at SciVerse ScienceDirect

Behaviour Research and Therapy 50 (2012) 544e550

Contents lists available

Behaviour Research and Therapy

journal homepage: www.elsevier .com/locate/brat

Therapeutic alliance in guided internet-delivered cognitive behavioural treatmentof depression, generalized anxiety disorder and social anxiety disorder

Gerhard Andersson a,b,*, Björn Paxling a, Maria Wiwe a, Kristofer Vernmark c, Christina Bertholds Felix d,Lisa Lundborg d, Tomas Furmark d, Pim Cuijpers e, Per Carlbring f

aDepartment of Behavioural Sciences and Learning, Swedish Institute for Disability Research, Linköping University, SwedenbDepartment of Clinical Neuroscience, Psychiatry Section, Karolinska Institutet, Stockholm, Swedenc Psykologpartners, Private Practice, Linköping, SwedendDepartment of Psychology, Uppsala University, Uppsala, SwedeneDepartment of Clinical Psychology, VU University Amsterdam, and EMGO Institute, The NetherlandsfDepartment of Psychology, Umeå University, Umeå, Sweden

a r t i c l e i n f o

Article history:Received 5 January 2012Received in revised form8 May 2012Accepted 10 May 2012

Keywords:Internet-delivered cognitive behaviourtherapyTherapeutic allianceMajor depressionGeneralized anxiety disorderSocial anxiety disorder

* Corresponding author. Department of BehavioLinköping University, SE-581 83 Linköping, Sweden. Te13 28 21 45.

E-mail address: Gerhard.Andersson@liu.se (G. And

0005-7967/$ e see front matter � 2012 Elsevier Ltd.doi:10.1016/j.brat.2012.05.003

a b s t r a c t

Guided internet-delivered cognitive behaviour therapy (ICBT) has been found to be effective in severalcontrolled trials, but the mechanisms of change are largely unknown. Therapeutic alliance is a factor thathas been studied in many psychotherapy trials, but the role of therapeutic alliance in ICBT is less wellknown. The present study investigated early alliance ratings in three separate samples. Participants fromone sample of depressed individuals (N ¼ 49), one sample of individuals with generalized anxietydisorder (N ¼ 35), and one sample with social anxiety disorder (N ¼ 90) completed the Working AllianceInventory (WAI) modified for ICBT early in the treatment (weeks 3e4) when they took part in guidedICBT for their conditions. Results showed that alliance ratings were high in all three samples and that theWAI including the subscales of Task, Goal and Bond had high internal consistencies. Overall, correlationsbetween the WAI and residualized change scores on the primary outcome measures were small and notstatistically significant. We conclude that even if alliance ratings are in line with face-to-face studies,therapeutic alliance as measured by the WAI is probably less important in ICBT than in regular face-to-face psychotherapy.

� 2012 Elsevier Ltd. All rights reserved.

Introduction

The concept of therapeutic alliance, also known as the workingalliance, is often regarded as an important ingredient in psycho-therapy across different psychotherapy orientations (Lambert &Barley, 2002), but has been studied to a lesser extent in alterna-tive treatment formats such as group therapy and guided self-help.Bordin (1979) introduced a model for understanding therapeuticalliance in which he made distinctions between task, goal andbond, which together form the concept of alliance (Bordin, 1979).

During the last 15 years there has been a rapid development ofnew ways to deliver cognitive behaviour therapy (CBT), and ofthese new approaches the internet has probably been the format

ural Sciences and Learning,l.: þ46 13 28 58 40; fax: þ46

ersson).

All rights reserved.

with themost studies (Andersson, 2009). Many studies on internet-delivered CBT (ICBT) have involved guidance which tends to boostthe effect (Andersson & Cuijpers, 2009; Spek, Cuijpers et al., 2007),usually to the extent that guided ICBT and face-to-face therapyyield equivalent outcomes (Bergström et al., 2010; Hedman et al.,2011). Equal outcomes have been found in studies on panicdisorder (Bergström et al., 2010; Carlbring et al., 2005; Kiropouloset al., 2008), social anxiety disorder (Andrews, Davies, & Titov,2011; Hedman et al., 2011), and subclinical depression (Spek,Nyklicek et al., 2007), but also in studies on conditions liketinnitus (Kaldo et al., 2008).

The role of the therapeutic alliance in guided ICBT is not obviousas there is much less therapist contact than in face-to-face treat-ments (approximately 1/10 of the time), and that the client may noteven see the therapist in person. Still there is a therapeutic inter-action as the therapist responds to messages sent from the clientand uses both specific and common factors to encourage the clientto work with the ICBT (Paxling et al., in press). The development ofa therapeutic alliance may not necessarily require direct face-to-

G. Andersson et al. / Behaviour Research and Therapy 50 (2012) 544e550 545

face contact with a therapist. It is possible that alliance is at leastpartly based on client expectations regarding both tasks and goalsthat may exist before the therapy starts. While the bond betweenthe therapist and the client probably is different in ICBT than inface-to-face therapy questions regarding task, goal and bond arestill relevant. Therefore we believe that a therapeutic alliance canbe formed over the internet as the therapist in ICBT providesindividualized encouragement on progress and also responds todifficulties the client may confront when working with self-helpmaterial. Moreover, self-help texts can include aspects that helpform a therapeutic alliance and a client may perceive an under-standing clinician who is behind the text material (Richardson,Richards, & Barkham, 2010). In other words the total informationavailable for clients when rating the alliance can consist of inter-actions with a therapist online, interactions with a treatmentsystem, and text material that can possibly boost the alliance.

