MI lacks conceptual stability

One prominent critique of MI is that it lacks conceptual stability. So, what is MI? How do we know whether what we are testing in studies and teaching practitioners really is MI?

Both Björk (Reference Björk2014) and Atkinson and Woods (Reference Atkinson and Woods2017) fault the conceptual stability of MI, describing how definitions of MI have changed over time. For example, Miller and Rollnick (Reference Miller and Rollnick1991) first set forth five principles of MI: express empathy, develop discrepancy, avoid argumentation, roll with resistance, and support self-efficacy. In their second edition (Miller and Rollnick, Reference Miller and Rollnick2002) this was reduced to four principles, collapsing develop discrepancy and avoid argumentation into ‘roll with resistance’. Next, they vacated the principles altogether, instead describing four processes of MI (engaging, focusing, evoking, and planning) and deconstructing their previously used concept of resistance (Miller and Rollnick, Reference Miller and Rollnick2013). Atkinson and Woods also note other changes in the definition of MI, including that from 2003 onwards there was increased focus on client language by eliciting change talk and avoiding elaboration on the client’s reasons for maintaining status quo behaviors (sustain talk), and later increased emphasis on avoiding MI-inconsistent responses. There also appeared a new description of the underlying ‘spirit’ with which MI is to be practised (Rollnick and Miller, Reference Rollnick and Miller1995) that has received increasing emphasis through subsequent editions of the principal text (Miller and Rollnick, Reference Miller and Rollnick2023). Whereas subtitles of the 1991 and 2002 texts described preparing people for change, subsequent versions removed this preparatory emphasis, focusing instead more broadly on helping people change.

Some of the changes made in how MI is defined and described have been carefully explained with reference to data from emerging research. One such example is how client speech came to feature more prominently in descriptions of MI after 2003, particularly influenced by the psycholinguistic research of Paul Amrhein (Amrhein, Reference Amrhein2004; Amrhein et al., Reference Amrhein, Miller, Yahne, Palmer and Fulcher2003). Different categories of client change talk (and sustain talk) were recognized, and MI practitioners were advised how to respond to client speech as a result of emerging research showing that client in-session speech and how MI practitioners respond to it predict whether behaviour change will happen. It became clear from correlational, sequential, and experimental studies that counsellors influence the balance of clients’ change talk and sustain talk (Apodaca and Longabaugh, Reference Apodaca and Longabaugh2009; Gaume et al., Reference Gaume, Bertholet, Faouzi, Gmel and Daeppen2010; Glynn and Moyers, Reference Glynn and Moyers2010; Walthers et al., Reference Walthers, Janssen, Mastroleo, Hoadley, Barnett, Colby and Magill2019) which in turn predicts subsequent client behaviour change (Gaume et al., Reference Gaume, Bertholet, Faouzi, Gmel and Daeppen2013; Lindqvist et al., Reference Lindqvist, Forsberg, Enebrink, Andersson and Rosendahl2017; Magill et al., Reference Magill, Apodaca, Borsari, Gaume, Hoadley, Gordon, Tonigan and Moyers2018).

Other changes to MI’s definition or description, however, have not been explained in relation to research findings. Atkinson and Woods note, for example, that no clear explanation was given for the transition from describing MI in terms of principles to describing it in terms of processes. Another example of an unexplained theoretical shift is the varying relationship between MI and the transtheoretical model of change (TTM; Prochaska and DiClemente, Reference Prochaska and DiClemente1984) that seemed central in early descriptions of MI (Miller, Reference Miller1983; Miller and Rollnick, Reference Miller and Rollnick1991), then more distant in later editions. Frustration with such conceptual changes is understandable. They lead to uncertainty regarding what is in fact being tested in studies and what is being taught in MI training. TTM (Prochaska and DiClemente, Reference Prochaska and DiClemente1984) supports clinicians’ understanding of different types of ambivalence and responding to them appropriately. TTM’s formulation of change as a multi-faceted process including many types of ambivalence, rather than seeing change as a dichotomous process may have contributed to MI’s spread to areas where more complicated change is the focus, such as smoking, alcohol use, etc. Different types of ambivalence require different MI skills. However, TTM’s theoretical foundation in MI has been toned down and the reasoning for doing so has been unclear.

