The Asymmetric Correction Hypothesis

Attachment Deactivation, Corrective Feedback Insulation, and When Clinical Deference Becomes Iatrogenic

Author: Mik Idrizovic
Original date: July 2026
Revised: August 2026
Status: Hypothesis-generating theoretical proposal; no empirical work is reported; not clinical guidance

No empirical work is reported in this paper. The framework is offered as a falsifiable research proposal and should not be treated as clinical guidance.

Abstract

Attachment insecurity is commonly represented by two continuous dimensions, anxiety and avoidance. They are often treated as parallel deviations from security, one organized around hyperactivation and the other around deactivation. This paper proposes that they may be nonparallel in a narrower respect that matters for change: attachment strategies may differ in the probability that the costs they generate return to the person using them as information that is noticed, attributed to the strategy, retained, and acted upon.

The proposal is named the Asymmetric Correction Hypothesis (ACH). Its organizing construct is Corrective Feedback Insulation (CFI), defined as the multilevel set of processes that reduces the probability that strategy-generated costs become usable corrective information. CFI is not initially proposed as a unitary personality trait. Its candidate routes occur at different causal levels and need not covary: intrapersonal signal attenuation, dyadic cost displacement, corrective-system underexposure, attributional neutralization, and episode reset. The first three routes have meaningful empirical support in adjacent attachment literatures. The latter two remain comparatively speculative and are retained as explicit, separately falsifiable hypotheses rather than treated as established components.

Existing evidence supports several links in the proposed architecture. Avoidance is associated with reduced attention to emotional information, selective limitations in encoding and memory for attachment-relevant material, lower vulnerable disclosure, reduced mental-health-service engagement, and weaker initial therapeutic alliance. Anxiety is generally associated with greater attention to attachment-relevant information, greater service engagement, and a more legible distress signal. Treatment studies do not establish a universal advantage in healing for either dimension. Avoidance can change, sometimes substantially, and highly avoidant clients can benefit from treatment. The strongest defensible claim is therefore not that avoidant people are less capable of revision. It is that deactivating organization may make entry into, exposure to, and uptake of correction unusually conditional on measurement, relationship process, intervention form, and delivery style.

The paper also argues that clinical deference can become iatrogenic under specific conditions. Respect for autonomy is not the target. The target is the recurrent clinical pull to treat withdrawal as a self-interpreting instruction to reduce relational depth, disclosure demands, or feedback. Where withdrawal is defensive rather than protective, and where compliance further reduces corrective exposure, deference may participate in the maintenance loop it intends to respect. The proposed alternative is not confrontation and not engineered rupture. It is accountability under protected repair conditions: descriptive naming of impact, preservation of autonomy, sustained availability, monitoring of activation, and an explicit commitment to repair.

ACH is situated relative to Bowlby’s defensive exclusion, epistemic trust, self-verification, interpersonal complementarity, safety behavior, family accommodation, cognitive immunization, demand-withdraw dynamics, and alliance rupture-repair. Its proposed novelty lies not in discovering any single component, but in integrating them into a multilevel theory of how corrective information can be lost between the generation of relational cost and the revision of a working model. The paper supplies an evidence ladder, revised hypotheses, construct-development standards, a staged empirical program, falsification criteria, and strengthened safeguards against coercive or partner-led weaponization.

Keywords: adult attachment; attachment avoidance; attachment anxiety; deactivating strategies; corrective feedback; defensive exclusion; treatment engagement; therapeutic alliance; disclosure; demand-withdraw; rupture and repair; working-model change

1. Introduction: The Signal Must Return

Change requires more than pain. It requires information.

A strategy can impose substantial costs and still remain stable if those costs do not return to the person using it in a form they can recognize, attribute, remember, and use. Two people can suffer equally while receiving radically different amounts of actionable evidence about the source of their suffering. One enters a corrective sequence. The other remains outside it.

This paper asks whether the two major dimensions of adult attachment insecurity differ along that pathway. Hyperactivating strategies amplify attachment-relevant signals. They intensify vigilance, sustain proximity-seeking, and make distress conspicuous to the person, to partners, and often to clinicians. Deactivating strategies reduce attention to attachment-relevant information, constrain disclosure and dependence, and suppress or redirect attachment activation under threat (Mikulincer & Shaver, 2016). The distinction is familiar. The proposed asymmetry is not.

The Asymmetric Correction Hypothesis states that deactivating strategies may generate greater Corrective Feedback Insulation than hyperactivating strategies. CFI is the multilevel reduction in the probability that the costs of a strategy become usable evidence against the working model sustaining it. In schematic form, correction requires a chain:

cost generation → transmission → attentional capture → encoding and retention → attribution → interpersonal processing → action → revision

CFI can operate at any link. The cost may be generated but borne primarily by another person. Feedback may be transmitted but not disclosed, trusted, or encoded. It may be encoded but attributed to an incompatible partner, an intrusive clinician, or a culture that misunderstands independence. Treatment may be available but never entered. An alliance may form but remain too shallow for the relevant material to become discussable. A relationship may end, allowing the evidence to be reset before it accumulates.

The chain is presented as a heuristic, not as a claim that its links are statistically independent. Its value is analytic. It prevents the paper from treating “feedback” as a single event and asks where, exactly, correction fails.

The argument is deliberately narrower than a claim about moral character, suffering, or intrinsic capacity. It does not propose that avoidant people suffer less, care less, or cannot change. It does not propose that anxious behavior is benign. Hyperactivation can be coercive, destabilizing, exhausting, and self-perpetuating. Nor does it propose a universal difference in psychotherapy outcome. The existing literature does not support one.

The strongest current formulation is this: anxiety and avoidance may differ in how reliably they enter the conditions under which revision becomes possible. Anxiety is often costly in ways that are felt, visible, reportable, and actionable. Avoidance can be costly while remaining selectively gated, underdisclosed, relationally displaced, or underrepresented in the systems designed to deliver correction. Once avoidant organization is reached under the right conditions, it can move.

That distinction matters. A theory of incapacity predicts failure even under adequate access. A theory of insulation predicts heterogeneity. High-avoidance individuals with low insulation should change readily; high-avoidance individuals with multiple active insulation routes should not. Intervention effects should depend less on the intensity of challenge than on whether feedback becomes usable without being converted into evidence of intrusion. Measurement modality, treatment entry, disclosure, alliance trajectory, partner behavior, and cultural attribution should therefore matter more than a simple attachment label.

The paper also advances a conditional claim about clinical deference. Clinicians appropriately respect autonomy, consent, pacing, and requests for distance. Those principles are not being challenged. The concern is a recurrent interactional pull: a deactivating presentation may invite reduced depth, lower demands for elaboration, fewer direct observations, or premature agreement that distance is the complete meaning of the behavior. Compliance then appears neutral because tension falls. Yet if the withdrawal is part of the maintaining process, reducing corrective contact may strengthen the process.

“Deference becomes iatrogenic” is therefore not a general verdict on giving space. It is a causal hypothesis with strict boundary conditions. The request must function defensively rather than protectively; the accommodating response must reduce access to relevant information; a safer autonomy-preserving alternative must exist; and the accommodation must predict poorer subsequent engagement or change. Without those conditions, the claim fails.

The central scientific question is not whether avoidant people need harsher consequences. It is how accurate information about the costs of deactivation can remain reachable long enough to support revision without reproducing the intrusion, coercion, or abandonment against which deactivation is organized.

2. Construct Alignment: Attachment Style Is Not One Measurement Tradition

Adult attachment research contains several related but noninterchangeable constructs. Infant attachment classifications, Adult Attachment Interview states of mind, self-reported romantic attachment dimensions, and relationship-specific attachment representations do not measure the same object. The two-dimensional anxiety-avoidance structure is well supported within romantic self-report measurement, but dimensional scores should not be mistaken for natural kinds (Brennan, Clark, & Shaver, 1998; Fraley & Waller, 1998; Fraley, Waller, & Brennan, 2000). The weak convergence between interview and self-report traditions is not a technical footnote. It is a constraint on every claim in this paper (Roisman et al., 2007).

The Adult Attachment Interview assesses the organization of discourse about childhood attachment experiences. Dismissing states of mind are characterized by idealization, derogation, or restricted access to attachment-relevant memory and affect. Romantic self-report measures, including the Experiences in Close Relationships family, assess conscious expectations and behavior in adult close relationships along dimensions of anxiety and avoidance. A person can score high in romantic avoidance without receiving a dismissing AAI classification, and the reverse. Relationship-specific avoidance can also diverge from global avoidance (Overall, Fletcher, & Friesen, 2003). Attachment orientations are moderately stable rather than fixed, and within-person change coexists with substantial rank-order continuity (Fraley, Vicary, Brumbaugh, & Roisman, 2011).

Altmann et al. (2021) make the methodological problem concrete. In 148 SOPHO-Net patients, an observer-rated attachment prototype and a self-report classification showed almost no agreement, Cohen’s κ = .08, and yielded different clinical associations. That finding supports measurement dependence. It does not, by itself, prove that self-report directionally minimizes avoidance. ACH therefore separates cross-modal divergence from the stronger claim of directional signal attenuation.

Accordingly, this paper uses three terms with distinct functions:

1. Attachment avoidance refers to a score on dimensional adult attachment measures unless another method is specified.

2. Dismissing attachment representation refers to an interview-derived state of mind.

3. Deactivating process refers functionally to attention, memory, affect regulation, disclosure, or interpersonal behavior that reduces activation or accessibility of attachment-relevant material.

ACH is primarily a theory of deactivating processes. Attachment avoidance and dismissing representation are treated as imperfect indicators of those processes, not interchangeable diagnoses. Every empirical study should specify which tradition it uses, which relationship is being assessed, and whether the proposed mechanism is inferred from self-report, behavior, interview, physiology, or dyadic observation.

This alignment also changes the novelty claim. Bowlby’s concept of defensive exclusion already proposed that attachment-relevant information can be excluded from awareness when it threatens an established working model (Bowlby, 1980). Later work has examined attentional inhibition, encoding limitations, thought suppression, and physiological-verbal discrepancies. ACH does not discover filtering. Its contribution is to connect intrapersonal filtering to dyadic, clinical, and contextual pathways through which the costs of a strategy may fail to return as corrective information.

The distinction between description and mechanism remains load-bearing. “Avoidant” names a region in a dimensional space or a pattern in a particular measurement system. It does not specify where corrective uptake fails. Two equally avoidant people may differ sharply: one may feel distress but refuse disclosure; another may report little distress but show attentional avoidance; another may disclose in treatment but elicit a complementary clinical retreat; another may be fully reachable in one relationship and highly insulated in another.

A useful theory must explain that heterogeneity. CFI is proposed to do so.

3. The Core Proposal: Corrective Feedback Insulation as a Multilevel Architecture

3.1 Definition

Corrective Feedback Insulation is defined as the multilevel reduction in the probability that strategy-generated costs become information that the person can perceive, attribute to the strategy, retain, discuss, and translate into behavioral or representational revision.