There has been some previous research on the role of alliance ininternet treatments. Cook and Doyle (2002) studied allianceratings in a small sample (N ¼ 15) and compared the ratings withthose from a previous study sample who had received face-to-facetherapy. Contrary to expectations the client rated alliance washigher in the online therapy group (Cook & Doyle, 2002). Thisfinding was replicated in a study on e-mail counselling in whichthe authors found high ratings of alliance (D’Arcy, Reynolds, Stiles,& Grohol, 2006), that were within the range of what has beenfound in face-to-face studies. A limitation of both studies is thelack of a control group as they used previously reported results forthe face-to-face comparisons. In a study on posttraumatic stressdisorder (PTSD) Knaevelsrud and Maercker (2006) found highratings of alliance, even if alliance was not a strong predictor ofoutcome (Knaevelsrud & Maercker, 2006). However in a subse-quent report, the same authors found that alliance scores increasedduring treatment and that ratings of alliance at the end of treat-ment correlated with treatment outcome (N ¼ 41) (Knaevelsrud &Maercker, 2007). Klein, Austin, et al. (2009) and Klein, Mitchellet al. (2009) studied the role of frequency of therapist e-mailcontact in a trial on panic disorder. They found no differences inclient alliance ratings between the intensive contact condition(average therapist time 308 min) versus the infrequent contactcondition (average therapist time 205 min) (Klein, Austin et al.,2009). The authors concluded that the time spent with the ther-apist may not be a key variable when rating alliance. Overall,ratings of alliance were high, which the same researchers alsofound in another study on posttraumatic stress disorder (Klein,Mitchell, et al., 2009). Correlations with outcome were notreported.

In light of the previous findings which indicate that alliance canbe formed in ICBT we wanted to investigate alliance ratings byclients who were research participants in three different ICBTprograms. The first was a study on ICBT for depression (Vernmarket al., 2010) in which internet-delivered guided self-help wascompared with e-mail therapy. Ratings of therapeutic alliance werecollected from the participants between week 3 and 4 of thetreatment. The second data set was from a controlled trial on ICBTfor generalized anxiety disorder (GAD), in which guided ICBT wastested against a waiting-list control group (Paxling et al., 2011).Alliance ratings were collected during the third treatment week.The third data set was derived from a study on social anxietydisorder (SAD), in which ICBT was tested against a waiting-listcontrol group who participated in an online discussion forum(Andersson, Carlbring, Furmark, & on behalf of the SOFIE ResearchGroup, 2012). Alliance ratings were collected at the beginning ofthe fourth treatment week. We expected high ratings of allianceand investigated if ratings of alliance would be predictive oftreatment outcome.

Sample I - depression

Background

Data were collected in association with a controlled study onICBT for major depression (Vernmark et al., 2010). All participantswere interviewed live, but self-report instruments including alli-ance ratings were collected via the internet. Treatment wasprovided on the internet and each participant had an online ther-apist who followed them for the full 8 week treatment period.

Method

ProcedureA total of 88 persons with a confirmed diagnosis of major

depression (American Psychiatric Association, 2000) were includedin the trial following recruitment via advertisement (Vernmarket al., 2010). More details regarding procedure and treatmentoutcome are provided in the original report. Herewe report data forthe treated participants who were asked to send in alliance ratingsfollowing the third treatment week. By then participants had beeninterviewed in a live structured interview (not conducted by thetherapist), had started their treatment, and had been in contactwith their online therapist at least three times when sending inhomework assignments. The treatment provided was either e-mailtherapy or guided self-help. Briefly, the e-mail therapy was tailoredand did not use any prepared self-help texts (Vernmark et al., 2010).All e-mails were individually written for the unique client. The totalaverage time spent by each therapist on the participant in the e-mail therapy was 509 min (SD ¼ 176). Each therapist (N ¼ 6) wasidentified with name and a picture on the study web page. Theguided internet-based self-help consisted of text chapters dealingwith CBT components such as behavioural activation and cognitiverestructuring, and had been developed in a previous study(Andersson et al., 2005). Each therapist spent an average of 53 minper participant (SD ¼ 28) for the whole self-help treatment. Home-work assignments were given two both groups.

ParticipantsA total of 59 individuals were randomly allocated to either one

of two treatments and the remaining 29 were on a waiting-listcontrol group. Overall, dropout rate was low with 14% notattending the posttreatment interview. A total of 49 participants inthe treatment groups completed the alliance measure, with five ineach of the two treatment groups not responding to the ques-tionnaire sent out by the study coordinator. There were 25 partic-ipants in the e-mail group and 24 in the guided self-help group.Mean age for the 49 participants was 38.9 years (SD ¼ 13.5), and75% were women. There were no differences between the twotreatment groups in terms of age and gender.

MeasuresSeveral symptom-related measures were included in the trial,

but here we focus on the 21-item Beck Depression Inventory (BDI)(Beck, Ward, Mendelson, Mock, & Erbaugh, 1961), which hasestablished psychometric properties for online use (Carlbring,Brunt et al., 2007). This was the primary outcome measure in thetrial (Vernmark et al., 2010). In the analyses we calculated residualgain scores, which handle measurement error of repeated admin-istration of the instruments and the initial differences betweenindividuals at pretreatment (Steketee & Chambless, 1992). Theresidual gain scores were calculated by the formula z2 e (z1 * r1 2)(Steketee & Chambless, 1992), where z2 is the Z-transformedposttreatment score and z1 the transformed pretreatment score,and r1 2 the Pearson correlation between pre- and post

G. Andersson et al. / Behaviour Research and Therapy 50 (2012) 544e550546

assessments. Residual gain scores were reversed so that higherscores would indicate greater improvement.

