Do Brief Lab-Based Interventions Decrease Coercive Conflict between Couples and Parent-Child Dyads?

Abstract

Coercive conflicts between parents and children and between couples are implicated in the pathogenesis of a variety of psychological and physical health problems. Despite its seeming importance to population health, there are no widely available, easy-to-use methods with demonstrated efficacy to engage coercive conflict and reduce it. Identifying and testing potentially efficacious and disseminable micro-interventions (i.e., interventions that can be delivered in under 15 minutes via computer or paraprofessional) for targets with cross-cutting health implications such as coercive conflict is the focus of the National Institutes’ of Health “Science of Behavior Change” initiative. We experimentally tested four micro-interventions targeting coercive conflict in couple and parent-toddler dyads in a within-between design. There were mixed, but supportive findings for the efficacy of most of the micro-interventions. Attributional reframing, implementation intentions, and evaluative conditioning all reduced coercive conflict as assessed by some, but not all measures of observed coercion. No findings indicated any iatrogenic effects. Interpretation Bias Modification Treatment improved at least one measure of coercive conflict for couples, but not for parents and toddlers; additionally, it increased self-reported coercive conflict. Overall, these results are encouraging and suggest that very brief and highly disseminable micro-interventions for coercive conflict are a fruitful direction for inquiry. Optimizing micro-interventions and deploying them across the healthcare infrastructure could tremendously enhance family functioning and, in turn, health behaviors and health (ClinicalTrials.gov IDs: NCT03163082, NCT03162822).

Most individual behavior is transactional, occurring as a function of the social environments, past and present, in which learning occurs/occurred (e.g., Bandura, 1986). Arguably the most potent agent of socialization is the family system (Slep et al., 2018), which has implications for health — both direct (e.g., influencing cardiovascular and stress responsivity; Cartozian & Ybarra, 2005; Graham et al., 2013) and indirect — via promoting or undermining health-supportive behaviors (e.g., diet, activity level, sleep habits, medication compliance; Repetti et al., 2002). The relationship between dysfunctional family environments (i.e., high in conflict, low in warmth) and poor health is well established (Repetti et al., 2002; Tobin et al., 2013). Indeed, hostile parent-child and intimate-partner relationships have been negatively associated with health-supportive behaviors, immune response, and healthy neuroendocrine function; and positively linked with chronic illness (e.g., cardiac disease, metabolic syndrome), poor oral health, and premature mortality (Repetti et al., 2002; Tobin et al., 2013). Among the key mechanisms proposed to give rise to and sustain such destructive relationships is coercive dyadic process (“coercion” for short; Patterson, 1982). Research has found that hostile parent-child and partner interactions, and specifically coercion (operationalized below), contribute to the development of aggressive and antisocial behavior in children and are generally implicated in harsh parenting, escalated family conflict, and family violence (Dishion & Snyder, 2016a; Snyder & Dishion, 2016).

Coercion Theory (Patterson, 1982) is the moniker given to the conditioning that occurs through the dynamics of repeated interpersonal conflict that train each person to act in ever more hostile and aggressive ways over time, thereby possibly potentiating detrimental health effects. (Note that coercion in this context is not synonymous with conventional use of “coercion” meanting persuading someone to do something by using force or threats.) Initially developed to explain behavior in parent-child dyads, Coercion Theory posits that children learn oppositional behavior and aggression through conditioning regarding behavior that is effective at ending parent-child conflicts: escalating aversive behavior. Over time, if parents respond to children’s escalating aversive behavior and aggression by giving in, children learn to behave aggressively to get their way. Importantly, not only the child’s but also the parent’s behaviors are maintained through reinforcement. The child is negatively reinforced for persisting and escalating aversive behavior (via the parent giving in and ending the conflict). The parent is negatively reinforced for giving in (via terminating the child’s aversive behavior). Of course, the child doesn’t always win. At times, the parent responds to the child’s escalation with her own escalation, and the child backs down in response. Over time, these conflicts serve as learning trials. Both the parent and child learn to be coercive — or in this narrow context, to behave increasingly aversively — because escalation turns off the other’s aversive behavior. Thus, once a coercive process takes hold, each member of the dyad is faced with an unfortunate choice: (a) give in and lose the battle, or (b) win via out-escalating the other. In Patterson’s (1976, p. 1) exquisite phrasing, each partner is both the “victim and architect of a coercive system.” Terminating the aversive interaction may concurrently put an end to individuals’ unpleasant internal experiences of the conflict, (i.e., negative affect), providing additional sources of negative reinforcement (Slep et al., 2018). Although coercive conflict relies on negative reinforcement, it can be positively reinforced as well (e.g., a child turns to plays video games because of the parent’s disengagement/backing down).