CFI is not initially modeled as a reflective latent trait. The proposed routes live at different levels of analysis, have different causes, and need not be interchangeable manifestations of one hidden essence. A person may show strong attentional attenuation but high disclosure, or low treatment entry but accurate interoceptive awareness, or substantial dyadic cost displacement only in one relationship. Forcing those routes into a conventional reflective scale would assume that they covary because CFI causes them. That assumption is premature.

The stronger initial representation is a multilevel causal architecture or route profile. Each route should be measured separately. Only after longitudinal evidence shows that the routes jointly and incrementally predict corrective uptake should researchers consider a formative index or higher-order causal composite (Bollen & Bauldry, 2011; Diamantopoulos & Siguaw, 2006; Jarvis, MacKenzie, & Podsakoff, 2003).

3.2 The five candidate routes

Route 1: Intrapersonal signal attenuation

Attachment-relevant information may receive less attention, weaker encoding, reduced accessibility, delayed affective recognition, or a discrepant verbal readout. “Signal attenuation” is intentionally broader than “interoceptive suppression.” The evidence does not support the claim that avoidant individuals always generate a full internal signal and merely fail to feel it. Some studies suggest cross-channel dissociation; others suggest effective attentional or physiological down-regulation. The route concerns selective reduction in corrective capture, whatever the exact channel.

Route 2: Dyadic cost displacement

The relational costs of withdrawal may be disproportionately carried, interpreted, or managed by a partner. The hypothesis is not that avoidant people experience no cost, that anxious partners alone suffer, or that withdrawal always creates greater power for the withdrawer. The proposed asymmetry concerns visibility and return: a cost carried by another person may fail to become evidence for the person whose strategy contributed to it.

Route 3: Corrective-system underexposure

A person may be less likely to seek help, participate deeply, disclose vulnerable material, accept assistance, or remain in a sufficiently engaged alliance. The result is not failed treatment in the ordinary sense. It is reduced exposure to the conditions under which treatment outcome can be observed at all.

Route 4: Attributional neutralization

Corrective information may be redescribed, reattributed, or cognitively immunized before it challenges the working model. “Semantic laundering” is retained as a vivid shorthand for one possible subroute, especially when unannounced withdrawal is relabeled as self-care or boundary-setting without preserving the relational distinctions those concepts require. Direct evidence for a culture-wide valence asymmetry is currently weak. Attributional neutralization is therefore a hypothesis, not an established property of avoidance.

Route 5: Episode reset

Relationship termination, partner substitution, or a global “wrong fit” attribution may prevent failures from accumulating as evidence about a recurring strategy. This route was previously described as low exit cost. The revised term is narrower and more testable. Exit cost is not inherently low for avoidant individuals, and substitution dynamics are not unique to avoidance. The hypothesis concerns whether a new relational episode resets attribution before cross-episode regularities become visible.

3.3 Evidence status

Table 1

Evidence status of candidate CFI routes

The table is deliberately uneven. A theory becomes stronger when its speculative limbs are labeled as such. ACH does not require every route to survive. It requires at least one route-specific mechanism to predict corrective uptake beyond attachment avoidance itself, and it requires the proposed asymmetry to withstand direct comparison with anxiety.

4. Evidence Ladder

4.1 Intrapersonal signal attenuation

Bowlby’s defensive exclusion is the closest conceptual ancestor of the intrapersonal route. Information incompatible with an attachment strategy may be prevented from gaining full access to awareness or integrated memory. Modern findings support several forms of selective processing without establishing a single universal mechanism.

Peng et al. (2024) meta-analyzed 68 studies comprising 5,417 participants. Attachment avoidance was associated with decreased attention to emotional stimuli, whereas attachment anxiety was associated with increased attention, especially under stress, for attachment-relevant material, and during later stages of processing. The avoidant effect was small, d = -.129, but directionally stable across the tested moderators. This is precisely the kind of modest, distributed effect that could matter cumulatively without licensing a deterministic portrait.

Encoding and memory work offers a more attachment-specific signal. Fraley and Brumbaugh (2007) found that poorer recall of attachment-relevant information among avoidant adults was better explained by defensive processes operating at encoding than by retrieval alone. Monetary incentives did not restore the missing information. Edelstein (2006) similarly reported avoidance-linked working-memory deficits for attachment-related material but not for nonattachment emotional or neutral material. Edelstein and Gillath (2008) found reduced emotional Stroop interference for attachment-related words among avoidant adults, with the effect attenuated under cognitive load, consistent with an effortful inhibitory process.

Thought-suppression experiments complicate any simple claim that distress remains fully present behind a disconnected readout. Fraley and Shaver (1997) found that dismissing individuals could suppress attachment-related thoughts, and some physiological activation decreased during suppression. Mikulincer, Dolev, and Shaver (2004), by contrast, showed rebound effects and increased accessibility of vulnerable self-representations under cognitive load. Kohn, Rholes, and Schmeichel (2012) likewise found that self-regulatory depletion increased access to negative attachment memories among avoidant participants.

The psychophysiological literature therefore supports a narrower proposition than the original manuscript stated. Deactivating presentation can coexist with autonomic stress, reduced attention, restricted memory, or cross-channel discrepancy. Dozier and Kobak (1992) found heightened skin conductance during attachment-relevant portions of the Adult Attachment Interview among deactivating participants despite minimizing discourse. Roisman, Tsai, and Chiang (2004) found related mismatches across physiological, facial, and self-reported response. Yet systematic review indicates heterogeneity across measures, tasks, and attachment classifications (Eilert & Buchheim, 2023). Dewitte, De Houwer, Goubert, and Buysse (2010) also found that anxiety more consistently predicted cortisol reactivity and subjective distress, while avoidance appeared more clearly in subjective or behavioral channels and was moderated by partner attachment.

The strongest defensible conclusion is not that avoidant people lack access to all distress. It is that deactivating organization can reduce, delay, or selectively gate attachment-relevant information, and that the affected channel varies across persons and contexts.

ACH adds a downstream prediction. Signal attenuation should matter only if it reduces correction. Attention, memory, or physiology must therefore be linked prospectively to what happens after feedback: recall of the partner’s account, accuracy of impact attribution, willingness to discuss it, retention across time, and behavioral revision. A processing difference without a correction consequence supports deactivation theory but not CFI.

4.2 Corrective-system underexposure

The strongest current evidence for asymmetry appears upstream of treatment outcome.

Adams, Wrath, and Meng (2018) systematically reviewed 18 studies of adult attachment and mental-health-care utilization. Across heterogeneous populations and services, attachment anxiety was generally associated with greater engagement and participation, while avoidance was associated with less. Findings for treatment completion were less conclusive. This pattern does not prove that avoidance causes underutilization, and structural access, diagnosis, socioeconomic status, and prior treatment must be modeled. It nevertheless places the two dimensions on opposite sides of the treatment-entry and participation process across a literature broader than the original Dozier program.

Disclosure provides another route. Saypol and Farber (2010) found that security of attachment to the therapist was positively associated with overall disclosure and more positive feelings after disclosure among 117 psychotherapy patients. Emery, Gardner, Carswell, and Finkel (2018) found across five studies that attachment avoidance predicted lower self-concept clarity through reduced self-verification from partners; less disclosure and less trust in partner feedback helped explain the pathway. Sun and Jakubiak (2024), using event-level and relationship-specific measurement, found that avoidance predicted a lower probability of sharing events and more selective disclosure favoring positive, competence-signaling, less vulnerable material. The exact magnitude varied by global versus relationship-specific avoidance, which supports the need to measure the active relational representation rather than assume a fixed person-level style.

These findings sharpen CFI. The problem may not be only whether feedback is offered. It may be whether enough of the relevant internal and interpersonal context enters the room for feedback to become accurate. A clinician or partner responding to a strategically thinned account can be empathic and still be correcting the wrong model.

The Dozier program supplies direct clinical observations. Dozier (1990) found that greater deactivation predicted reduced help-seeking and more rejection of help in adults with serious psychopathology. Dozier, Cue, and Barnett (1994) found that case managers attended more to dependency needs and intervened more deeply with preoccupied than dismissing clients. Dozier, Lomax, Tyrrell, and Lee (2001) described restricted engagement and confusion in clinical exchanges with dismissing clients. These studies are influential but concentrated in particular populations and one research lineage. They should motivate replication, not function as universal law.

The updated alliance literature also corrects an overstatement. Notsu, Blansfield, Spina, and Levy (2025) meta-analyzed 33 studies and found small negative associations between client-rated alliance and both anxiety, r = -.09, and avoidance, r = -.13. Avoidance was somewhat more strongly related to alliance, but the difference was modest. ACH should therefore not rest on a large alliance-strength asymmetry once treatment has begun.

The more interesting signal is trajectory. Igra et al. (2026) analyzed 549 clients treated by 53 therapists. Higher pretreatment avoidance predicted lower initial alliance and was marginally associated with steeper early alliance growth. Steeper alliance growth across the first four sessions predicted greater subsequent improvement in avoidance, with no analogous association for anxiety. The design was naturalistic and does not establish mediation. Still, it suggests that an initially weak alliance is not destiny and that avoidant change may depend on whether the relationship becomes progressively usable.

Corrective-system underexposure thus has at least four separable components: entry, participation, disclosure, and relational usability. Lumping them into a single self-report score would obscure the mechanism. Each should be measured directly.

4.3 Dyadic cost displacement

Demand-withdraw research provides clear evidence that withdrawal has relational consequences. Foundational work located the pattern partly in social structure and in asymmetries over who seeks change (Christensen & Heavey, 1990). Meta-analytic work links demand-withdraw patterns with lower relational satisfaction and poorer communicative outcomes (Schrodt, Witt, & Shimkowski, 2014). Longitudinal studies find that the pattern predicts later dissatisfaction beyond negative communication generally (Caughlin & Huston, 2002). Attachment anxiety often corresponds to the demanding position and avoidance to the withdrawing position, but the mapping is approximate. Demand and withdrawal are relational roles shaped by topic, power, gendered expectations, desired change, and situational incentives. They are not attachment diagnoses.

Power-dependence accounts and the principle of least interest have been read as implying a general power advantage for avoidant partners (Waller, 1938; Emerson, 1962). The dyadic evidence does not support that reading. Körner et al. (2026), across five studies and 1,256 dyads, found that both anxiety and avoidance were associated with lower actor power. Avoidance was also associated with lower partner power, and avoidant actors withdrew more in conflict and daily life; withdrawal predicted lower partner power. The pattern is not “the avoidant partner wins.” It is closer to a relational field in which withdrawal can reduce the partner’s agency while not conferring subjective power or wellbeing on the withdrawer.

That distinction strengthens the humane interpretation of ACH. Dyadic externalization is not an advantage awarded to one side. It is a failure of information routing. The anxious partner may carry overt uncertainty, protest, and interpretive labor. The avoidant partner may carry loneliness, reduced mattering, self-concept costs, and attachment activation that is underprocessed or unshared. Both can lose while only one receives a legible indictment of the strategy.