In order to measure therapeutic alliance we used an adaptedversion of theWorking Alliance Inventory (WAI), which contains 36items (Horvath & Greenberg,1989). Each item in theWAI is rated ona 7 point scale, with higher scores indicating higher alliance. Theadaptation consisted of minor reframing of items to yield internettreatment instead of face-to-face treatment. The WAI is scored intothree subscales measuring Task, Goal and Bond. The Task and Goalscales are intended to measure agreement between the client andtherapist with regard to Tasks and Goals of the treatment. The Bondsubscale aims to measure the empathic bond between the clientand the therapist (Horvath & Greenberg, 1989). We only measuredalliance from the participant point of view and not therapistratings. A total score of the WAI was also calculated. Psychometricproperties for the online version used here was a ¼ .81 for the Tasksubscale, a ¼ .84 for the Goal subscale, and a ¼ .90 for the Bondsubscale. For the total score the internal consistency was a ¼ .91. Inaddition, there were high intercorrelations between the total scoreand the subscales (r ¼ .67 to r ¼ .92).

Results

Both groups improved on the BDI with within group effect sizesbeing d ¼ 2.18 and d ¼ 1.39 for the e-mail therapy and guided self-help groups respectively. First, we checked if the non-completers ofthe WAI differed from the completers. Non-completers did notdiffer with regard to age, pretreatment BDI, or residual changescores on the BDI. Mean alliance ratings on the WAI includingcorrelations with residualized change scores on the BDI are pre-sented in Table 1. There were no between group differences withthe exception of the WAI subscale Goal for which the e-mailtherapy group scored significantly higher, t(47) ¼ 2.17, p ¼ .035.None of the correlations were statistically significant. The WAI wasfinally correlated with age and pretreatment BDI scores. Whilethere was a significant correlation betweenWAI total score and age(r ¼ .33, p ¼ .02), there was no significant correlation withpretreatment BDI scores (r ¼ �.21). Results presented in Table 1were unchanged when adding age as a covariate.

Sample II e generalized anxiety disorder

Background

Data in sample II were collected during a controlled study onICBT for GAD (Paxling et al., 2011). Participants were screenedonline and thereafter interviewed over the telephone. As in sampleI, self-report instruments and alliance ratings were collected via theinternet. Treatment was provided via the internet and eachparticipant had an online therapist who guided them through the 8week treatment period.

Method

ProcedureFollowing screening and interview we included 89 persons

diagnosed with GAD (American Psychiatric Association, 2000). The89 participants were randomized to either a control (N ¼ 45) ora treatment (N ¼ 44). Post treatment measures were obtained from38 out of 44 randomized participants in the treatment group (86%).More details regarding procedures and treatment outcomes areprovided in Paxling et al. (2011). Alliance ratings using the shortversion of theWAI (Busseri & Tyler, 2003; Tracey & Kokotovic, 1989)were completed during the third treatment week. As in sample Iparticipants had been in contact with the research staff and their

therapist at least three times and had begun to send in theirhomework assignments. The online therapists (N ¼ 3) were iden-tified with name and a picture on the study web page (as in sampleI). The treatment in theGAD studywas guided self-helpwithweeklyfeedback sent to the participants, but also occasional answers toquestions that could arise during the treatment. The treatment wastext-based and contained modules on psychoeducation, appliedrelaxation, worry time, cognitive restructuring, problem solving,worry, interpersonal problem solving, sleep management andfinally relapse prevention (Paxling et al., 2011). The therapist timedevoted to each client ranged from 10 to 15 min per week, with anaverage total time of M ¼ 97 min (SD ¼ 52) per participant.

ParticipantsA total of 37 participants in the treatment group completed the

alliance measure. Two participants did not complete the post-treatment measures and were hence not included yielding 35participants out of 44 (79.5%) completing both alliance andoutcome measures.

Mean age for the 35 participants was 40.0 years (SD ¼ 11.2), and80.6% were women.

MeasuresThe primary outcome measure in the GAD trial was the Penn

State Worry Questionnaire (PSWQ) (Meyer, Miller, Metzger, &Borkovec, 1990). The PSWQ has 16 items and is designed tocapture the generality, excessiveness, and uncontrollability ofpathological worry. It has good psychometric properties (Meyeret al., 1990), and has been used in previous ICBT studies on GAD(Titov et al., 2009). Separate psychometric properties for online usehave not been reported.

As a measure of therapeutic alliance we used the short versionof theWAI (Andrusyna, Tang, DeRubeis, & Luborsky, 2001; Tracey &Kokotovic, 1989), which contains 12 items. This version has beenfound to be exchangeable with the longer 36-item version (Busseri& Tyler, 2003) and has similar psychometric properties. The shortWAI can also be divided into the three subscales of Task, Goal andBond, with 4 items for each subscale. Psychometric properties forthe online version of the short WAI was a ¼ .84 for the Tasksubscale, a ¼ .69 for the Goal subscale, and a ¼ .83 for the Bondsubscale. For the total score the internal consistency was a¼ .91. Aswith the longer version of the WAI used in sample I, there werehigh intercorrelations between the total score and the subscales(r ¼ .69 to r ¼ .93).

Results

As reported in the trial the treatment group improved, and forthe group who completed the short WAI the within group effectsize on the PSWQwas d¼ 1.17.We checked if the non-completers ofthe WAI differed from the ones who completed the WAI and therewere no significant differences with regards to age, pretreatmentPSWQ or residual change scores on the PSWQ (unpaired t-tests).Mean alliance ratings on the short WAI including correlations withresidualized change scores on the BDI are presented in Table 2.Correlations were again not statistically significant. The WAI wasfinally correlated with age and pretreatment PSWQ scores and age.We found no significant correlations.

Sample III e social anxiety disorder

Background

Data for sample III were collected in association witha controlled study on ICBT for SAD (Andersson, Carlbring, &

Table 1Mean working alliance inventory ratings and Pearson correlations with residualized change scores for the depression sample.