In isolation, a minor coercive conflict is relatively innocuous. Indeed, all relationships involve some degree of periodic low-level coercion (Snyder, 2002). However, as coercive exchanges increase in frequency — spurred by adversity or major life stress (Conger et al., 1994) or because de-escalating behaviors are not rewarded (Slep et al., 2016) — they concurrently increase in severity (Patterson, 2002, 2016). Moreover, they become overlearned and eventually automatic (Dumas, 2005). As such, generalized, entrenched hostile relationships can be traced back to the earlier accumulation of seemingly trivial dysfunctional episodes (i.e., “micro-origins of coercive processes” Patterson & Cobb, 1971).

Once the coercive process takes hold, aversive escalation behaviors tend to generalize across topics of disagreement and relationships. Although Coercion Theory was originally developed to explain the development of antisocial behavior across childhood, it appears that these dynamics generalize beyond parent-child relationships and can affect other close dyadic relationships as well, such as intimate partners (e.g., Slep et al., 2016).

Direct health effects likely operate through the physiological activation that is part and parcel of escalating (e.g., Repetti et al, 2002). Indirect effects likely operate through the potentiation of parents and partners giving in and giving up on supporting the acquisition of healthy routines for children or partners (Lorber et al., 2014). The insidiousness and ubiquity of coercion, as well as the breadth of deleterious sequelae, have led to calls to make it a target of public health intervention (Biglan, 2016; Slep et al., 2018). Indeed, we believe coercive conflict is a modifiable transdiagnostic mechanism of health risk in close family relationships.

We propose that coercion is part of the “black box” (Nielsen et al., 2018) that (a) explains noxious interpersonal and health outcomes and (b) can be modified for improved functioning through successful intervention. Although coercion is likely a mechanistic factor underlying numerous maladaptive outcomes (Dishion & Snyder, 2016b), coercive process has only thoroughly validated in the etiology of childhood conduct problems and aggression, and more recently implicated as a mechanism contributing to intimate partner violence (Slep et al., 2016). Yet, the literature abounds with parenting and couple interventions that implicate coercive process in the etiology of interpersonal problems in that they propose directly target coercion (which they likely do) by teaching parents or couples different and non-escalating ways to respond to partner/child escalation, but fail to specifically measure the impact of intervention elements on coercive process specifically. This is emblematic of much of intervention science, so much so that the U.S. National Institute of Health’s “Science of Behavior Change” (SOBC) initiative promoted a mechanism-explicating “experimental-medicine approach to behavior change” (Nielsen et al., 2018; Riddle, 2015), Undoubtedly, in the case of coercion, this is due in part to the fact that methods for measuring coercion (microanalytic observational coding), while precise, are slow and costly (Biglan, 2016; Mitnick et al., 2021). As such, changes in dyadic coercive conflict have only been measured across a handful of treatment studies (e.g., Forgatch et al., 2005; Sitnick et al., 2015), and only one recent study has identified which components of these multifaceted treatments act on coercive processes (Holtrop et al., 2021). Figure 1 depicts how coercive conflict is hypothesized to function within the context of the experimental medicine approach to behavior change.

Figure 1.

Experimental Medicine Approach to Behavior Change Model of Coercion and Health Behaviors.

Moreover, successful interventions with explicit foci of targeting parent-child — such as Parent Management Training-Oregon Model (Forgatch, 1994; Patterson, 2005), Triple P (Sanders et al., 2002), Incredible Years (Webster-Stratton, 2005) — and couple coercion (e.g. Integrative Behavioral Couple Therapy; Jacobson & Christensen, 1996) are unavailable to most families due to logistical, financial, and access barriers (Doss et al., 2017). In addition to an accessibility problem, there may be a willingness one. Research suggests that even when such programs are made available to at-risk parents as a resource, they are poorly attended (Chacko et al., 2016). It has been suggested that program duration, location, scheduling, and stigma are all contributors (DeGarmo & Jones, 2019). As DeGarmo and Jones (2019) note, the onus is on clinical researchers to determine the appropriate balance of required resources for implementation and optimal dose to effectively treat specific populations. Addressing coercion through a public-health lens dictates the development of an array of evidence-based, scalable interventions — starting with the most economical, portable, and brief and progressing to more intensive treatments — that emphasize matching need and intensity, thus maximizing feasibility and minimizing financial and time burdens (e.g., Sanders & Mazzucchelli, 2017).

In the service of developing truly scalable, disseminable interventions, SOBC’s experimental-medicine approach to behavior change allows for assessing the short-term efficacy of brief ‘micro-interventions’, while concurrently testing whether coercion, a putative mechanism of health behavior change, is specifically engaged. Proof of concept support for this strategy comes from research on brief interventions and correlates of the coercive process. Similar to SOBC’s goal of target engagement, Gottman et al. (2005) proposed the use of “proximal change experiments” to elicit immediate measurable change in behaviors that imply coercion among distressed couples’ following brief specific interventions. Babcock et al. (2011) used the proximal change experiment approach to test interventions lasting under 10 minutes in duration for couples reporting male-to-female physical IPV. Men receiving the microinterventions, compared with controls, showed fewer signs of coercion (i.e., lower expressed and experienced aggression); their female partners, who underwent no intervention, reported experiencing less aggressive affect (an amalgam of anger, distrust, jealousy, and vengefulness) toward their partners (Babcock et al., 2011). Among parent-child dyads, several processes that can be altered via micro-intervention impact indicators of coercion (e.g., use of time-out, consistency in parenting responses, increasing positive parent-child interaction; see Slep et al., 2018). As such, it seems likely that coercion can respond to brief, easily implemented interventions.