The empirical target should therefore be cost distribution and visibility rather than a priori claims about who suffers more. Ecological momentary assessment can ask, after a withdrawal episode: Who experiences distress? Who understands its source? Who attempts repair? Who reports the episode to others or to a clinician? Whose interpretation becomes the dominant account? Does the withdrawing partner remember the partner’s reported impact days later? Does physiological activation converge with self-report? Does the episode alter future behavior?

Partner effects are central. Emery et al. (2018) show that reduced trust in partner feedback can deprive avoidant individuals of self-verifying information. Ybarra and Seedall (2024) found actor and partner links between avoidance, lower perceived mattering, communication, and wellbeing. These are not direct tests of CFI, but they establish that attachment organization changes the informational environment of both members of the dyad.

The route should be considered supported only if partner-borne costs fail to return as accurate, retained, strategy-attributed feedback. Partner distress by itself is not CFI. The theory is about the missing return path.

4.4 Clinical complementarity and feedback dose

Attachment patterns can recruit responses that preserve themselves. In interpersonal terms, the client’s presentation may pull the clinician toward complementarity: distance for the distancing client, intensified involvement for the preoccupied client. Dozier and Tyrrell (1998) argued that such responsiveness can be clinically counterproductive. Tyrrell, Dozier, Teague, and Fallot (1999) found stronger alliances under certain noncomplementary client-clinician pairings among people with serious psychiatric disorders. Bernier and Dozier (2002) framed noncomplementarity as part of a corrective emotional experience.

This line of work is important prior art, but it is not settled. The samples are specialized, replication is limited, and broader reviews of therapist attachment and treatment process remain heterogeneous (Degnan, Seymour-Hyde, Harris, & Berry, 2016). The updated systematic review by Horne, Liu, and Aafjes-van Doorn (2024) identified 42 empirical studies of therapist attachment or introject. Only two were rated moderate in quality and 40 were rated weak. More secure therapist attachment appeared more consistently related to alliance and process than to patient outcome. A preliminary dyadic study by Marmarosh et al. (2014) also suggests client-therapist attachment combinations matter, but it does not justify a universal matching rule. The clinician side of CFI is plausible and empirically thin, exactly the kind of claim that needs direct behavioral replication rather than inheritance by citation.

The strongest claim available is interactional rather than prescriptive. Dismissing or avoidant presentation can alter the depth, style, and timing of clinical response. Whether that response is maintaining or corrective depends on what it does to feedback dose, disclosure, alliance growth, and the client’s experience of autonomy.

Emotionally focused couple therapy offers a mechanistically aligned example. Withdrawer re-engagement treats withdrawal as a relational process to be approached, not as a self-interpreting instruction for the therapist to disengage. Myung et al. (2022) described the emotional architecture of withdrawer re-engagement and blamer-softening change events. EFT outcome evidence is meaningful but should not be oversold. Rathgeber, Bürkner, Schiller, and Holling (2019) found medium effects for both behavioral couple therapy and emotionally focused couple therapy across randomized trials, with no clear superiority and concerns about publication bias and long-term evidence. Spengler, Lee, Wiebe, and Wittenborn (2024) reported substantial aggregate effects across controlled and uncontrolled EFCT studies, but the evidence base includes heterogeneous designs. EFT is one of the better-supported couple therapies, not a settled winner.

Delivery style is the load-bearing moderator. Overall, Simpson, and Struthers (2013) found that avoidant targets of influence showed greater anger and withdrawal and less successful conflict discussions. Within the same people, those responses were attenuated when partners used softening strategies that conveyed value and respected autonomy. Girme, Overall, Simpson, and Fletcher (2015) likewise found nonlinear effects of support for avoidant recipients: low or moderate support could produce worse responses, while high support under some conditions functioned as unusually clear evidence of availability. Neither study licenses indiscriminate intensity. Together they show that corrective input can fail when it is weak enough to remain ambiguous or forceful enough to confirm intrusion.

CFI therefore predicts an interaction, not a main effect of confrontation. Naming impact should improve corrective uptake only when paired with autonomy preservation, sustained availability, and repair capacity. Naming without those conditions should increase deactivation. Pure deference may reduce immediate tension while decreasing feedback dose. The optimum is neither pressure nor disappearance. It is high-fidelity information delivered in a relationship that remains safe enough to process it.

4.5 Attributional neutralization and episode reset

The cultural and attributional routes are the least established parts of the theory.

Negative attribution bias is associated with both anxiety and avoidance, with medium meta-analytic effects (Li, Carnelley, & Rowe, 2023). This is a warning against claiming that avoidant organization uniquely reattributes relational difficulty. Anxiety can also neutralize correction through catastrophic interpretation, selective attention, reassurance cycles, and protest behavior. ACH requires route specificity, not a morality play in which only one dimension distorts evidence.

Cognitive immunization offers useful adjacent language. Festinger’s (1957) dissonance framework already predicts that contradiction can be reduced through routes cheaper than revision of a core belief. In depression research, disconfirming positive information can be discounted or reinterpreted so that negative expectations remain intact (Kube et al., 2019, 2022). Findings are not perfectly consistent across samples, which is exactly why immunization should be treated as a testable process rather than assumed. The attachment analogue would be a shift from “my withdrawal contributed to this outcome” to “the partner was too demanding,” “the clinician was intrusive,” or “the relationship was wrong.” Those attributions may sometimes be correct. The scientific task is to identify rigid, cross-contextual reattribution that persists despite independently observed regularities.

The “semantic laundering” claim requires especially careful handling. There is good reason to suspect that identical behaviors receive different moral descriptions depending on whether they are framed as independence, boundaries, neediness, or pursuit. Haslam’s (2016) concept-creep account and sociological analyses of individualism and detachment provide framing, not direct evidence (Bellah et al., 1985; Illouz, 2007, 2019). Cultural-fit findings are mixed. Friedman et al. (2010) found that avoidance was more strongly associated with relationship problems in more collectivist settings than in the United States, consistent with context-dependent penalization. Other work does not find individualism to moderate all avoidance-related symptom associations. The available evidence supports cultural contingency, not the broad statement that contemporary culture uniformly congratulates avoidance.

The revised construct is attributional neutralization. Semantic laundering is one possible observable form. It should be tested through preregistered content analysis and experiments in which identical patterns of unannounced, intimacy-contingent withdrawal are randomly labeled as “boundary-setting,” “self-care,” “avoidance,” or neutrally described behavior. Outcomes should include moral evaluation, perceived responsibility, recommended partner response, and predicted need for change, crossed with actor gender, relationship type, and attachment profile.

Episode reset is similarly unproven. Investment-model research shows that satisfaction, alternatives, and investment shape commitment and persistence, and attachment effects operate partly through those variables (Etcheverry, Le, Wu, & Wei, 2013). Anxiously attached individuals can also use new romantic attention to disengage from former partners (Spielmann, MacDonald, & Wilson, 2009). Partner substitution is therefore not an avoidance-specific mechanism.

The ACH-specific prediction is narrower: when a relationship ends, do people attribute the failure to a recurring strategy or reset the explanatory count at zero? A prospective design would follow participants across relationships, assess narrative attribution after each dissolution, and test whether repeated cross-partner feedback predicts change only when episodes are mentally linked. Until such data exist, episode reset remains a candidate contextual moderator, not a core established route.

5. Adjacent Prior Art and the Precise Novelty Claim

ACH occupies a crowded conceptual neighborhood. The paper is stronger when it names the neighbors and states exactly what remains unbuilt.

5.1 Defensive exclusion

Bowlby’s defensive exclusion is direct prior art for selective inaccessibility of attachment-relevant information. The attentional, encoding, memory, and suppression findings reviewed above elaborate that tradition. ACH should not claim novelty for the proposition that deactivation filters attachment information. Its extension is multilevel: the same corrective sequence can be interrupted not only within cognition, but through disclosure, partner cost distribution, treatment entry, clinician response, attribution, and relationship discontinuity.

5.2 Epistemic trust and vigilance

Epistemic trust concerns openness to the authenticity and personal relevance of socially transmitted knowledge (Fonagy, Campbell, & Bateman, 2017). It offers a close account of why accurate information may not become learnable merely because another person says it. ACH overlaps wherever corrective feedback depends on trusting a partner or clinician as a source.

The constructs are not identical. Epistemic trust concerns receptivity to social communication broadly. CFI concerns the probability that costs generated by a strategy return through the entire correction chain. A person may trust a clinician generally while withholding the precise material needed for correction. Another may hear the material but fail to encode it. Another may encode it and attribute it elsewhere. Epistemic trust is therefore a likely mediator or moderator within CFI, not a synonym.

The Epistemic Trust Rating System provides useful measurement precedent because it codes sharing, “we-mode” interaction, and receptiveness to learning rather than relying entirely on self-report (Fisher, Guralnik, Fonagy, & Zilcha-Mano, 2025). In an initial validation sample of 118 psychotherapy patients, the observer-rated system showed strong interrater reliability and modest convergent associations. It is promising rather than definitive, but its architecture is directly relevant: ACH should build on observer-coded transmission and receptivity rather than asking participants to certify their own insulation.

5.3 Self-verification and partner feedback

Self-verification theory treats close partners as important sources of identity-relevant feedback. Emery et al. (2018) show that avoidance is associated with less disclosure, lower trust in partner feedback, reduced self-verification, and lower self-concept clarity. This is close prior art for the dyadic route. ACH extends the question from verification of identity to correction of strategy and predicts that partner feedback should matter most when it is accurately transmitted, trusted, remembered, and linked to behavior.

5.4 Safety behaviors, experiential avoidance, and accommodation

Clinical theories of safety behavior provide a strong analogue. Behaviors that reduce immediate threat can prevent disconfirmation and maintain anxiety, although contemporary reviews emphasize that safety behaviors are not uniformly harmful and can sometimes support graded engagement (Blakey & Abramowitz, 2016; Goodson et al., 2026). The lesson is not “avoidance must always be blocked.” It is that the function and timing of protection determine whether it preserves fear or enables learning.

Family accommodation offers another analogue. Accommodation reliably covaries with symptom severity in obsessive-compulsive disorder, and it often decreases during effective treatment. Yet a 2024 meta-analysis found that baseline accommodation did not predict symptom change across treatment (Hermida-Barros et al., 2024). This is an important gotcha. Correlation with a maintaining system does not establish that reducing the accommodating behavior causes improvement.

ACH must meet the same burden. It is not enough to show that clinicians give more distance to avoidant clients or that partners absorb uncertainty. The theory must show that these responses reduce corrective uptake and that changing them, under safe conditions, improves the relevant outcome.

Experiential avoidance is broader still. It concerns attempts to alter or escape unwanted private events and has extensive transdiagnostic relevance. CFI should demonstrate incremental validity beyond experiential avoidance, alexithymia, self-concealment, reflective functioning, social desirability, and distress severity. Otherwise, it risks renaming a general avoidance process with attachment vocabulary.