Atkinson and Woods also point out that available treatment manuals tend to describe not MI on its own – or ‘pure’ MI – but rather how MI can be combined with other interventions as in motivational enhancement therapy (MET; Miller et al., Reference Miller, Zweben, DiClemente and Rychtarik1992) or a combined behavioural intervention (Miller, Reference Miller2004). Similarly, other researchers have described how MI should be done in concert with another study intervention (e.g. Naar et al., Reference Naar, Pennar, Wang, Brogan-Hartlieb and Fortenberry2021), often referred to as adaptations of MI (Burke et al., Reference Burke, Arkowitz and Menchola2003). ‘Pure’ MI had not been manualized until recently (Hurlocker et al., Reference Hurlocker, Moyers, Hatch, Curran, McCrady, Venner and Witkiewitz2023). It is indeed the case that by far the most common use of MI in controlled clinical trials has been in combination with other evidence-based treatment, sometimes as a pre-treatment intervention (Westra et al., Reference Westra, Arkowitz and Dozois2009) but more often integrated in less specified ways. In the largest randomized trial of treatments for alcohol use disorders (Project Match, Babor and Del Boca, Reference Babor and Del Boca2003), MI in combination with objective health feedback was associated with behaviour change, with developing discrepancy on the basis of negative objective health feedback being an important ingredient.

In MI, the focus now is on how ambivalence is expressed in language, even if ambivalence can also be expressed in feelings, body language, silence, and behaviour. The concept of ambivalence can be seen as having been given an operational definition in MI in client speech concerning change versus client speech favouring the status quo.

Atkinson and Woods’ criticism that MI lacks a stable definition and a manual for its use is echoed by Björk (Reference Björk2014) who noted that there is often no definition of MI in the scientific studies that have been conducted on its efficacy and effectiveness. Björk, too, notes that when defined in research study manuals, MI tends to be used in combination with another intervention rather than as ‘pure MI’. Both argue that the absence of clear definition gives rise to important methodological and practical problems for the method and its practice, as well as for research on its efficacy and effectiveness. For example, the lack of definition makes it uncertain exactly what has been delivered in different studies and to what extent MI is the same across studies. Changes in how MI was described in the early 1990s versus 20 years later may have caused differences in how MI was done in studies. MI is often described as an evidence-based method for changing behaviour. However, it is likely that the intervention referred to as ‘MI’ has been done differently across the many studies that make up its evidence basis. This concern was also raised by Miller and Rollnick (Reference Miller and Rollnick2014) with regard to MI and behavioural interventions more generally.

Björk also notes, however, that while MI has lacked a clear and consistent definition, the method has been stabilized by other kinds of efforts. One such effort is the development and widespread use of MI fidelity assessment tools, the first of which was the Motivational Interviewing Skills Code (MISC; DeJonge et al., Reference DeJonge, Schippers and Schaap2005; Miller and Mount, Reference Miller and Mount2001). A simplified Motivational Interviewing Treatment Integrity (MITI) code followed, with demonstrated reliability and validity, and which continues to be updated (Moyers et al., Reference Moyers, Rowell, Manuel, Ernst and Houck2016), with practice samples more recently subjected to automated machine coding via voice recognition (e.g. Tanana et al., Reference Tanana, Hallgren, Imel, Atkins and Srikumar2016). It remains unclear what levels of proficiency practitioners need on such measures in order to improve clients’ outcomes. It is also clear that in most research and clinical applications of MI to date there has been little or no use of such tools to document fidelity (Lundahl et al., Reference Lundahl, Moleni, Burke, Butters, Tollefson, Butler and Rollnick2013), so their overall effect on stabilizing MI practice may be minimal (Atkinson and Woods, Reference Atkinson and Woods2017).