E-mail therapy Guided self-help Total group

M (SD) r Range M (SD) r Range M (SD) r Range

WAI total 5.58 (.82) .09 3.22e6.53 5.25 (.82) .20 4.03e6.61 5.41 (.83) .18 3.22e6.21WAI Task 5.23 (.83) .06 2.58e6.17 5.19 (.84) .27 3.17e6.92 5.21 (.82) .17 2.58e6.92WAI Goal 5.63 (.86) .14 3.33e6.75 5.08 (.92)a .17 3.58e6.75 5.37 (.92) .20 3.33e6.75WAI Bond 5.86 (.91) .05 3.75e6.83 5.47 (.97) .12 3.58e7.00 5.67 (.95) .12 3.58e7.00

a p < .05 (difference between e-mail and guided self-help).

G. Andersson et al. / Behaviour Research and Therapy 50 (2012) 544e550 547

Furmark, & on behalf of the SOFIE Research Group, 2012). Similar tothe sample II participants were screened online and then inter-viewed over the telephone to obtain a DSM-IV diagnosis of socialphobia (American Psychiatric Association, 2000). Self-reportinstruments including the short WAI were collected online. Theguided ICBT treatment lasted for 9 weeks and each participant hada personal online therapist during the trial.

Method

Procedure and treatmentFollowing screening and interview we included 204 individuals

diagnosed with SAD. The participants were randomized to eithera control (N ¼ 102) or a treatment (N ¼ 102) group. Post treatmentmeasures were obtained from 94 out of 102 randomized partici-pants in the treatment group (92%). More details regarding proce-dures and treatment outcomes are provided in Andersson et al.(2012). Alliance ratings using the short version of the WAI werecompleted during the fourth treatment week. The reason foradministrating the short WAI one week later was that the partici-pants had been given feedback on their goal formulation by thattime, but also that the treatment was one week longer than insamples I and II. This meant that they had been in contact with thestudy staff at least four times, but the major work in the treatmenthad not begun (i.e., exposure). As with samples I and II the therapistwas identified by name and a picture on the study web page. TheICBT treatment for SAD has been tested in several previouscontrolled trials (Andersson et al., 2006; Carlbring, Gunnarsdóttiret al., 2007; Furmark et al., 2009; Hedman et al., 2011), andconsists of an introductory module describing SAD and facts aboutCBT. Moreover, remaining modules cover a cognitive model forSAD, introduce cognitive restructuring, exposure exercises andattention training. Finally, the last modules include social skills andrelapse prevention. The treatment is guided with weekly feedbacksent to the participants and on occasion answers if problems arise.On average, internet therapists (N ¼ 13) spent 15 min per weekproviding feedback to each participant via the online contacthandling system (exact time not collected).

ParticipantsA total of 91 participants in the treatment group completed the

alliance measure, but of one participant had no posttreatment datayielding a total of 90 persons with complete data for this sample

Table 2Mean working alliance inventory short version ratings and Pearson correlationswith residualized change scores for the GAD sample.

Guided self-help

M (SD) Range r

WAI total 5.63 (.94) 3.75e7.00 .13WAI Task 5.60 (1.03) 3.50e7.00 .26WAI Goal 5.88 (.94) 4.25e7.00 .16WAI Bond 5.43 (1.12) 2.75e7.00 .03

(88.2%). Mean age of the 90 participants was 37.7 years (SD¼ 11.42),and 59.3% were women.

MeasuresThe primary outcome in the SAD trial was the Liebowitz Social

Anxiety Scale self-report version (LSAS-SR) (Baker, Heinrichs, Kim,& Hofmann, 2002; Fresco et al., 2001). The LSAS-SR measures fearin and avoidance of 24 social situations (13 performance- and 11interaction situations), that are assumed to be difficult for peoplesuffering from social anxiety disorder. Fear and avoidance in eachsituation is rated on a four-point scale ranging from 0 (no fear/never avoid) to 3 (severe fear/usually avoid). The LSAS-SR has goodpsychometric properties, and has also been validated for online use(Hedman et al., 2010).

We used the same 12-item online version of the WAI as insample II (Andrusyna et al., 2001; Tracey & Kokotovic, 1989).Psychometric properties for the online version of the short WAI forthis sample of persons with SAD was a ¼ .88 for the Task subscale,a ¼ .79 for the Goal subscale, and a ¼ .89 for the Bond subscale. Forthe total score the internal consistency was a¼ .94. Therewere alsohigh intercorrelations between the total score and the subscales(r ¼ .81 to r ¼ .94), suggesting that the subscales and total scaleoverlap substantially.

Results

In terms of treatment outcome the treatment group improvedwith a within-group effect size of d ¼ .97 on the LSAS-SR. The fewnon-responders to theWAI did not differ from the completers (withregards to age, pretreatment LSAS-SR or residual change scores onthe LSAS-SR). Mean alliance ratings on the short WAI includingcorrelations with residualized change scores on the LSAS-SR arepresented in Table 3. In line with samples I and II correlations wereagain not statistically significant. The WAI was finally correlatedwith age and pretreatment LSAS-SR scores. We found no significantcorrelations.

Discussion

The aim of this study was to investigate the role of therapeuticalliance in guided ICBT. We included data from three studysamples diagnosed with major depression, generalized anxietydisorder, or social anxiety disorder. The rationale for the study was

Table 3Mean working alliance inventory short version ratings including SD’s and ranges,and Pearson correlations with residualized change scores for the SAD sample.