5.5 Cognitive immunization and motivated reasoning

Cognitive immunization describes the reinterpretation of disconfirming evidence in ways that preserve prior expectations. Motivated reasoning, belief perseverance, and schema assimilation all occupy adjacent territory. ACH narrows the target to attachment-generated costs and asks whether deactivation predicts a characteristic route through which those costs are reattributed or neutralized.

The novelty is not “people protect beliefs.” The novelty would be evidence that a specific multilevel route profile explains why attachment avoidance predicts reduced corrective uptake, and that the profile predicts working-model revision better than attachment avoidance, general defensiveness, or symptom severity alone.

5.6 Interpersonal complementarity and corrective emotional experience

Interpersonal theory has long examined how one person’s behavior evokes reciprocal or complementary responses. The noncomplementary treatment tradition explicitly proposes that clinicians sometimes need to decline the interpersonal role a client solicits. Corrective emotional experience likewise predates ACH by decades. Intensive short-term dynamic psychotherapy is mechanistically aligned in its explicit attention to defenses, but its evidence base is smaller and more heterogeneous than the couple-therapy and alliance literatures invoked here (Abbass, Town, & Driessen, 2012).

ACH’s contribution is to connect those clinical ideas to an information model. Noncomplementarity is useful only if it increases accurate, tolerable, retained feedback. A surprising relational response that is not understood or integrated is novelty without correction. The proposed outcome is not merely an emotional experience but an observable increase in corrective uptake.

5.7 Novelty statement

The component findings are not novel. Defensive exclusion, deactivation, low disclosure, treatment underutilization, demand-withdraw dynamics, clinical complementarity, epistemic trust, alliance rupture-repair, and cognitive immunization all have established literatures.

The proposed novelty is their synthesis into a multilevel correction architecture with four distinctive commitments:

1. The primary asymmetry concerns access to correction rather than intrinsic capacity to change.

2. CFI is route-specific and multilevel rather than a single attachment trait.

3. The theory predicts where information is lost between strategy-generated cost and working-model revision.

4. The theory can be falsified by showing that the routes do not differ by attachment dimension, do not predict corrective uptake, or add nothing beyond established constructs.

The claim is narrow by design. Its value depends on the routes being separately measurable and separately refutable.

6. What the Change Literature Shows

No identified study provides a clean population-level estimate of the probability that anxiously versus avoidantly attached adults “heal” or become secure. The remembered image of a large percentage gap should not enter the paper without an exact source and an outcome definition. Attachment change can mean lower self-reported anxiety or avoidance, a shift in AAI classification, stronger security, symptom remission, relationship improvement, or movement in a latent trajectory. Those are different endpoints.

The current literature supports five conclusions.

First, attachment representations can change. Security-enhancing relationship conditions can revise working models over time, although the operative ingredients vary across people and contexts (Arriaga, Kumashiro, Simpson, & Overall, 2018). Taylor, Rietzschel, Danquah, and Berry (2015) reviewed 14 psychotherapy studies and concluded that security generally increased and anxiety decreased, while findings for avoidance were unclear. Levy, Kivity, Johnson, and Gooch (2018), synthesizing 36 studies and 3,158 patients, found that greater pretreatment security predicted better psychotherapy outcome and that increases in security were associated with improvement. The review also suggested that clients with lower security may benefit more from therapies emphasizing relationships and interpersonal process.

Second, anxiety appears to decline more consistently in some longitudinal and treatment literatures. Chopik, Edelstein, and Grimm (2019) found that anxiety declined curvilinearly across a 59-year span, with pronounced decreases in middle and later adulthood, whereas avoidance declined more slowly and approximately linearly. This is compatible with ACH but is not evidence for its mechanism.

Third, the best within-study trajectory signal remains SOPHO-Net. Strauß et al. (2018) assessed 495 adults with social anxiety disorder before treatment, after CBT or psychodynamic therapy, and at six- and twelve-month follow-up. CBT produced significant pre-post reductions in both anxiety and avoidance. After treatment, anxiety continued to change, while avoidance was comparatively stable. The authors did not preregister or formally test the ACH-specific cross-dimension slope contrast. The pattern is suggestive, not dispositive.

Fourth, avoidance can change and can coexist with strong symptom response. A-Tjak, Morina, Boendermaker, Topper, and Emmelkamp (2020) found reductions in both attachment dimensions during ACT and CBT for depression, and greater baseline avoidance predicted greater symptom reduction. Exploratory SOPHO-Net analyses likewise suggested that highly avoidant patients could benefit substantially from CBT. Abapolnikova et al. (2026), in an 80-patient randomized comparison of interpersonal psychotherapy and CBT, found a small increase in security across treatments, d = .23, and linked early increases in security with lower posttreatment depression. The change did not differ by treatment. These studies contradict any simple claim that avoidance is intrinsically immutable.

Fifth, anxiety does not confer superior prognosis once treatment begins. Levy, Ellison, Scott, and Bernecker (2011) found that attachment anxiety predicted worse posttherapy outcome, security predicted better outcome, and avoidance was not significantly associated with outcome. Hyperactivation may produce legible distress without producing efficient learning. A loud signal can still be misattributed, converted into protest, or repeatedly discharged through reassurance and relationship cycling (Dailey et al., 2009).

The newest direct counterexample is especially useful. Veler-Poleg, Tchizick, and Zilcha-Mano (2026) followed alliance and depression weekly in 118 patients with major depressive disorder receiving short-term psychodynamic treatment. Neither anxiety nor avoidance predicted depression or alliance slopes, and alliance change did not mediate attachment-outcome associations. In exploratory response-surface analysis, patients high in avoidance and low in anxiety showed the greatest depression improvement. The result is exploratory and sample-specific, but it lands squarely against any monotonic claim that more avoidance means less benefit. It also warns that anxiety-avoidance configurations may carry information lost when the dimensions are modeled only as additive main effects.

The resulting formulation is sharper:

Attachment anxiety may generate more actionable distress without producing better outcome conditional on treatment. Attachment avoidance may reduce exposure to correction without reducing capacity to benefit once correction becomes usable.

This formulation explains why outcome studies alone cannot settle ACH. They condition on entry, participation, measurement, and retention. If avoidance predicts underexposure at those stages, the population of avoidant people represented in treatment studies is selected. The least engaged and least disclosed cases may be absent or attenuated before outcomes are estimated.

Selection is a testable proposition, not a rescue device. Population-first studies must recruit independently of treatment status, measure distress through multiple channels, and model entry and dropout explicitly. If avoidance does not predict underexposure after structural access, severity, comorbidity, prior treatment, and social resources are controlled, the selection component should be abandoned.

The clinical claim should therefore be stated in terms of conditional accessibility. Avoidance is not an iron door. It is a door whose hinges, locks, and handles may differ by person and relationship. The empirical job is to identify which mechanism is active and whether the intervention reaches it.

7. The Demarcation Problem: Deactivation Versus Legitimate Distance

A theory that questions accommodation of withdrawal can be weaponized almost instantly. “Your boundary is avoidance” is a sentence available to anyone who refuses another person’s autonomy. The paper is scientifically and ethically unacceptable unless it makes that misuse harder.

The proposed demarcation is not a diagnostic tool and cannot be applied to a single episode. It is a functional pattern analysis requiring time, context, and competing explanations. A request for distance can be emotionally costly to a partner and still be legitimate. A boundary does not become illegitimate because another person dislikes it. Conversely, calling a behavior a boundary does not settle what function it serves or how its costs are negotiated.

Table 2

Functional indicators for distinguishing communicated distance from a deactivating pattern

The dimensions are probabilistic, not dispositive. Work demands, caregiving, illness, disability, neurodivergence, sensory overload, grief, cultural norms, religious practice, communication style, and ordinary preferences can produce patterns that resemble withdrawal. Safety concerns override the framework entirely. No-contact after coercion, violence, stalking, manipulation, or abuse is protective action, not evidence of deactivation.

The most informative dimensions are intimacy contingency and discussability. A stable preference for solitude need not increase after closeness. A communicated boundary can usually survive a request for information, even when the answer remains no. Yet neither criterion is sufficient alone. A traumatized person may find inquiry difficult while setting a necessary boundary. A controlling partner may use “discussion” as an endless appeal process intended to defeat refusal.

The proper use of the demarcation is therefore clinical and research-oriented, not partner-diagnostic. A clinician may respect a client’s requested distance behaviorally while exploring its timing, function, costs, and alternatives. Respect and curiosity are compatible. The clinician does not override the distance. The clinician declines to treat the client’s first explanation as the last scientifically relevant fact.

For research, H5 below operationalizes the strongest distinction: within-person distance-seeking that rises reliably after intimacy should predict different outcomes from distance-seeking of equivalent quantity that is stable across relational temperature. If contingency adds no predictive value, the proposed demarcation fails.

8. When Clinical Deference Becomes Iatrogenic

Clinical deference is not a unitary intervention. It can mean respect for consent, slower pacing, reduced session frequency, acceptance of a client’s stated priorities, or avoidance of direct engagement with a defense. Some forms are indispensable. ACH concerns the last form and only under specified conditions.

A recurrent clinical pull may operate as follows:

1. The client minimizes attachment need, distress, or relational impact.

2. The clinician experiences low urgency, reduced emotional invitation, or concern about intrusion.

3. Intervention becomes less relationally deep or less direct.

4. Immediate tension decreases for both parties.

5. Reduced tension negatively reinforces the complementary response.

6. The client receives less discrepancy, less disclosure support, or less opportunity to examine the pattern.

This sequence is plausible, not yet established as a population-level norm. The phrase “standard clinical response” should be retired. The empirical question is whether avoidant presentation predicts a lower dose of relevant feedback after controlling for presenting problem, client preference, alliance quality, clinician orientation, and safety considerations.

The Dozier findings support the possibility. The broader alliance literature shows small rather than dramatic associations. Igra et al. (2026) suggest that growth in alliance, not simply initial strength, matters for subsequent avoidance change. Taken together, the evidence points toward process sensitivity rather than blanket deference or blanket challenge.

The iatrogenic claim requires four linked findings:

1. Presentation effect: deactivating presentation changes clinician behavior.

2. Dose effect: the changed behavior reduces exposure to accurate corrective information or relational depth.

3. Mechanism effect: lower exposure predicts weaker corrective uptake.

4. Outcome effect: an autonomy-preserving alternative improves uptake or change without unacceptable harm.

If any link fails, “iatrogenic” is too strong. The term belongs in the title because the paper makes that causal wager, but the title now says “when” rather than treating harm as the inevitable consequence of respect.

9. Accountability Under Protected Repair Conditions

The original phrase “engineered rupture with guaranteed repair” had force but made two indefensible promises. Rupture should not be manufactured for its own sake, and repair cannot be guaranteed by one party. What can be guaranteed is conduct: availability, nonretaliation, autonomy preservation, monitoring, and a serious attempt to repair.

The revised intervention hypothesis is accountability under protected repair conditions. It has five elements.

9.1 Consent and safety

The relationship must be safe enough for discrepancy work. Clinicians should screen for coercion, violence, stalking, acute instability, and contexts in which distance is protective. The client should understand that relational patterns may be explored and should retain the right to pause or refuse an exercise. In couple contexts, both partners require independent safety assessment.

