sandell, R., Blomberg, J., & Lazar, A. (2002). Time matters: On
the temporal interactions in long-term follow-up of long-term
psychotherapies. Psychotherapy Research, 12, 39�59. doi:10.1093/
ptr/12.1.39.
Singer, J.D., & Willett, J.B. (2003). Applied longitudinal data
analysis: Modeling change and event occurrence. New York, NY:
Oxford University Press.
Shapiro, D.A., Barkham, M., Stiles, W.B., Hardy, G.E., Reynolds,
S., & Startup, M. (2003). Time is of the essence: A selective
review of the fall and rise of brief therapy research. Psychology
and Psychotherapy: Theory, Research and Practice, 76, 211�235.
doi:10.1348/147608303322362460.
Shimokawa, K., Lambert, M.J., & Smart, D.W. (2010). Enhan-
cing treatment outcome of patients at risk of treatment failure:
Meta-analytic and mega-analytic review of a psychotherapy
quality assurance system. Journal of Consulting and Clinical
Psychology, 78, 298�311. doi:10.1037/a0019247.
Spielmans, G.I., Masters, K.S., & Lambert, M.J. (2006).
A comparison of rational versus empirical methods in the
prediction of psychotherapy outcome. Clinical Psychology and
Psychotherapy, 13, 202�214. doi:10.1002/cpp.491.
Takuya, M., Davies, R.D., Tierney, S.C., Bettman, J.E., Mc-
Award, S.M., Averill, L.A., . . . Wampold, B.E. (2009).
Preliminary evidence of the effectiveness of psychological
treatments delivered at a university counseling center. Journal
of Counseling Psychology, 56, 309�320. doi:10.1037/a0015398.
Thompson, B. (1996). AERA editorial policies regarding statis-
tical significance testing: Three suggested reforms. Educational
Researcher, 25, 26�30. doi:10.3102/0013189X025002026.
Umphress, V.J., Lambert, M.J., Smart, D.W., Barlow, S.H., &
Clouse, G. (1997). Concurrent and construct validity of the
outcome questionnaire. Journal of Psychoeducational Assessment,
15, 40�55. doi:10.1177/073428299701500104.
Whipple, J.L., Lambert, M.J., Vermeersch, D.A., Smart, D.W.,
Nielsen, S.L., & Hawkins, E.J. (2003). Improving the effects of
psychotherapy: The use of early identification of treatment
failure and problem-solving strategies in routine practice.
Journal of Counseling Psychology, 50, 59�68. doi:10.1037/
0022-0167.50.1.59.
Psychotherapy dose-response 619
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a, S
an F
ranc
isco
] at
01:
59 1
6 D
ecem
ber
2014