Guided self-help

M (SD) Range r

WAI total 5.45 (1.05) 2.83e6.92 .10WAI Task 5.24 (1.16) 2.75e7.00 .08WAI Goal 5.61 (.98) 2.75e7.00 .10WAI Bond 5.5 (1.22) 2.50e7.00 .11

G. Andersson et al. / Behaviour Research and Therapy 50 (2012) 544e550548

that guided ICBT has been found to be effective (Andrews, Cuijpers,Craske, McEvoy, & Titov, 2010), but the mechanisms driving theeffects are largely unknown (Andersson, 2010). Therapeutic alli-ance is often regarded as an important factor in psychotherapyoutcome research, with a robust but modest correlation withoutcome (Horvath, Del Re, Fluckiger, & Symonds, 2011). Thetemporal association between alliance as an effect of treatmentrather than as a predictor has however been debated (Barber,Connolly, Crits-Christoph, Gladis, & Siqueland, 2000; Feeley,DeRubeis, & Gelfand, 1999), and it has been argued that alliancecould be an outcome and rather than a predictor (Feeley et al.,1999). Moreover, the different components of the alliance maybe differently associated with outcome in CBT than in otherpsychotherapy orientations (Webb et al., 2011). In addition to thepossible differences between CBT and other forms of psycho-therapy, the role of the alliance in other delivery formats of CBT isnot well investigated. However, there are examples of studies inwhich high alliance ratings had been obtained in ICBT(Knaevelsrud & Maercker, 2007; Ljótsson et al., 2011). In all threesamples we found high ratings of alliance with mean scores for thetotal and subscales in line with what is usually found in face-to-face treatments (Busseri & Tyler, 2003). Moreover, the psycho-metric properties of the WAI were good across all samples withhigh internal consistency for all subscales and the total score, againin line with face-to-face administration (Busseri & Tyler, 2003) andprevious research investigating how well self-report measures canbe transferred to online administration (Carlbring, Brunt et al.,2007; Hedman et al., 2010; Holländare, Andersson, & Engström,2010). In sample I we found one statistically significant differ-ence in favour of e-mail therapy on the WAI goal subscale. Thisdifference is most likely explained by the lower contact time in theguided self-help group which was less than in samples II and III.The mean goal subscale score for the e-mail group was similar towhat was found in samples II and III.

The association between alliance ratings and outcome was non-significant across all three samples. Since very few researchers haveinvestigated if early alliance is predictive of outcome in ICBT it isdifficult to compare our findings against previous studies. In linewith most results reported by Knaevelsrud and Maercker (2006)correlations in our three samples were in a positive direction butnot statistically significant. Our study samples are different in thatwe had less therapist contact and larger samples than in theKnaevelsrud and Maercker trial. Statistical power could howeverhave an impact on our findings, but when we pooled all threesamples there was no significant association between WAI totalscore and residualized change scores on the main outcomemeasures (r ¼ .06). Another possibility is restriction of range whichwould decrease the likelihood of finding significant associations.However, the alliance scores showed a variation in all three samplesmaking this less likely.

The findings in this study calls for an explanation and also haveimplications for the broader field of psychotherapy outcomeresearch. First, in contrast to most studies on unguided ICBT (Spek,Cuijpers et al., 2007), therewere robust and large treatments effectsof guided ICBT which is in linewith what has been found in face-to-face trials. Therefore lack of treatment effects cannot explain thefindings.

Second, given theminimal amount of therapist contact in ICBT itwould not be surprising if alliance ratings had been lower if thealliance measure had focused on what occurs in the therapy roomonly. However, the WAI is focused on concrete aspects of thetherapy (task, goal, bond) and it may be that the client perceiveagreement on theses aspects by reading texts inwhich the therapistis ”present” in the form of author of the self-help text. Indeed, it hasbeen suggested that alliance fostering aspects are included in self-

help texts as we alluded to in the introduction (Richardson et al.,2010). However, a few of our participants indicated that it wasa bit strange to complete the WAI as they had received littleinformation from their therapists. One possible explanation for thiscould be the fact that some clients in ICBT engage less with thetherapy which includes not interacting much with either thetherapists or reading text material carefully (Bendelin et al., 2011).It is possible that more specific measures of therapeutic alliance inICBT needs to be developed in line with the reasoning in researchon group treatment where alliance may be more to a group (i.e.,cohesion) rather than to a therapist (Burlingame, Fuhriman, &Johnson, 2002).

A third interesting angle on the observation that high allianceratings can be obtained with an online therapist is the possible lackof negative therapist effects. It has been argued that therapisteffects explain around 5% of the outcome variance in therapy(Wampold & Brown, 2005). However, we have not found anyindications that it makes much difference who the therapist is inICBT, even if therapist behaviours are not arbitrary (Almlöv,Carlbring, Berger, Cuijpers, & Andersson, 2009; Almlöv et al.,2011). Horvath and Bedi (2002) mentioned that some therapistbehaviour in face-to-face treatments may lead to poor or deterio-rating alliance (Horvath & Bedi, 2002). Most therapist interaction inICBT consists of encouragement and validation of homeworkcompletion, and it is possible that more negative therapist behav-iours such as ”taking charge” are more absent in ICBT. One way tostudy this factor would be to collect alliance ratings from thetherapists, which we did not do. Indeed, it has consistently beenfound that clients rate the alliance as higher than ratings by theirtherapist (Tryon, Blackwell, & Hamel, 2007), but whether this istrue for ICBT as well is not known.

Fourth, in this study we decided to study both the WAI totalscore and the subscores. We found no major differences betweenthe subscales in terms of their predictive value, but in line withprevious research (Webb et al., 2011) and for exploratory reasonswe decided to report data on both the total and subscore scales.

A fifth consideration concerns the theoretical orientation of theICBT treatment, which was cognitive-behavioural. While there islittle to suggest that the alliance is less important in CBT versusother therapy orientations (Horvath et al., 2011), we do not know ifthis is the case in guided internet treatments. Indeed, in a study ontwo forms of telephone-administered treatments for depression itwas found that alliance moderated the outcome in the CBT condi-tion, but not in the supportive emotion-focused therapy condition(Beckner, Vella, Howard, & Mohr, 2007). Telephone-administeredCBT is very different from guided ICBT and one major differenceis the fact that the therapist is only available off-line in ICBT (i.e., notin real time). There is however at least one large study on real timeonline CBT for depression showing good outcome (Kessler et al.,2009), but the role of therapeutic alliance was not investigated.Overall, theway alliance is conceptualized differs between differenttherapy orientations (Catty, 2004), but in behavioural terms thetherapist is often viewed as a provider of social reinforcement(Follette, Naugle, & Callaghan, 1996), which is very much thetherapist role in ICBT.