9.2 Descriptive naming

Feedback is framed as observed behavior and impact, not diagnosis. “When contact stopped for three weeks without a return point, I did not know whether the relationship still existed” is information. “That is your avoidant attachment” is a contestable identity claim that invites debate over labeling rather than contact with consequence.

Naming should be specific enough to resist semantic drift. The target is not the need for distance. It is the manner, timing, predictability, or unacknowledged transfer of cost.

9.3 Autonomy preservation

The other person’s right to choose distance, decline contact, or reject the interpretation remains intact. Autonomy preservation is not decorative warmth around a predetermined demand. It must be behaviorally real. The person can say no without punishment, diagnosis, or escalating pursuit.

Overall et al. (2013) provide the clearest support for this moderator. Influence attempts were less likely to trigger avoidant anger and withdrawal when partners conveyed value and respected autonomy. The feedback did not disappear. Its delivery changed.

9.4 Sustained availability

Naming is not followed by retaliatory withdrawal, contempt, or threat. The relationship remains available for a bounded period and under stated conditions. This closes one easy reattribution route: “I was abandoned because I was criticized.” Sustained availability does not mean endless access or absence of consequences. It means the feedback source does not vanish at the moment discrepancy appears.

9.5 Monitoring and repair

The clinician or partner tracks activation, understanding, memory, and alliance. Misattunement is expected. The task is to identify whether the person heard the intended message, whether autonomy felt real, and whether the interaction produced curiosity, shutdown, rage, compliance, or confusion. Repair is an active process, not a promise that rupture will resolve.

The rupture-repair literature supports the importance of successful resolution. Eubanks, Muran, and Safran (2018) found a moderate association between rupture resolution and positive outcome. It does not follow that more rupture is better. The implication is that avoiding all tension may forgo useful discrepancy, while tension without repair can be harmful.

The pressure objection is empirical, not merely ethical. Reactance theory predicts resistance when freedom feels threatened (Brehm, 1966). Miller, Benefield, and Tonigan (1993) found that a confrontational style increased client resistance and predicted worse one-year drinking outcomes, while Patterson and Forgatch (1985) showed resistance rising and falling as therapists shifted between directive and reflective behavior. These studies are not attachment-specific, but they define a serious failure mode for ACH. Any intervention that produces more shutdown, dropout, coerced agreement, or symptom deterioration has failed even if the therapist believes the feedback was accurate.

The intervention can be summarized as high-fidelity discrepancy under protected relational conditions. It is not pressure. It is an attempt to prevent accurate impact information from being lost to vagueness, diagnosis, retaliation, or coercion.

10. Revised Hypotheses

H1a: Cross-modal divergence

Estimates of attachment avoidance will show lower convergence across self-report, interview or observer rating, partner report, behavioral observation, and physiological measurement than estimates of attachment anxiety, after reliability, relational target, and method variance are modeled.

H1b: Directional signal attenuation

During attachment-relevant stress, self-reported activation among highly avoidant participants will be lower relative to matched behavioral, partner-reported, interview-derived, attentional, or physiological indices than the corresponding discrepancy among highly anxious participants. Generic disagreement among measures is insufficient. The direction of discrepancy is the test.

H2: Corrective-system entry and participation

At equivalent levels of independently assessed distress and impairment, avoidance will predict lower probability of help-seeking, treatment entry, vulnerable disclosure, participation, and acceptance of assistance, while anxiety will predict higher entry or participation. Structural access, diagnosis, socioeconomic resources, prior treatment, and stigma must be controlled.

H3: Vulnerability-selective disclosure

Avoidance will predict selective rather than uniformly low disclosure. Positive, competence-signaling, and low-dependence material will be more likely to be shared than attachment need, shame, dependency, or partner-impact material. Selectivity will predict poorer clinician and partner accuracy about the active working model.

H4: Dyadic cost distribution and visibility

In mixed-profile dyads, withdrawal episodes will produce actor and partner costs, but the partner’s distress and interpretive labor will be less likely to return as accurately recalled, strategy-attributed information to the withdrawing actor. The ACH prediction concerns visibility and attribution, not a universal difference in total suffering.

H5: Intimacy-contingent distance

Within-person distance-seeking that increases after closeness, dependency, successful repair, or expressed need will predict poorer relational and individual outcomes than distance-seeking of equal duration that is stable across relational temperature. Contingency will add predictive value beyond attachment scores and baseline preference for solitude.

H6: Clinical complementarity and feedback dose

Deactivating presentation will predict shallower therapist exploration, fewer direct relational observations, or reduced disclosure support in at least a subset of clinicians. Those responses will mediate lower corrective-feedback dose. Therapist orientation, attachment, experience, client preference, and presenting problem will moderate the effect.

H7: Delivery moderation

Descriptive naming of withdrawal impact will improve subsequent corrective uptake and reduce deactivation when paired with autonomy preservation and sustained availability. Naming without those elements will increase withdrawal, anger, dropout, or alliance deterioration. Delivery style will moderate outcome more strongly than nominal confrontation intensity.

H8: Alliance trajectory

Among clients high in avoidance, early alliance growth will predict later reduction in deactivation more strongly than initial alliance level. This is a prediction about attachment change, not a claim that alliance growth universally mediates symptom outcome. In an experimental or sequential design, alliance growth will partially carry the effect of autonomy-preserving feedback on later corrective uptake, with symptom change modeled as a competing explanation.

H9: Episode-reset moderation

Repeated relational failure will predict revision only when cross-episode regularities are linked and attributed to a recurring strategy. Relationship substitution or “wrong fit” attribution will weaken that association. The prediction is about explanatory reset, not number of partners or moral judgment about leaving.

H10: Incremental validity

Route-specific CFI measures will predict corrective uptake and subsequent working-model change beyond attachment avoidance, anxiety, experiential avoidance, alexithymia, self-concealment, epistemic trust, reflective functioning, social desirability, symptom severity, and treatment expectancy.

H10 is the construct’s survival test. If CFI adds nothing beyond avoidance, it is redundant language. If the routes predict correction equally across attachment dimensions, the multilevel CFI model may survive while the asymmetric hypothesis fails.

11. Proposed Empirical Program

11.1 Define the outcome before building the scale

The dependent variable is not symptom reduction in general. It is corrective uptake: the degree to which discrepant, strategy-relevant information is accurately captured, retained, attributed, discussed, and translated into behavior or representation.

A multi-method corrective-uptake battery could include:

• immediate and delayed recall of partner or clinician feedback;

• recognition of the specific behavior-impact link;

• attribution of outcome to self, partner, context, and recurring strategy;

• willingness to discuss the feedback without disappearance or coercive compliance;

• observer-rated mentalization and epistemic receptivity;

• behavior change in a later analogous interaction;

• partner or clinician report of whether the same pattern recurs;

• longitudinal change in attachment measures or narrative coherence.

No single item should be allowed to stand in for revision. Agreement in the moment may be submission. Distress may be activation without learning. Self-reported insight may not generalize. Corrective uptake is a sequence and should be scored across time.

11.2 Stage 0: Causal map and adversarial specification

Before item generation, the theory should be represented as a directed causal graph. Each route needs a specified exposure, mediator, outcome, confounders, and plausible competing model. Examples:

• avoidance → reduced vulnerable disclosure → lower clinician accuracy → lower corrective uptake;

• avoidance → lower help-seeking → treatment nonentry → unobserved outcome;

• intimacy event → withdrawal → partner distress → feedback → attributional neutralization → no revision;

• therapist complementarity → lower feedback dose → flatter alliance growth → weaker avoidance change.

Alternative graphs should be preregistered. Symptom severity may cause both avoidance scores and underengagement. Poor alliance may produce withdrawal rather than the reverse. Partner criticism may cause both withdrawal and partner distress. A coercive clinician may make “deflection” accurate rather than defensive. The design must permit those possibilities.

Construct development should begin with qualitative interviews from multiple perspectives: people high in avoidance, people high in anxiety, partners, clinicians from different orientations, attachment researchers, cultural psychologists, neurodivergent participants, and survivor advocates. The goal is not to validate the author’s vocabulary. It is to discover observable routes and false-positive conditions.

11.3 Stage 1: Population-first selection and multimethod measurement

A first major study should recruit from the population rather than treatment settings and stratify on independently assessed distress and impairment. A sample of roughly 600 to 1,000 adults would support multilevel subgroup and selection analyses, but power should be determined through simulation based on the final measurement model and expected attrition.

Participants would complete:

• dimensional global and relationship-specific attachment measures;

• an interview-based attachment or narrative-coherence measure in a subsample;

• clinical diagnostic and functioning interviews;

• treatment history and current utilization;

• measures of structural access, insurance, cost, stigma, prior treatment, and social support;

• experiential avoidance, alexithymia, self-concealment, reflective functioning, epistemic trust, and social desirability;

• behavioral help-seeking and help-acceptance tasks;

• attachment-relevant attention, memory, disclosure, and feedback-recall tasks;

• physiological or ambulatory stress indices in a sufficiently powered subsample.

H1a and H1b must be separated. A longitudinal multitrait-multimethod model can estimate trait, method, relational target, and state components. Cross-modal disagreement supports measurement dependence. It supports CFI’s directional signal claim only if avoidance predicts lower apparent activation or change in self-report relative to appropriately matched non-self-report indicators.

Treatment entry should be modeled as selection. Inverse-probability weighting, doubly robust estimators, or explicit selection models can reduce observed confounding, but none solves unmeasured selection. Sensitivity analyses should quantify how strong an unmeasured factor would need to be to erase the avoidance-entry association. Missingness and dropout should be treated as potentially informative, not automatically missing at random.

11.4 Stage 2: Dyadic intensive longitudinal study

The dyadic mechanism requires repeated within-person data. Recruit approximately 200 to 300 couples with oversampling of mixed anxiety-avoidance profiles and follow them for 21 to 30 days using ecological momentary assessment. Laboratory sessions should include both closeness induction and a conflict or influence discussion, with ambulatory physiology where feasible.

Events should be timestamped:

1. closeness, dependency, disagreement, or repair;

2. urge for distance;

3. withdrawal behavior;

4. actor and partner affect;

5. feedback attempt;

6. interpretation of feedback;

7. re-entry or continued withdrawal;

8. later recall and behavior.

Dynamic structural equation modeling or multilevel vector autoregression can estimate within-person lags while separating stable between-person differences. Actor-partner interdependence models should test both actor and partner outcomes. Person-mean centering is essential so that the intimacy-contingency hypothesis is not reduced to the claim that avoidant people are generally more distant.

The decisive test is not whether withdrawal predicts partner distress. That is already plausible. It is whether partner distress fails to become retained, strategy-attributed information for the actor, and whether the route differs from anxiety after conflict intensity, partner behavior, and baseline dissatisfaction are modeled.

