NOTE-1456: Stockton et al. — Mechanisms of change in ACT

Contribution

It updated Hayes et al.’s 2006 summary of ACT mediation studies (LIT-555) with a systematic search, inclusion rules and a mediation-quality checklist. It confined itself to trials against an active treatment, so that a mediator could be tested for being ACT’s own. What is true after it is that the mediation evidence for ACT’s model rests mostly on overall flexibility and on acceptance. The other processes are thinly tested or untested, and the 2006 complaints about method had largely gone unheeded.

Key insight

A process that mediates an outcome in ACT and equally in the treatment it is compared with is real but not ACT’s. On that test only acceptance stands out. Defusion and committed action work, but not specially. The self as context, a process ACT names as one of six, has never been put to the test.

Assumptions

• Mediation as evidence for mechanism. Following Kazdin (2007), a mediator “may be a guide that highlights possible mechanisms of change, but is not necessarily a bone fide mechanism” (p. ms 4).
• Specificity by moderated mediation. A mediator whose path is moderated by treatment type is taken as specific to ACT. One that mediates in both arms is taken as shared with the comparison (pp. ms 8, 16–17).
• Inclusion (p. ms 10). Clinical trials of ACT against a different active treatment, 2006 to 2015, in English and peer-reviewed, using mediation analysis or testing treatment type’s interaction with the mediation path. Excluded: waitlist comparisons, overlapping arms (ACT against ACT plus something), mixed-approach interventions and theses. Searched in January 2015 in PsycINFO, Medline and Web of Science, on abstracts only.
• Quality. Rhodes and Pfaeffli’s 11-item mediation checklist (low 0–4, moderate 5–8, high 9–11), with inter-rater κ = .59 on five studies. Three statistical criteria were added: a modern technique (bootstrapping or longitudinal mediation), a reported indirect effect, and control for time and therapist (pp. ms 11–12).

Key results

• Twelve studies from the US, Sweden, the UK and South Africa, with 27 to 222 participants. Six used mediation analysis and six tested moderated mediation or mediated moderation (p. ms 12). The included studies, from the reference list: Arch et al. 2012; Bach et al. 2013; Bricker et al. 2013; Flaxman & Bond 2010; Forman et al. 2007; Forman et al. 2012; Hesser et al. 2013; Lundgren et al. 2008; Niles et al. 2014; Rost et al. 2012; Wicksell et al. 2011; Zettle et al. 2011.
• Self as context never tested. “The mediating variables investigated were limited to five of the six ACT processes … Therefore, self as context has been consistently overlooked as a potential mediator” (p. ms 12). “There were no studies that examined the ‘self-as-context’ aspect” (p. ms 18). The authors suggest this may reflect difficulty measuring it, and point to the Self Experiences Questionnaire (Yu, Norton & McCracken 2016) and to dismantling or component studies (pp. ms 18, 21–22).
• Psychological flexibility mediated mental-health outcomes in four of five studies (Niles 2014; Rost 2012; Flaxman & Bond 2010; Forman 2007). Evidence was limited for physical health and functioning and mixed for quality of life (p. ms 14). In the discussion it “strongly mediated” mental health, functioning and physical symptoms (p. ms 17).
• Acceptance mediated mental-health outcomes in four of five studies (Hesser 2013; Forman 2012; Rost 2012; Wicksell 2011), and functioning in two. There was some evidence for quality of life and behaviour (Arch 2012; Bricker 2013). It was “the only process for which mediation/moderated mediation was found across type of outcome”, and is called “a primary mechanism of change during ACT” (pp. ms 14, 16–17).
• Defusion. Of eight mediation analyses in four studies, seven found mediation, but four of five moderated-mediation analyses found it not specific to ACT. The two studies that found nothing used non-standardized measures, while Bach et al. (a behavioural outcome, rehospitalization) and Zettle et al. (a validated measure) were positive (pp. ms 14–15, 17). The results section and discussion state this differently. The results say three of four studies found no mediation for mental-health outcomes. The discussion says only one of eight analyses failed. The tables that would reconcile them are not in the deposit.
• Committed action mediated in both ACT and cognitive therapy in two studies, so it is not distinct to ACT. Present-moment contact (Forman 2007) and values (Lundgren 2008) were tested in one study each (p. ms 15).
• Non-ACT processes. Change in dysfunctional cognitions, tested in seven studies, did not consistently mediate. Symptom frequency, symptom distress, self-efficacy and pain intensity did not mediate (p. ms 15).
• Quality (pp. ms 13, 18–19). All twelve were “moderate quality”. Only Lundgren et al. 2008 was adequately powered, and the rest did not report power. Only three studies (Hesser 2013; Arch 2012; Rost 2012) “ascertained whether changes in mediator variables preceded changes in outcome”. Two studies measured the mediator once (Lundgren 2008; Zettle 2011). Only Forman 2007 and 2012 accounted for both time and therapist. Reliability went unreported for five mediator and seven outcome measures. “Studies were over dependant on self-report measures.”
• Earlier reviews, as reported (pp. ms 4–5). Hayes et al. 2006 summarized eight mediation studies with weaknesses that included “process measures being taken after outcomes have significantly improved”. Hayes, Villatte, Levin and Hildebrandt 2011 found about half of between-group differences at follow-up explained by post-treatment flexibility.