A final sixth consideration relates to the decision in this study tofocus on the primary outcomemeasures in the three trials. This wasdone to prevent type I error but can be questioned as it has beensuggested that WAI ratings may have more overlap with generalimprovement scores rather than symptom-specific measures(Busseri & Tyler, 2003). Overall, we conclude that while ICBT iseffective for the primary complaints for persons with depression,GAD and SAD seek help for, alliance is not the explanation.

There are limitations that should be noted. First, all threesamples were self-recruited and therapeutic alliance may be either

G. Andersson et al. / Behaviour Research and Therapy 50 (2012) 544e550 549

more or less important in samples recruited from in clinicallyrepresentative settings. Data on ICBT in effectiveness studies oftenreplicate the findings in efficacy studies (Aydos, Titov, & Andrews,2009; Bergström et al., 2010; Hedman et al., 2011), but allianceratings have not been collected in these studies. Second, the WAIwas developed for the assessment of face-to-face relations betweena therapist and a client it is not clear if the minor adaptations wemade (mainly changing from “therapist” to “online therapist”),were enough. Third, we have no control over temporal confounds.Alliance ratings were collected early in the treatment process, but itis still possible that the high ratings are a consequence of thetreatment rather than a sign of a therapeutic alliance (Webb et al.,2011). It is for example possible that WAI ratings would have beenthe same even in the absence of therapist contact (e.g., expectedalliance), and measures of alliance at several time points could helpto disentangle the temporal confounds. Moreover, alliance mayfluctuate over the therapy process (Safran & Muran, 2000) whichwould require repeated measures of the WAI to establish if fluc-tuations influence outcome.

We conclude that alliance, as it is conceptualized in Bordin’spantheoretical model (Bordin, 1979), is probably less important inICBT than in standard face-to-face therapies. Future researchshould be informed by how clients perceive ICBT (Bendelin et al.,2011), as it is unlikely that factors that make a difference in face-to-face therapies are equally important in ICBT. However, we alsoconclude that participants in ICBT trials tend to rate the alliance ashigh, which raises many questions regarding what theWAI actuallymeasures and if therapist factors really are that important togenerate good outcome as has been stated in the psychotherapyliterature (Ahn & Wampold, 2001).

Acknowledgements

This research was sponsored in part by grants from the Swedishcouncil for working and life research, the Swedish research council,and Linköping University. We thank all coworkers in the Klara,Origo and Sofie 6 trials for the contributions.

References

Ahn, H. N., & Wampold, B. E. (2001). Where oh where are the specific ingredients? Ameta-analysis of component studies in counseling and psychotherapy. Journalof Counseling Psychology, 48, 251e257.

Almlöv, J., Carlbring, P., Berger, T., Cuijpers, P., & Andersson, G. (2009). Therapistfactors in internet-delivered CBT for major depressive disorder. CognitiveBehaviour Therapy, 38, 247e254.

Almlöv, J., Carlbring, P., Källqvist, K., Paxling, B., Cuijpers, P., & Andersson, G. (2011).Therapist effects in guided Internet-delivered CBT for anxiety disorders.Behavioural and Cognitive Psychotherapy, 39, 311e322.

American Psychiatric Association. (2000). Diagnostic and statistical manual of mentaldisorders (4th ed., text revision ed.). Washington, DC: American Psychiatric Press.

Andersson, G. (2009). Using the internet to provide cognitive behaviour therapy.Behaviour Research and Therapy, 47, 175e180.

Andersson, G. (2010). The promise and pitfalls of the internet for cognitivebehavioural therapy. BMC Medicine, 8(82).

Andersson, G., Bergström, J., Holländare, F., Carlbring, P., Kaldo, V., & Ekselius, L.(2005). Internet-based self-help for depression: a randomised controlled trial.British Journal of Psychiatry, 187, 456e461.

Andersson, G., Carlbring, P., Furmark, T., & on behalf of the SOFIE Research Group.(2012). Therapist experience and knowledge acquisition in Internet-deliveredCBT for social anxiety disorder: A randomized controlled trial. PloS ONE, 7(5),e37411.

Andersson, G., Carlbring, P., Holmström, A., Sparthan, E., Furmark, T., Nilsson-Ihrfelt, E., et al. (2006). Internet-based self-help with therapist feedback and in-vivo group exposure for social phobia: a randomized controlled trial. Journal ofConsulting and Clinical Psychology, 74, 677e686.

Andersson, G., & Cuijpers, P. (2009). Internet-based and other computerizedpsychological treatments for adult depression: a meta-analysis. CognitiveBehaviour Therapy, 38, 196e205.

Andrews, G., Cuijpers, P., Craske, M. G., McEvoy, P., & Titov, N. (2010). Computertherapy for the anxiety and depressive disorders is effective, acceptable andpractical health care: a meta-analysis. PloS ONE, 5(10), e13196.

Andrews, G., Davies, M., & Titov, N. (2011). Effectiveness randomized controlled trialof face to face versus Internet cognitive behaviour therapy for social phobia.Australian and New Zealand Journal of Psychiatry, 45, 337e340.

Andrusyna, T. P., Tang, T. Z., DeRubeis, R. J., & Luborsky, L. (2001). The factorstructure of the working alliance inventory in cognitive-behavioral therapy. TheJournal of Psychotherapy Practice and Research, 10, 173e178.

Aydos, L., Titov, N., & Andrews, G. (2009). Shyness 5: the clinical effectiveness ofinternet-based clinician-assisted treatment of social phobia. AustralasianPsychiatry, 17, 488e492.