11.5 Stage 3: Psychotherapy process study

A naturalistic psychotherapy study should recruit at least 300 clients nested within enough therapists to estimate therapist-level variance, preferably 60 or more. Sessions should be recorded and coded during the first eight sessions for:

• client vulnerable disclosure;

• therapist depth of inquiry;

• therapist direct relational observation;

• validation and autonomy language;

• withdrawal, topic shift, minimization, or confusion;

• rupture markers and repair attempts;

• epistemic trust or learning markers;

• alliance after every session.

Multilevel growth models can distinguish initial alliance from alliance trajectory. The analysis should test whether deactivating presentation predicts a lower feedback dose and whether therapist behavior mediates the association between avoidance and corrective uptake. Therapist attachment, theoretical orientation, experience, professional self-doubt, and caseload should be modeled as moderators rather than treated as noise.

Video coding is critical. Client self-report alone cannot validate a theory partly concerned with underdisclosure and measurement dependence. Therapist report alone risks complementarity bias. Observer ratings, session transcripts, and delayed client recall provide independent channels.

11.6 Stage 4: Experimental delivery studies

The original four-arm clinical trial proposed deliberately withholding elements of repair from some participants. That is too ethically ambitious for an unvalidated construct. The safer sequence begins with analogue and microprocess experiments.

Stage 4A: Vignette and analogue experiments

Randomize descriptions of identical withdrawal patterns by:

• naming versus no naming;

• autonomy-preserving versus controlling language;

• sustained availability versus threatened withdrawal;

• “boundary” versus “avoidance” versus neutral labels;

• high versus low intimacy contingency.

Measure perceived intrusion, responsibility, clarity, willingness to re-engage, memory for impact, and recommended response. These studies can establish whether the proposed semantic and delivery effects exist before they are imported into treatment.

Stage 4B: Laboratory dyadic microintervention

In consenting couples without current abuse or severe instability, train partners or facilitators to deliver a constant impact statement under randomized autonomy and availability conditions. Use immediate and delayed outcomes, including physiological activation, feedback recall, re-engagement, and later behavior. Stop rules should be based on distress and coercion indicators.

Stage 4C: Sequential clinical design

Only after safety and signal are established should therapy research randomize microinterventions. A micro-randomized trial or sequential multiple-assignment design is preferable to assigning clients to a rigid confrontational style. At decision points, clinicians can randomize among validated, ethically acceptable feedback forms and adapt based on alliance and activation.

Causal mediation claims require caution. Alliance growth, disclosure, and symptom change influence each other. Sequential randomization or strong longitudinal assumptions are needed before claiming that alliance growth mediates avoidance change. Ordinary cross-lagged panels should be avoided when random-intercept cross-lagged or dynamic multilevel models are feasible.

11.7 Construct scoring

Route measures should initially remain separate. A profile may include:

• signal attenuation score;

• disclosure and help-exposure score;

• dyadic return-path score;

• attributional neutralization score;

• episode-linking score.

A formative CFI index is justified only if each route contributes causally and nonredundantly to corrective uptake. Weights should be estimated and cross-validated, not assigned by intuition. A reflective factor should be rejected if removal of one route changes the construct’s meaning, if routes do not covary, or if interventions can alter one route without affecting the others.

Measurement invariance should be tested across gender, culture, sexual orientation, neurodivergence, relationship type, and clinical status. A construct framed as “insulation” could otherwise encode majority communication norms as pathology.

12. Clinical and Normative Implications

If supported, ACH would change what researchers measure, what clinicians count as an intervention, and what relational discourse is asked to distinguish.

Outcome research would need to move upstream. Treatment response cannot answer a theory about treatment entry, disclosure, and measurement selection by itself. Studies would need population-first recruitment, multiple methods, and explicit models of who becomes observable.

Clinicians would treat accommodation as an active choice rather than the absence of one. Respecting a client’s request for distance may remain the correct decision. The change is epistemic: the clinician would ask what the decision does to disclosure, alliance, feedback dose, and long-term patterning. “I complied” would no longer be mistaken for “I made no intervention.”

Relational discourse would be asked to defend boundaries by drawing them more clearly. The target is not solitude, autonomy, or ending relationships. It is the collapse of announced, discussable, bounded distance and unannounced, intimacy-contingent disappearance into one morally protected category. Better distinctions protect genuine boundaries from conceptual dilution while making relational costs speakable.

The humane core of ACH is not that avoidant individuals are getting away with something. It is that a person can carry attachment-relevant costs in a system that repeatedly helps those costs remain unformulated, underdisclosed, displaced, or mismeasured. That is not advantage. It is a form of informational abandonment.

13. Falsification Criteria

ACH should be revised or abandoned under the following conditions.

1. No incremental validity: Route-specific CFI measures do not predict corrective uptake or working-model change beyond avoidance and adjacent constructs.

2. No asymmetry: Anxiety and avoidance show equivalent route profiles after distress, method, and relationship context are controlled. In that case, a general CFI model may survive, but ACH does not.

3. No directional measurement effect: Avoidance does not show the predicted direction of discrepancy between self-report and matched non-self-report indicators.

4. No entry effect: Avoidance fails to predict help-seeking, participation, disclosure, or help acceptance after structural access and severity are controlled.

5. No dyadic return failure: Partner-reported costs return to withdrawing actors with equal accuracy, retention, and strategy attribution.

6. No intimacy contingency: Distance-seeking after closeness predicts no different outcome than equivalent stable solitude.

7. No clinical complementarity: Deactivating presentation does not alter therapist feedback dose, relational depth, or disclosure support.

8. No delivery interaction: Autonomy preservation and sustained availability do not moderate responses to descriptive naming.

9. No alliance-trajectory effect: Early alliance growth does not predict later avoidance change once symptom improvement and retention are modeled.

10. No episode reset: Linking versus separating cross-relationship episodes does not alter strategy attribution or change.

The iatrogenic claim has an additional burden. Even if deference reduces feedback dose, the paper is wrong clinically if autonomy-preserving alternatives produce equal or greater harm, dropout, coercion, or deterioration. A mechanism can be real while its proposed intervention is unacceptable.

The most informative positive result would be a double dissociation: high-avoidance people with low CFI change readily, while lower-avoidance people with high CFI do not. That result would show that the proposed architecture explains revision better than the attachment label that generated it.

14. Ethical Boundaries and Weaponization Risk

This theory can be used badly with almost no effort. Its safeguards must therefore be part of the theory, not an appendix of manners.

First, partners do not receive diagnostic authority. A person may describe behavior and impact, set their own limits, or leave. They may not convert resistance to their preferred level of contact into proof of another person’s pathology.

Second, autonomy is not conditional on psychological insight. A person does not need to persuade a partner or clinician that their reasons are healthy before declining contact. Clinical exploration can continue only within consent and role boundaries.

Third, the framework does not apply to withdrawal from coercive, violent, exploitative, or abusive contexts. Protective distance may be abrupt, nondiscussable, and indefinite because discussion itself is unsafe. The demarcation table will fail exactly where power is most unequal unless safety is assessed first.

Fourth, communication differences require protection. Neurodivergence, disability, cultural norms, sensory needs, and trauma can alter pacing, eye contact, language, re-entry, and tolerance for synchronous discussion. Measurement must distinguish inaccessible social convention from reduced corrective uptake.

Fifth, clinicians cannot use ACH to justify paternalism. Treating a request as clinical material does not mean refusing the request. The legitimate claim is smaller: compliance has effects and should be examined like any other intervention.

Sixth, research must monitor deterioration, not merely average benefit. Delivery studies need stopping rules, independent safety review, partner-specific consent, and adverse-event reporting. Researchers should report subgroup harm, not hide it inside a favorable mean.

Seventh, “accountability” cannot become enforced confession. Corrective uptake includes the freedom to consider feedback and reject it. The theory predicts accurate processing, not compulsory agreement.

The ethical standard is simple to state and difficult to fake: no one is entitled to another person’s proximity. Everyone is entitled to describe what distance cost them. ACH asks whether those truths can coexist without either one being erased.

15. Limitations

The central construct is unvalidated. CFI may prove redundant with experiential avoidance, self-concealment, epistemic mistrust, alexithymia, or reflective-functioning deficits. Its route structure is theoretical, and the proposed distinction between route profile and formative index requires empirical adjudication.

The evidence base is stitched across measurement traditions that converge weakly. AAI dismissing states, romantic avoidance, relationship-specific avoidance, and observed deactivation cannot be pooled casually. Some supporting studies use specialized psychiatric samples; others use convenience samples of couples or students. Generalization must be earned.

Many effects are small. The attention effect for avoidance is modest; alliance associations are small; treatment findings vary by disorder, method, and intervention. A multilevel theory can be important with small component effects, but only if the components add predictive value and replicate.

Causal direction is often unclear. Withdrawal can cause partner demand, and demand can cause withdrawal. Poor alliance can follow avoidance, and poor treatment fit can produce avoidance-like behavior. Distress can increase attachment insecurity and service utilization simultaneously. Longitudinal and experimental work is necessary.

The cultural route remains weak. Context-dependent penalization is plausible, but the paper lacks direct evidence that clinical or popular discourse systematically codes deactivation as virtue and hyperactivation as pathology. The proposed content analysis and framing experiments are prerequisites for strong claims.

Episode reset is also speculative. Relationship termination can be adaptive, necessary, or unrelated to deactivation. The route concerns attribution across repeated episodes, not the morality or frequency of exit.

The clinical evidence does not establish that deference is a widespread standard or that noncomplementarity is generally superior. The title makes an intervention-level hypothesis that must survive therapist-level, client-level, and safety-level moderators.

The dissonance model remains underspecified. Corrective information may change working models through prediction error, emotional learning, mentalization, behavioral reinforcement, narrative integration, or social learning. “Dissonance discharge” is a useful organizing metaphor but not yet a demonstrated mechanism.

Finally, the theory emerged partly from lived relational experience. That origin is not disqualifying, but it increases the need for adversarial design. The framework should be tested by researchers who expect it to fail, using outcomes and comparison constructs selected before results are known.

16. Conclusion

The Asymmetric Correction Hypothesis began with a simple observation: strategies may differ not only in the suffering they produce, but in whether that suffering returns as usable information.

The revised theory no longer treats anxiety and avoidance as groups with fixed healing rates. It proposes an asymmetry in access to correction. Hyperactivation often makes distress legible and actionable without guaranteeing good outcome. Deactivation can reduce attention, disclosure, treatment exposure, relational return, or attribution without eliminating the capacity to change. When an avoidant person is reached through a usable alliance and an intervention that preserves autonomy, meaningful change is possible.

Corrective Feedback Insulation is therefore not one hidden trait and not five equally proven valves. It is a multilevel architecture of candidate routes through which strategy-generated costs can be lost before they revise the strategy. Some routes are already supported; some are plausible; some are deliberately on probation.

The clinical wager is equally conditional. Deference becomes iatrogenic only when it participates in insulation. The alternative is not confrontation. It is accountability under protected repair conditions: accurate naming, real autonomy, continued availability, close monitoring, and willingness to repair.