In psychotherapy, the idea of a standardized treatment manual is a relatively recent by-product of funding for clinical trials that require specification of the interventions to be evaluated. When developing new psychological treatments, it has become common to first develop a step-by-step therapist manual to guide practice and later the dissemination of the intervention. Such homogenization of practice is neither common nor particularly welcome in most clinical service settings, but is it a good idea? Atkinson and Woods note that in the Lundahl et al. (Reference Lundahl, Moleni, Burke, Butters, Tollefson, Butler and Rollnick2013) meta-analysis of trials in health care settings, studies that measured MI fidelity produced lower effect sizes compared with those that did not document fidelity. Another meta-analysis (Hettema et al., Reference Hettema, Steele and Miller2005) found that intervention effect size was significantly lower when the delivery of MI was manual-guided. These two findings indicate that attempts to stabilize the fidelity of MI and minimize its variability may reduce its effectiveness. In a clinical trial for which William Miller personally wrote the standardized treatment manual and personally trained and supervised the therapists, there was no effect of MET on client outcomes (Miller et al., Reference Miller, Yahne and Tonigan2003). One explanation could be that treatment integrity measured in terms of fidelity to a manual or treatment integrity assessment tool might miss important MI skills. Perhaps process studies can shed some light on this issue, highlighting a need to understand MI better. The fidelity issue may be a signal that manuals have lacked a component that is otherwise easier to perform, and that what we measure in MI perhaps does not correlate enough with well-performed MI. Clearly there is more to learn about the content of MI.

In his critique, Björk observes that in contrast to manual-guided treatment, MI has evolved in a manner similar to how technological innovations are often developed and disseminated (cf. Rogers, Reference Rogers2003). Such innovations are often co-created by many people using a methodology that seeks to understand how the innovation works and why. Björk notes that the invention of a technology is seldom an isolated event but rather a long-term process whereby people involved in research and in practical applications test applications in new areas. Technological innovations often begin with attempts to measure features that are hypothesized to be active components. MI seems to mirror this in the way in which researchers and practitioners of MI have both been involved in developing the method and applying it in many different fields and contexts.

In MI, this broad diffusion has happened by training people to teach others in how to use the method. A first Training of New Trainers (TNT) was offered by Miller and Rollnick in 1993. In 1995, a loose network of MI trainers was formed that would later be formally organized as the Motivational Interviewing Network of Trainers (MINT). As Björk notes, MINT established ties between researchers, trainers, and practitioners, developing a culture that encourages its members to ‘give more than you take’. Materials and methods were freely shared to facilitate MI practice, training, and research. First on an email listserv and later on a web-based platform, many kinds of MI-related challenges and solutions were discussed among MINT members. Once a year, new MI trainers were trained, and in connection with the TNT there developed an informal MI conference known as the MINT Forum, where members shared updates, views, and innovations. As new research findings emerged these, too, were disseminated via MINT, influencing future training, practice, and research. This organized collaboration between practitioners, trainers and researchers for more than 30 years has stabilized what MI is and supported the practice and training of MI in a unique way for a psychotherapeutic intervention and its development.

In essence, MI has been analogous to open-source software – freely available for those who are interested in trying it. The authors made no attempt to trademark, franchise, copyright, or otherwise control or restrict its use (Miller, Reference Miller2023; Miller and Rollnick, Reference Miller and Rollnick2023). This may account in part for its widespread adoption across different problems, settings, practitioner groups, nations, and languages (Bjőrk, Reference Björk2014). Atkinson and Woods also note that MI as an intervention is flexible and capable of being applied in many different contexts with a wide range of clients. Like the person-centred approach of Carl Rogers on which it is based (Miller and Moyers, Reference Miller and Moyers2017), MI has been applied in many fields including education (Herman et al., Reference Herman, Reinke, Frey and Shepard2021; Rollnick et al., Reference Rollnick, Kaplan and Rutschman2016), negotiation (Amador, Reference Amador2022), pastoral care (Clarke et al., Reference Clarke, Giordano, Cashwell and Lewis2013; Miller et al., Reference Miller, Forcehimes, O’Leary and LaNoue2008), leadership and management (Marshall and Nielsen, Reference Marshall and Nielsen2020; Organ, Reference Organ2021), and social work (Forrester et al., Reference Forrester, Wilkins and Whittaker2021; Hohman, Reference Hohman2021).

A necessary consequence of open sourcing is a lack of consistency and quality assurance in MI delivery and training. No permission or certification is required to practise MI or claim to do so. The same is true, of course, for nearly all forms of therapy, counselling, and coaching. Even if it were desirable for a treatment method to be unilaterally defined and unchanging, that is not the reality of psychotherapies. Their processes and outcomes vary with the person providing them (Miller and Moyers, Reference Miller and Moyers2021).