Baker, S. L., Heinrichs, N., Kim, H. J., & Hofmann, S. G. (2002). The Liebowitz socialanxiety scale as a self-report instrument: a preliminary psychometric analysis.Behaviour Research and Therapy, 40, 701e715.

Barber, J. P., Connolly, M. B., Crits-Christoph, P., Gladis, L., & Siqueland, L. (2000).Alliance predicts patients’ outcome beyond in-treatment change in symptoms.Journal of Consulting and Clinical Psychology, 68, 1027e1032.

Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventoryfor measuring depression. Archives of General Psychiatry, 4, 561e571.

Beckner, V., Vella, L., Howard, I., & Mohr, D. C. (2007). Alliance in two telephone-administered treatments: relationship with depression and health outcomes.Journal of Consulting and Clinical Psychology, 75, 508e512.

Bendelin, N., Hesser, H., Dahl, J., Carlbring, P., Zetterqvist Nelson, K., & Andersson, G.(2011). Experiences of guided internet-based cognitive-behavioural treatmentfor depression: a qualitative study. BMC Psychiatry, 11, 107.

Bergström, J., Andersson, G., Ljótsson, B., Rück, C., Andréewitch, S., Karlsson, A., et al.(2010). Internet- versus group-administered cognitive behaviour therapy forpanic disorder in a psychiatric setting: a randomised trial. BMC Psychiatry, 10,54.

Bordin, E. S. (1979). The generalizability of the psychoanalytic concept of theworking alliance. Psychotherapy: Theory, Research and Practice, 16, 252e260.

Burlingame, G. M., Fuhriman, A., & Johnson, J. (2002). Cohesion in group psycho-therapy. In J. C. Norcross (Ed.), Psychotherapy relationships that work (pp.71e87). Oxford: Oxford University Press.

Busseri, M. A., & Tyler, J. D. (2003). Interchangeability of the working allianceinventory and working alliance inventory, short form. Psychological Assessment,15, 193e197.

Carlbring, P., Brunt, S., Bohman, S., Austin, D., Richards, J. C., Öst, L.-G., et al. (2007).Internet vs. paper and pencil administration of questionnaires commonly usedin panic/agoraphobia research. Computers in Human Behavior, 23, 1421e1434.

Carlbring, P., Gunnarsdóttir, M., Hedensjö, L., Andersson, G., Ekselius, L., &Furmark, T. (2007). Treatment of social phobia: randomized trial of Internetdelivered cognitive behaviour therapy and telephone support. British Journal ofPsychiatry, 190, 123e128.

Carlbring, P., Nilsson-Ihrfelt, E., Waara, J., Kollenstam, C., Buhrman, M., Kaldo, V.,et al. (2005). Treatment of panic disorder: live therapy vs. self-help via Internet.Behaviour Research and Therapy, 43, 1321e1333.

Catty, J. (2004). ’The vehicle of success’: theoretical and empirical perspectives onthe therapeutic alliance in psychotherapy and psychiatry. Psychology andPsychotherapy: Theory, Research and Practice, 77, 255e272.

Cook, J. E., & Doyle, C. (2002). Working alliance in online therapy as compared toface-to-face therapy: preliminary results. Cyberpsychology & Behavior, 5,95e105.

D’Arcy, J., Reynolds, D. J., Stiles, W. B., & Grohol, J. M. (2006). An investigation ofsession impact and alliance in Internet based psychotherapy: preliminaryresults. Counselling and Psychotherapy Research, 6, 164e168.

Feeley, M., DeRubeis, R. J., & Gelfand, L. A. (1999). The temporal relation of adher-ence and alliance to symptom change in cognitive therapy for depression.Journal of Consulting and Clinical Psychology, 67, 578e582.

Follette, W. C., Naugle, A. E., & Callaghan, G. M. (1996). A radical behavioralunderstanding of the therapeutic relationship in effecting change. BehaviorTherapy, 27, 623e641.

Fresco, D. M., Coles, M. E., Heimberg, R. G., Liebowitz, M. R., Hami, S., Stein, M. B.,et al. (2001). The Liebowitz social anxiety scale: a comparison of the psycho-metric properties of self-report and clinician-administered formats. Psycho-logical Medicine, 31, 1025e1035.

Furmark, T., Carlbring, P., Hedman, E., Sonnenstein, A., Clevberger, P., Bohman, B.,et al. (2009). Guided and unguided self-help for social anxiety disorder: rand-omised controlled trial. British Journal of Psychiatry, 195, 440e447.

Hedman, E., Andersson, G., Ljótsson, B., Andersson, E., Rück, C., Mörtberg, E., et al.(2011). Internet-based cognitive behavior therapy vs. cognitive behavioralgroup therapy for social anxiety disorder: a randomized controlled non-inferiority trial. PloS ONE, 6(3), e18001.

Hedman, E., Ljótsson, B., Rück, C., Furmark, T., Carlbring, P., Lindefors, N., et al.(2010). Internet administration of self-report measures commonly used inresearch on social anxiety disorder: a psychometric evaluation. Computers inHuman Behavior, 26, 736e740.

Holländare, F., Andersson, G., & Engström, I. (2010). A comparison of psychometricproperties between Internet and paper versions of two depression instruments(BDI-II and MADRS-S) administered to clinic patients. Journal of Medical InternetResearch, 12(5), e49.

Horvath, A. O., & Bedi, R. P. (2002). The alliance. In J. C. Norcross (Ed.), Psychotherapyrelationships that work (pp. 37e69). Oxford: Oxford University Press.

Horvath, A. O., Del Re, A. C., Fluckiger, C., & Symonds, D. (2011). Alliance in indi-vidual psychotherapy. Psychotherapy, 48, 9e16.

Horvath, A. O., & Greenberg, L. S. (1989). The development and validation of theworking alliance inventory. Journal of Counseling Psychology, 36, 223e233.