The paper’s empirical burden is now clear. Show where correction is lost. Show that the loss differs by strategy. Show that it predicts revision beyond existing constructs. Show that closing a route improves uptake without coercion or harm. If those findings do not appear, the theory should contract or die.

The simplest formulation remains the strongest:

A strategy cannot be revised by information that never becomes reachable.

References

Abbass, A. A., Town, J. M., & Driessen, E. (2012). Intensive short-term dynamic psychotherapy: A systematic review and meta-analysis of outcome research. Harvard Review of Psychiatry, 20(2), 97-108. [https://doi.org/10.3109/10673229.2012.677347]{.underline}

Abapolnikova, M., Coyne, A. E., Glushka, L., Liu, C. J., Constantino, M. J., Atkinson, L. R., Bagby, R. M., Ravitz, P., & McBride, C. (2026). Does attachment change serve as a theory-specific mediator of depression reduction in interpersonal psychotherapy versus cognitive behavioral therapy? Psychotherapy Research. Advance online publication. [https://doi.org/10.1080/10503307.2026.2656698]{.underline}

Adams, G. C., Wrath, A. J., & Meng, X. (2018). The relationship between adult attachment and mental health care utilization: A systematic review. The Canadian Journal of Psychiatry, 63(10), 651-660. [https://doi.org/10.1177/0706743718779933]{.underline}

Altmann, U., Nodop, S., Dinger, U., Ehrenthal, J. C., Schauenburg, H., Dymel, W., Willutzki, U., & Strauß, B. M. (2021). Differential effects of adult attachment in cognitive-behavioural and psychodynamic therapy in social anxiety disorder: A comparison between a self-rating and an observer rating. Clinical Psychology & Psychotherapy, 28(2), 373-383. [https://doi.org/10.1002/cpp.2513]{.underline}

Arriaga, X. B., Kumashiro, M., Simpson, J. A., & Overall, N. C. (2018). Revising working models across time: Relationship situations that enhance attachment security. Personality and Social Psychology Review, 22(1), 71-96. [https://doi.org/10.1177/1088868317705257]{.underline}

A-Tjak, J. G. L., Morina, N., Boendermaker, W. J., Topper, M., & Emmelkamp, P. M. G. (2020). Explicit and implicit attachment and the outcomes of acceptance and commitment therapy and cognitive behavioral therapy for depression. BMC Psychiatry, 20, Article 155. [https://doi.org/10.1186/s12888-020-02547-7]{.underline}

Bellah, R. N., Madsen, R., Sullivan, W. M., Swidler, A., & Tipton, S. M. (1985). Habits of the heart: Individualism and commitment in American life. University of California Press.

Bernier, A., & Dozier, M. (2002). The client-counselor match and the corrective emotional experience: Evidence from interpersonal and attachment research. Psychotherapy: Theory, Research, Practice, Training, 39(1), 32-43. [https://doi.org/10.1037/0033-3204.39.1.32]{.underline}

Blakey, S. M., & Abramowitz, J. S. (2016). The effects of safety behaviors during exposure therapy for anxiety: Critical analysis from an inhibitory learning perspective. Clinical Psychology Review, 49, 1-15. [https://doi.org/10.1016/j.cpr.2016.07.002]{.underline}

Bollen, K. A., & Bauldry, S. (2011). Three Cs in measurement models: Causal indicators, composite indicators, and covariates. Psychological Methods, 16(3), 265-284. [https://doi.org/10.1037/a0024448]{.underline}

Bowlby, J. (1980). Attachment and loss: Vol. 3. Loss: Sadness and depression. Basic Books.

Brehm, J. W. (1966). A theory of psychological reactance. Academic Press.

Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measurement of adult attachment: An integrative overview. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 46-76). Guilford Press.

Caughlin, J. P., & Huston, T. L. (2002). A contextual analysis of the association between demand-withdraw and marital satisfaction. Personal Relationships, 9(1), 95-119. [https://doi.org/10.1111/1475-6811.00006]{.underline}

Chopik, W. J., Edelstein, R. S., & Grimm, K. J. (2019). Longitudinal changes in attachment orientation over a 59-year period. Journal of Personality and Social Psychology, 116(4), 598-611. [https://doi.org/10.1037/pspp0000167]{.underline}

Christensen, A., & Heavey, C. L. (1990). Gender and social structure in the demand-withdraw pattern of marital conflict. Journal of Personality and Social Psychology, 59(1), 73-81. [https://doi.org/10.1037/0022-3514.59.1.73]{.underline}

Dailey, R. M., Pfiester, A., Jin, B., Beck, G., & Clark, G. (2009). On-again/off-again dating relationships: How are they different from other dating relationships? Personal Relationships, 16(1), 23-47. [https://doi.org/10.1111/j.1475-6811.2009.01208.x]{.underline}

Degnan, A., Seymour-Hyde, A., Harris, A., & Berry, K. (2016). The role of therapist attachment in alliance and outcome: A systematic literature review. Clinical Psychology & Psychotherapy, 23(1), 47-65. [https://doi.org/10.1002/cpp.1937]{.underline}

Dewitte, M., De Houwer, J., Goubert, L., & Buysse, A. (2010). A multi-modal approach to the study of attachment-related distress. Biological Psychology, 85(1), 149-162. [https://doi.org/10.1016/j.biopsycho.2010.06.006]{.underline}

Diamantopoulos, A., & Siguaw, J. A. (2006). Formative versus reflective indicators in organizational measure development: A comparison and empirical illustration. British Journal of Management, 17(4), 263-282. [https://doi.org/10.1111/j.1467-8551.2006.00500.x]{.underline}

Dozier, M. (1990). Attachment organization and treatment use for adults with serious psychopathological disorders. Development and Psychopathology, 2(1), 47-60. [https://doi.org/10.1017/S0954579400000584]{.underline}

Dozier, M., Cue, K. L., & Barnett, L. (1994). Clinicians as caregivers: Role of attachment organization in treatment. Journal of Consulting and Clinical Psychology, 62(4), 793-800. [https://doi.org/10.1037/0022-006X.62.4.793]{.underline}

Dozier, M., & Kobak, R. R. (1992). Psychophysiology in attachment interviews: Converging evidence for deactivating strategies. Child Development, 63(6), 1473-1480. [https://doi.org/10.1111/j.1467-8624.1992.tb01708.x]{.underline}

Dozier, M., Lomax, L., Tyrrell, C. L., & Lee, S. W. (2001). The challenge of treatment for clients with dismissing states of mind. Attachment & Human Development, 3(1), 62-76. [https://doi.org/10.1080/14616730010000858]{.underline}

Dozier, M., & Tyrrell, C. (1998). The role of attachment in therapeutic relationships. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 221-248). Guilford Press.

Edelstein, R. S. (2006). Attachment and emotional memory: Investigating the source and extent of avoidant memory impairments. Emotion, 6(2), 340-345. [https://doi.org/10.1037/1528-3542.6.2.340]{.underline}

Edelstein, R. S., & Gillath, O. (2008). Avoiding interference: Adult attachment and emotional processing biases. Personality and Social Psychology Bulletin, 34(2), 171-181. [https://doi.org/10.1177/0146167207310024]{.underline}

Eilert, D. W., & Buchheim, A. (2023). Attachment-related differences in emotion regulation in adults: A systematic review on attachment representations. Brain Sciences, 13(6), 884. [https://doi.org/10.3390/brainsci13060884]{.underline}

Emery, L. F., Gardner, W. L., Carswell, K. L., & Finkel, E. J. (2018). You can’t see the real me: Attachment avoidance, self-verification, and self-concept clarity. Personality and Social Psychology Bulletin, 44(8), 1133-1146. [https://doi.org/10.1177/0146167218760799]{.underline}

Emerson, R. M. (1962). Power-dependence relations. American Sociological Review, 27(1), 31-41. [https://doi.org/10.2307/2089716]{.underline}

Etcheverry, P. E., Le, B., Wu, T.-F., & Wei, M. (2013). Attachment and the investment model: Predictors of relationship commitment, maintenance, and persistence. Personal Relationships, 20(3), 546-567. [https://doi.org/10.1111/j.1475-6811.2012.01423.x]{.underline}

Eubanks, C. F., Muran, J. C., & Safran, J. D. (2018). Alliance rupture repair: A meta-analysis. Psychotherapy, 55(4), 508-519. [https://doi.org/10.1037/pst0000185]{.underline}

Festinger, L. (1957). A theory of cognitive dissonance. Stanford University Press.

Fisher, S., Guralnik, T., Fonagy, P., & Zilcha-Mano, S. (2025). The development of the Epistemic Trust Rating System (ETRS). Psychotherapy Research, 35(3), 412-423. [https://doi.org/10.1080/10503307.2023.2299213]{.underline}

Fonagy, P., Campbell, C., & Bateman, A. (2017). Mentalizing, attachment, and epistemic trust in group therapy. International Journal of Group Psychotherapy, 67(2), 176-201. [https://doi.org/10.1080/00207284.2016.1263156]{.underline}

Fraley, R. C., & Brumbaugh, C. C. (2007). Adult attachment and preemptive defenses: Converging evidence on the role of defensive exclusion at the level of encoding. Journal of Personality, 75(5), 1033-1050. [https://doi.org/10.1111/j.1467-6494.2007.00465.x]{.underline}

Fraley, R. C., & Shaver, P. R. (1997). Adult attachment and the suppression of unwanted thoughts. Journal of Personality and Social Psychology, 73(5), 1080-1091. [https://doi.org/10.1037/0022-3514.73.5.1080]{.underline}

Fraley, R. C., Vicary, A. M., Brumbaugh, C. C., & Roisman, G. I. (2011). Patterns of stability in adult attachment: An empirical test of two models of continuity and change. Journal of Personality and Social Psychology, 101(5), 974-992. [https://doi.org/10.1037/a0024150]{.underline}

Fraley, R. C., & Waller, N. G. (1998). Adult attachment patterns: A test of the typological model. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 77-114). Guilford Press.

Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An item response theory analysis of self-report measures of adult attachment. Journal of Personality and Social Psychology, 78(2), 350-365. [https://doi.org/10.1037/0022-3514.78.2.350]{.underline}

Friedman, M., Rholes, W. S., Simpson, J., Bond, M., Diaz-Loving, R., & Chan, C. (2010). Attachment avoidance and the cultural fit hypothesis: A cross-cultural investigation. Personal Relationships, 17(1), 107-126. [https://doi.org/10.1111/j.1475-6811.2010.01256.x]{.underline}

Girme, Y. U., Overall, N. C., Simpson, J. A., & Fletcher, G. J. O. (2015). “All or nothing”: Attachment avoidance and the curvilinear effects of partner support. Journal of Personality and Social Psychology, 108(3), 450-475. [https://doi.org/10.1037/a0038866]{.underline}

Goodson, J. T., Patel, T. A., Zech, J. M., Sala, M. C., & Cougle, J. R. (2026). The standalone effect of safety behavior manipulations: A systematic review and meta-analysis. Clinical Psychology Review, 124, 102703. [https://doi.org/10.1016/j.cpr.2026.102703]{.underline}

Haslam, N. (2016). Concept creep: Psychology’s expanding concepts of harm and pathology. Psychological Inquiry, 27(1), 1-17. [https://doi.org/10.1080/1047840X.2016.1082418]{.underline}

Hermida-Barros, L., Primé-Tous, M., García-Delgar, B., Forcadell, E., Lera-Miguel, S., Fernández de la Cruz, L., Vieta, E., Radua, J., Lázaro, L., & Fullana, M. A. (2024). Family accommodation in obsessive-compulsive disorder: An updated systematic review and meta-analysis. Neuroscience & Biobehavioral Reviews, 161, 105678. [https://doi.org/10.1016/j.neubiorev.2024.105678]{.underline}

Horne, S. J., Liu, S., & Aafjes-van Doorn, K. (2024). The role of therapists’ attachment and introject in their treatment process and outcome: A systematic review. Clinical Psychology & Psychotherapy, 31(5), e3043. [https://doi.org/10.1002/cpp.3043]{.underline}

Igra, L., Mayo, O., Kivity, Y., Jacobsen, C. F., Lunn, S., & Poulsen, S. (2026). Early alliance growth and changes in attachment styles in psychotherapy. Psychotherapy. Advance online publication. [https://doi.org/10.1037/pst0000622]{.underline}

Illouz, E. (2007). Cold intimacies: The making of emotional capitalism. Polity Press.

Illouz, E. (2019). The end of love: A sociology of negative relations. Oxford University Press.

Jarvis, C. B., MacKenzie, S. B., & Podsakoff, P. M. (2003). A critical review of construct indicators and measurement model misspecification in marketing and consumer research. Journal of Consumer Research, 30(2), 199-218. [https://doi.org/10.1086/376806]{.underline}

Kohn, J. L., Rholes, W. S., & Schmeichel, B. J. (2012). Self-regulatory depletion and attachment avoidance: Increasing the accessibility of negative attachment-related memories. Journal of Experimental Social Psychology, 48(1), 375-378. [https://doi.org/10.1016/j.jesp.2011.06.020]{.underline}

Körner, R., Overall, N. C., Chang, V. T., Hammond, M. D., Sasaki, E., Schütz, A., & Zverling, E. (2026). The relational nature of attachment and power: Attachment avoidance and withdrawal limit partners’ power. Personality and Social Psychology Bulletin, 52(7), 2098-2115. [https://doi.org/10.1177/01461672251333165]{.underline}

Kube, T., Glombiewski, J. A., Gall, J., Touissant, L., Gärtner, T., & Rief, W. (2019). How to modify persisting negative expectations in major depression? An experimental study comparing three strategies to inhibit cognitive immunization against novel positive experiences. Journal of Affective Disorders, 250, 231-240. [https://doi.org/10.1016/j.jad.2019.03.027]{.underline}

Kube, T., Kirchner, L., Lemmer, G., & Glombiewski, J. A. (2022). How the discrepancy between prior expectations and new information influences expectation updating in depression: The greater, the better? Clinical Psychological Science, 10(3), 430-449. [https://doi.org/10.1177/21677026211024644]{.underline}

Levy, K. N., Ellison, W. D., Scott, L. N., & Bernecker, S. L. (2011). Attachment style. Journal of Clinical Psychology, 67(2), 193-203. [https://doi.org/10.1002/jclp.20756]{.underline}

Levy, K. N., Kivity, Y., Johnson, B. N., & Gooch, C. V. (2018). Adult attachment as a predictor and moderator of psychotherapy outcome: A meta-analysis. Journal of Clinical Psychology, 74(11), 1996-2013. [https://doi.org/10.1002/jclp.22685]{.underline}

Li, D., Carnelley, K. B., & Rowe, A. C. (2023). Insecure attachment orientation in adults and children and negative attribution bias: A meta-analysis. Personality and Social Psychology Bulletin, 49(12), 1679-1694. [https://doi.org/10.1177/01461672221117690]{.underline}

Marmarosh, C. L., Kivlighan, D. M., Bieri, K., LaFauci Schutt, J. M., Barone, C., & Choi, J. (2014). The insecure psychotherapy base: Using client and therapist attachment styles to understand the early alliance. Psychotherapy, 51(3), 404-412. [https://doi.org/10.1037/a0031989]{.underline}

Mikulincer, M., Dolev, T., & Shaver, P. R. (2004). Attachment-related strategies during thought suppression: Ironic rebounds and vulnerable self-representations. Journal of Personality and Social Psychology, 87(6), 940-956. [https://doi.org/10.1037/0022-3514.87.6.940]{.underline}

Mikulincer, M., & Shaver, P. R. (2016). Attachment in adulthood: Structure, dynamics, and change (2nd ed.). Guilford Press.

Miller, W. R., Benefield, R. G., & Tonigan, J. S. (1993). Enhancing motivation for change in problem drinking: A controlled comparison of two therapist styles. Journal of Consulting and Clinical Psychology, 61(3), 455-461. [https://doi.org/10.1037/0022-006X.61.3.455]{.underline}

Myung, J., Furrow, J. L., Palmer, G., Sandberg, J., Bradley, B., & Johnson, S. M. (2022). Understanding the emotional landscape in the withdrawer re-engagement and blamer-softening EFCT change events. Journal of Marital and Family Therapy, 48(3), 815-834. [https://doi.org/10.1111/jmft.12583]{.underline}

Notsu, H., Blansfield, R. E., Spina, D. S., & Levy, K. N. (2025). An updated meta-analysis of the relation between adult attachment style and working alliance. Psychotherapy Research, 35(5), 721-734. [https://doi.org/10.1080/10503307.2024.2370344]{.underline}

Overall, N. C., Fletcher, G. J. O., & Friesen, M. D. (2003). Mapping the intimate relationship mind: Comparisons between three models of attachment representations. Personality and Social Psychology Bulletin, 29(12), 1479-1493. [https://doi.org/10.1177/0146167203251519]{.underline}

Overall, N. C., Simpson, J. A., & Struthers, H. (2013). Buffering attachment-related avoidance: Softening emotional and behavioral defenses during conflict discussions. Journal of Personality and Social Psychology, 104(5), 854-871. [https://doi.org/10.1037/a0031798]{.underline}

Patterson, G. R., & Forgatch, M. S. (1985). Therapist behavior as a determinant for client noncompliance: A paradox for the behavior modifier. Journal of Consulting and Clinical Psychology, 53(6), 846-851. [https://doi.org/10.1037/0022-006X.53.6.846]{.underline}

Peng, X., Gillath, O., Jiang, M., Wang, B., Zhang, J., & Wu, L. (2024). Attachment style and attention bias to emotional information: The moderating effect of stress, stimulus characteristics, and attention stage. Journal of Personality, 92(5), 1315-1340. [https://doi.org/10.1111/jopy.12891]{.underline}

Rathgeber, M., Bürkner, P.-C., Schiller, E.-M., & Holling, H. (2019). The efficacy of emotionally focused couples therapy and behavioral couples therapy: A meta-analysis. Journal of Marital and Family Therapy, 45(3), 447-463. [https://doi.org/10.1111/jmft.12336]{.underline}

Roisman, G. I., Holland, A., Fortuna, K., Fraley, R. C., Clausell, E., & Clarke, A. (2007). The Adult Attachment Interview and self-reports of attachment style: An empirical rapprochement. Journal of Personality and Social Psychology, 92(4), 678-697. [https://doi.org/10.1037/0022-3514.92.4.678]{.underline}

Roisman, G. I., Tsai, J. L., & Chiang, K.-H. S. (2004). The emotional integration of childhood experience: Physiological, facial expressive, and self-reported emotional response during the Adult Attachment Interview. Developmental Psychology, 40(5), 776-789. [https://doi.org/10.1037/0012-1649.40.5.776]{.underline}

Saypol, E., & Farber, B. A. (2010). Attachment style and patient disclosure in psychotherapy. Psychotherapy Research, 20(4), 462-471. [https://doi.org/10.1080/10503301003796821]{.underline}

Schrodt, P., Witt, P. L., & Shimkowski, J. R. (2014). A meta-analytical review of the demand-withdraw pattern of interaction and its associations with individual, relational, and communicative outcomes. Communication Monographs, 81(1), 28-58. [https://doi.org/10.1080/03637751.2013.813632]{.underline}

Spengler, P. M., Lee, N. A., Wiebe, S. A., & Wittenborn, A. K. (2024). A comprehensive meta-analysis of emotionally focused couple therapy. Couple and Family Psychology: Research and Practice, 13(2), 81-99. [https://doi.org/10.1037/cfp0000233]{.underline}

Spielmann, S. S., MacDonald, G., & Wilson, A. E. (2009). On the rebound: Focusing on someone new helps anxiously attached individuals let go of ex-partners. Personality and Social Psychology Bulletin, 35(10), 1382-1394. [https://doi.org/10.1177/0146167209341580]{.underline}

Strauß, B., Altmann, U., Manes, S., Tholl, A., Koranyi, S., Nolte, T., Beutel, M. E., Wiltink, J., Herpertz, S., Hiller, W., Hoyer, J., Joraschky, P., Nolting, B., Ritter, V., Stangier, U., Willutzki, U., Salzer, S., Leibing, E., Leichsenring, F., & Kirchmann, H. (2018). Changes of attachment characteristics during psychotherapy of patients with social anxiety disorder: Results from the SOPHO-Net trial. PLOS ONE, 13(3), e0192802. [https://doi.org/10.1371/journal.pone.0192802]{.underline}

Sun, E. R., & Jakubiak, B. K. (2024). Attachment avoidance predicts limited and selective sharing of personal events in close relationships. Personal Relationships, 31(2), 397-419. [https://doi.org/10.1111/pere.12537]{.underline}

Taylor, P. J., Rietzschel, J., Danquah, A., & Berry, K. (2015). Changes in attachment representations during psychological therapy. Psychotherapy Research, 25(2), 222-238. [https://doi.org/10.1080/10503307.2014.886791]{.underline}

Tyrrell, C. L., Dozier, M., Teague, G. B., & Fallot, R. D. (1999). Effective treatment relationships for persons with serious psychiatric disorders: The importance of attachment states of mind. Journal of Consulting and Clinical Psychology, 67(5), 725-733. [https://doi.org/10.1037/0022-006X.67.5.725]{.underline}

Veler-Poleg, N., Tchizick, A., & Zilcha-Mano, S. (2026). Attachment orientations and treatment outcomes in psychotherapy: Does the alliance mediate the association? Clinical Psychology & Psychotherapy, 33(3), e70296. [https://doi.org/10.1002/cpp.70296]{.underline}

Waller, W. (1938). The family: A dynamic interpretation. Dryden Press.

Ybarra, J. A., & Seedall, R. B. (2024). Fostering connection: A dyadic analysis of the relationships between mattering, attachment, and mental health. Journal of Marital and Family Therapy, 50(3), 649-668. [https://doi.org/10.1111/jmft.12720]{.underline}