The MINT offers to practitioners and trainers updated training methods and exercises, new research findings, and innovative applications of MI. MINT’s collaborative and supportive culture of sharing and giving back have contributed to the development and dissemination of MI. MINT also promotes interaction among professionals in research, practice, and training. As noted above, a potential downside of such free exchange is a lack of control over how MI is spread and used. In combination with conceptual changes over time, there is understandable concern about what is actually being delivered in practice, taught in training, and tested in studies of MI.

MI lacks a theoretical foundation

A second criticism of MI is that it has no consistent or coherent theoretical basis (Atkinson and Woods, Reference Atkinson and Woods2017). Implicit in this critique is an ideal that a psychotherapy should be deductively derived from and guided by a pre-existing theory of personality or therapy. Proponents of this critique have been less clear about what disadvantages are bestowed by the lack of an a priori theoretical foundation. Are atheoretical therapies inherently more difficult to learn or more variable in practice? Effective medications are sometimes discovered by accident without a theoretical reason for or understanding of mechanisms of their efficacy.

In becoming President of the American Psychological Association in 1947, Carl Rogers argued that clinical psychology should be an empirical science with measurable therapeutic process and outcomes. His work was a nascent clinical science and a forerunner of current research on active ingredients and mechanisms of change in psychotherapy (e.g. Magill et al., Reference Magill, Kiluk, McCrady, Tonigan and Longabaugh2015). His person-centred approach was derived not from a pre-existing theory but abductively through close observation of clinical practice to develop and test hypotheses about what therapeutic factors actually help clients change (Kirschenbaum, Reference Kirschenbaum2009). Theories arose later to explain the results being observed (e.g. Gendlin, Reference Gendlin1961; Rogers, Reference Rogers and Koch1959).

MI similarly began from close examination and discussion of therapeutic practice (Miller, Reference Miller1983) stimulated in part by the incidental finding of an unexpectedly large effect of counsellor empathy on cognitive behaviour therapy outcomes (Miller et al., Reference Miller, Taylor and West1980). There was no predominant theory guiding its development; MI has been described as quintessentially pragmatic (Carr, Reference Carr2023). Early hypotheses were operationalized and tested in clinical trials (Brown and Miller, Reference Brown and Miller1993; Miller et al., Reference Miller, Sovereign and Krege1988; Miller et al., Reference Miller, Benefield and Tonigan1993; Moyers et al., Reference Moyers, Miller and Hendrickson2005), eventually integrating Amrhein’s psycholinguistic findings (Amrhein, Reference Amrhein2004; Amrhein et al., Reference Amrhein, Miller, Yahne, Palmer and Fulcher2003). Tentative theories began to emerge (de Almeida Neto, Reference de Almeida Neto2017; Markland et al., Reference Markland, Ryan, Tobin and Rollnick2005; Miller and Rose, Reference Miller and Rose2009) along with the development of causal chain predictions of client outcomes (Magill et al., Reference Magill, Apodaca, Barnett and Monti2010, Reference Magill, Apodaca, Borsari, Gaume, Hoadley, Gordon, Tonigan and Moyers2018; Moyers et al., Reference Moyers, Martin, Houck, Christopher and Tonigan2009).

As with programmatic studies of the person-centred approach (Truax and Carkhuff, Reference Truax and Carkhuff1967), the above-described lines of research have provided increasingly clear guidelines for clinical practice of MI. Has the absence of prior theory impaired delivery and learning of MI? Although MI can be simplified conceptually (Miller and Rollnick, Reference Miller and Rollnick2023), the available evidence on training does indicate that MI can be challenging to learn, with large variability in mastery across individuals. One obstacle is that without reliable performance feedback, clinicians can substantially over-estimate their proficiency with MI, undermining motivation to continue learning (Miller and Mount, Reference Miller and Mount2001). Another challenge is difficulty in unlearning prior MI-inconsistent habits of practice (Dunn et al., Reference Dunn, Darnell, Atkins, Hallgren, Imel, Bumgardner, Owens and Roy-Byrne2016; Madson et al., Reference Madson, Loignon and Lane2009; Miller et al., Reference Miller, Yahne, Moyers, Martinez and Pirritano2004). These obstacles are not unique to MI, nor is the need for substantial time to develop mastery of a psychotherapy. Regardless of whether these are greater difficulties for MI and whether they have any connection to its lack of a theoretical foundation, the evidence that learning MI is both challenging and variable is reason enough to question whether it is indeed ‘MI’ that has spread widely to so many settings and practitioner groups. It remains unclear what elements of MI have been disseminated, and ‘the efficacy of MI approaches is unclear given the inconsistency of MI descriptions and intervention components’ (Morton et al., Reference Morton, Beauchamp, Prothero, Joyce, Saunders, Spencer-Bowdage, Dancy and Pedlar2015).