Claims

id	claim	strength	support
C1	No ACT mediation study from 2006 to 2015 tested the self as context	strong as a report of the search	the review’s inclusion set (p. ms 12, 18)
C2	Overall psychological flexibility mediates ACT’s effects on mental-health outcomes	moderate	four of five studies; moderate quality; measures not stated in the prose
C3	Acceptance is a mechanism of change specific to ACT	weak to moderate	mediation in four of five studies for mental health, across outcome types, moderated by treatment; temporal order shown in three studies overall
C4	Defusion and committed action mediate outcomes but are not specific to ACT	weak	few studies; the review’s two accounts of defusion differ
C5	The ACT mediation literature has not met the methodological standards Hayes et al. 2006 set	strong as a report	quality assessment

Concepts

• Mediator, mechanism, moderator — after Kazdin (2007): a mediator “accounts (statistically)” for the treatment–outcome relation; a mechanism is the process “responsible for therapeutic change”; a moderator changes the relation’s direction or size (p. ms 2).
• Moderated mediation and mediated moderation — treatment type moderating the mediator–outcome path only, or the treatment–mediator path (p. ms 8).
• Psychological flexibility (hexaflex) — defusion, acceptance, contact with the present moment, values, committed action and self as context, this last glossed as “development of an non-evaluative, observing self” (p. ms 3).

Connections

It is the systematic successor to the mediation section of Hayes et al. 2006 (LIT-555, read in NOTE-431 from an early manuscript). It cites Levin et al.’s 2012 meta-analysis of laboratory component studies for acceptance’s large effect against inactive conditions. It recommends the MPFI (Rolffs, Rogge & Wilson 2016), the multidimensional flexibility measure from Rogge’s group, whose 2024 meta-analysis (LIT-543) is the one the record’s ACT accounts name. Its search ends in January 2015, so it predates that work.

Bearing on the record

• THEORY-050: partly, and not what its promote_when asks. The account’s test (1) is “a first-hand reading of the 2024 meta-analysis … showing that change in avoidance, fusion or flexibility precedes and predicts change in outcome across more than one kind of disorder”. This is a different and earlier review. It finds flexibility and acceptance mediating mental-health outcomes across anxiety, depression, social anxiety, tinnitus, chronic pain and cancer adjustment. But it shows the mediator changing first in only three studies. It says the studies relied on self-report, and its prose does not say which measures were used, so it cannot show that they were not the global AAQ that promote_when excludes. It also finds defusion, the account’s “fusion”, not specific to ACT. So it supports the account’s process claim as mediation without temporal order. That is the state THEORY-053 calls the right kind of evidence with the order not established, and THEORY-050 already records that state from Hayes et al. 2006. It does not meet the condition.
• THEORY-051: not at all, and it confirms the gap. The account waits on a study that changes or measures the self as context apart from defusion and acceptance. This review found none: “self as context has been consistently overlooked as a potential mediator”. It suggests a measure (the SEQ) and dismantling designs, which is the test the account’s promote_when already names. Nothing here bears for or against it.
• LIT-543 stays the test both accounts name. This review is not a substitute for it.
• No instruction for machine-learning practice; nothing for the anthology.

Limitations

• The tables are missing from the copy read. Study-by-study methods, measures, outcomes and quality scores (Tables 2–4, Appendix 1) were not seen. Claims about which measures were used cannot be checked here.
• Narrative synthesis. There is no pooled estimate and no count of null analyses beyond the prose, and the two accounts of defusion differ.
• Narrow window and search. 2006 to 2015 only, abstracts only, three databases, English only.
• Inclusion by active comparator answers specificity but leaves out waitlist trials, which the authors grant would have widened the evidence (p. ms 20).
• Checklist reliability was moderate (κ = .59), and the authors say its statistical item is too broad (p. ms 19).

Open questions

• Does the self as context mediate outcome when measured apart from defusion and acceptance, for example with the SEQ, or isolated in a dismantling design?
• Does acceptance change before outcome in more than the three studies here, and with measures other than global self-report?
• What does Macri and Rogge’s 2024 meta-analysis (LIT-543), with the MPFI domains, add to this?