G. Andersson et al. / Behaviour Research and Therapy 50 (2012) 544e550550

Kaldo, V., Levin, S., Widarsson, J., Buhrman, M., Larsen, H. C., & Andersson, G. (2008).Internet versus group cognitive-behavioral treatment of distress associatedwith tinnitus. A randomised controlled trial. Behavior Therapy, 39, 348e359.

Kessler, D., Lewis, G., Kaur, S., Wiles, N., King, M., Weich, S., et al. (2009). Therapist-delivered internet psychotherapy for depression in primary care: a randomisedcontrolled trial. Lancet, 374, 628e634.

Kiropoulos, L. A., Klein, B., Austin, D. W., Gilson, K., Pier, C., Mitchell, J., et al. (2008).Is internet-based CBT for panic disorder and agoraphobia as effective as face-to-face CBT? Journal of Anxiety Disorders, 22, 1273e1284.

Klein, B., Austin, D., Pier, C., Kiropoulos, L., Shandley, K., Mitchell, J., et al. (2009).Internet-based treatment for panic disorder: does frequency of therapistcontact make a difference? Cognitive Behaviour Therapy, 38, 121e131.

Klein, B., Mitchell, J., Gilson, K., Shandley, K., Austin, D., Kiropoulos, L., et al. (2009).A therapist-assisted internet-based CBT intervention for posttraumatic stressdisorder: preliminary results. Cognitive Behaviour Therapy, 38, 121e131.

Knaevelsrud, C., & Maercker, A. (2006). Does the quality of the working alliancepredict treatment outcome in online psychotherapy for traumatized patients?Journal of Medical Internet Research, 8(4), e31.

Knaevelsrud, C., & Maercker, A. (2007). Internet-based treatment for PTSD reducesdistress and facilitates the development of a strong therapeutic alliance:a randomized controlled clinical trial. BMC Psychiatry, 7, 13.

Lambert, M. J., & Barley, D. E. (2002). Research summary on the therapeutic rela-tionship and psychotherapy outcome. In J. C. Norcross (Ed.), Psychotherapyrelationships that work (pp. 17e32). Oxford: Oxford University Press.

Ljótsson, B., Hedman, E., Andersson, E., Hesser, H., Lindfors, P., Hursti, T., et al. (2011).Internet-delivered exposure based treatment vs. stress management for irri-table bowel syndrome: a randomized trial. American Journal of Gastroenterology,106, 1481e1491.

Meyer, T. J., Miller, M. L., Metzger, R. L., & Borkovec, T. D. (1990). Development andvalidation of the Penn state worry questionnaire. Behaviour Research andTherapy, 28, 487e495.

Paxling, B., Almlöv, J., Dahlin, M., Carlbring, P., Breitholtz, E., Eriksson, T., et al. (2011).Internet-delivered cognitive behaviour therapy for generalized anxiety disorder.A randomized controlled trial. Cognitive Behaviour Therapy, 40, 159e173.

Paxling, B., Lundgren, S., Norman, A., Almlöv, J., Carlbring, P., Cuijpers, P., & et al.Therapist behaviours in Internet-delivered cognitive behaviour therapy -

Analyses of e-mail correspondence in the treatment of generalized anxietydisorder. Behavioural and Cognitive Psychotherapy, in press.

Richardson, R., Richards, D. A., & Barkham, M. (2010). Self-help books for peoplewith depression: the role of the therapeutic relationship. Behavioural andCognitive Psychotherapy, 38, 67e81.

Safran, J. D., & Muran, J. C. (2000). Negotiating the therapeutic alliance. New York:Guilford Press.

Spek, V., Cuijpers, P., Nyklicek, I., Riper, H., Keyzer, J., & Pop, V. (2007). Internet-based cognitive behaviour therapy for symptoms of depression and anxiety:a meta-analysis. Psychological Medicine, 37, 319e328.

Spek, V., Nyklicek, I., Smits, N., Cuijpers, P., Riper, H., Keyzer, J., et al. (2007).Internet-based cognitive behavioural therapy for subthreshold depression inpeople over 50 years old: a randomized controlled clinical trial. PsychologicalMedicine, 37, 1797e1806.

Steketee, G., & Chambless, D. L. (1992). Methodological issues in prediction oftreatment outcome. Clinical Psychology Review, 12, 387e400.

Titov, N., Andrews, G., Robinson, E., Schwencke, G., Johnston, L., Solley, K., et al.(2009). Clinician-assisted Internet-based treatment is effective for generalizedanxiety disorder: randomized controlled trial. Australian and New ZealandJournal of Psychiatry, 43, 905e912.

Tracey, T. J., & Kokotovic, A. M. (1989). Factor structure of the working allianceinventory. Psychological Assessment, 1, 207e2010.

Tryon, G. S., Blackwell, S. C., & Hamel, E. F. (2007). A meta-analytic examination ofclient-therapist perspectives of the working alliance. Psychotherapy Research, 17,629e642.

Vernmark, K., Lenndin, J., Bjärehed, J., Carlsson, M., Karlsson, J., Öberg, J., et al.(2010). Internet administered guided self-help versus individualized e-mailtherapy: a randomized trial of two versions of CBT for major depression.Behaviour Research and Therapy, 48, 368e376.

Wampold, B. E., & Brown, G. S. (2005). Estimating the variability outcome attrib-utable to therapists: a naturalistic study of outcomes in managed care. Journalof Consulting and Clinical Psychology, 73, 914e923.

Webb, C. A., DeRubeis, R. J., Amsterdam, J. D., Shelton, R. C., Hollon, S. D., &Dimidjian, S. (2011). Two aspects of the therapeutic alliance: differential rela-tions with depressive symptom change. Journal of Consulting and ClinicalPsychology, 79, 279e283.