MI is just common factors

If MI has lacked theoretical grounding, could it be nothing more than general components of good practice that are sometimes referred to as ‘common’ or ‘non-specific’ factors in psychotherapy? These two terms can themselves be misleading. Allegedly ‘common’ factors are not universal practices found in all therapies or therapists. Neither does ‘non-specific’ mean that these factors are unspecifiable or unmeasurable. The meaning of both ‘non-specific’ and ‘common’ is that these practices are not unique or limited to any particular theoretical orientation in psychotherapy. Perhaps a better term, then, would be therapeutic factors (Kivlighan and Holmes, Reference Kivlighan and Holmes2004).

So, what are these skills of more effective therapists? Miller and Moyers (Reference Miller and Moyers2021) reviewed 70 years of psychotherapy research to identify therapeutic skills that distinguish clinicians whose clients show better outcomes compared with those treated by their peer practitioners working within the same setting, theoretical orientation, or delivering the same specific and even manualized treatment. These therapist factors often have substantially more impact on client outcomes than specific treatment procedures that are being delivered (Imel et al., Reference Imel, Wampold, Miller and Fleming2008; Wampold and Brown, Reference Wampold and Brown2005). Miller and Moyers identified eight such factors empirically associated with more effective therapists: accurate empathy, acceptance, positive regard, genuineness, focus, hope, evocation, and offering information and advice.

Of these eight factors, seven have been explicitly described and taught in MI since its inception. Only genuineness was unmentioned, an omission corrected in the most recent edition of the source text (Miller and Rollnick, Reference Miller and Rollnick2023). In this sense, MI does appear to embody what renders helpers more helpful, operationalizing and combining these common non-specific therapeutic skills. This is consistent with the finding in addiction research that MI can improve client outcomes when added to other evidence-based treatments (Hettema et al., Reference Hettema, Steele and Miller2005). It also suggests testable hypotheses about what is actually being ‘added’ by MI, coming full circle to the aforementioned seminal finding that therapist empathy substantially improved outcomes of cognitive behaviour therapy (Miller et al., Reference Miller, Taylor and West1980). MI calls attention to often ignored therapist factors that can improve client outcomes across a wide array of treatment methods and clinical problems.

Is MI merely a compilation of these non-specific therapeutic skills? This is a testable question. However, MI has something more in its treatment bag. Beyond the person-centred relational element of MI there is also a technical component related to client language known as change talk and sustain talk (Magill et al., Reference Magill, Apodaca, Borsari, Gaume, Hoadley, Gordon, Tonigan and Moyers2018; Miller and Rose, Reference Miller and Rose2009). Specifically training this aspect of MI in addition to the relational skills has been shown to differentially impact clients’ in-session speech that has been linked to subsequent change (Moyers et al., Reference Moyers, Houck, Glynn, Hallgren and Manual2017). The person-centred relational skills of MI can themselves influence change and sustain talk (DeVargas and Stormshak, Reference DeVargas and Stormshak2020). Three experimental trials have compared MI with a non-directive person-centred condition embodying the relational aspect of MI without seeking to evoke change talk. In two of these studies the MI condition (which included differential responding to change and sustain talk) yielded significantly greater (Sellman et al., Reference Sellman, Sullivan, Dore, Adamson and MacEwan2001) or faster change (Morgenstern et al., Reference Morgenstern, Kuerbis, Amrhein, Hail, Lynch and McKay2012) whereas the third found no 8-week difference in outcome between the directive and non-directive conditions (Morgenstern et al., Reference Morgenstern, Kuerbis, Houser, Levak, Amrhein, Shao and McKay2017).