Thus, consistent with the SOBC agenda, the purpose of our study was to test the ability brief, easy-to-implement micro-interventions to successfully engage and reduce coercive conflict within participants with whom we could also investigate the impact of coercive conflict on health behavior. We sought to test micro-interventions that could be used both with couples and parent-child dyads, had high likelihood of reducing coercive conflict in a laboratory setting, and had been demonstrated to result in behavior change. Two of the micro-interventions tested required a conversation with a college-educated, trained facilitator (i.e., attributional reframing, implementation intentions) and two were computer-administered (i.e., evaluative conditioning, interpretation bias modification).

Micro-Interventions

In controlled experiments, attributional reframing (replacing hostile attributions for another’s behavior with benign ones) impacts harsh parenting and negative couple behavior (e.g., Fincham & Bradbury, 1988; Slep & O’Leary, 1998), and is a component of empirically supported dyad-focused treatments (e.g., Epstein & Baucom, 2002). In fact, repeated attributional reframing reduced child abuse among at-risk mothers (Bugental et al., 2002).

Implementation intentions (Gollwitzer, 1999) are a strategy for setting and implementing a behavioral goal. Conceptually, they are “if-then” plans intended to help override behavioral automaticity by indicating how, where, and when a new response will be enacted. Overriding automaticity is necessary to changing well-practiced coercive exchanges (e.g., Gottman et al., 2005). Individuals consciously learn the implementation intention by selecting a cue, in this case, one that interferes with enacting coercive responses. When the cue is experienced, the goal-directed response is enacted immediately without conscious intent (Gollwitzer, 1999).

Evaluative conditioning (Milner et al., 2017) employs pictures of children with ambiguous facial expressions. A classical conditioning paradigm is used: ambiguous pictures of children’s faces (conditioned stimuli) are paired repeatedly with positive descriptive adjectives (e.g., “sweet”, “cooperative;” unconditioned stimuli). Participants are presented with a focal point cross for 1000 ms, and then the ambiguous pictures and positive word descriptors for 20 ms each). In a linked set of six studies (Milner et al., 2017), Evaluative conditioning significantly increased positive attitudes toward children, decreased negative attitudes, and decreased expectations of the need for harsh discipline. In a randomized controlled trial, those in the treatment group, compared with controls, reduced attributions of hostile intent, hostile-intent related feelings of anger, and expectations about use of harsh verbal and physical discipline. These findings imply that this intervention would reduce coercive conflict.

Interpretation bias modification treatment targets the tendency to interpret ambiguous cues from others as a threat, often referred to as a “hostile interpretation bias” (e.g., Leibenluft & Stoddard, 2013). In this computer-based treatment protocol (Stoddard et al., 2016), a participant is presented repeatedly with pictures of the face of the same adult with varying “morphed” facial expressions (i.e., very slight changes across a range of pictures). Using feedback via an instrumental conditioning paradigm, the intervention aims to move the individual’s set point toward more benign interpretations of ambiguous facial expressions. In a series of three experiments, youth diagnosed with Disruptive Mood Dysregulation Disorder (DMDD), compared to controls, were found to have a baseline tendency to classify ambiguous faces as angry (Stoddard et al., 2016). Interpretation Bias Modification Treatment decreased irritability and was associated with brain-based changes (e.g., increased activation in the lateral orbitofrontal cortex). These outcomes were of interest because depression is correlated with both hostile interpretation biases and coercion (Ha & Kim, 2016; Reuben & Shaw, 2016). As such it seemed plausible that proximal changes to hostile interpretation biases could reduce coercion.

Because the SOBC initiative focuses on putative mechanisms of behavior change within the context of health behaviors, our samples were selected to facilitate research on health behaviors as well (although these findings are not the focus of this paper). As such, we focused on parent-child dyads in which the child was at elevated risk for early childhood caries (ECC) and couples in which at least one adult had, or was at risk for, type II diabetes (T2D). As we review elsewhere (Slep et al., 2018), we selected both these conditions as health foci because of their high prevalences, tremendous impact, and shared risk pathways (e.g., sugar consumption); children’s sugar consumption is related to family conflict; and medical regimen adherence among those with T2D is linked with a variety of couple relationship variables that implicate coercive conflict.

In this study, we hypothesized that each of these micro-interventions would reduce observer-coded and self-reported coercion (i.e., within couples and parent-child dyads). We used a within-between randomized design. Dyads were randomized to one of the interventions and served as their own control, with the order of intervention and control interactions also being randomized. The exception to this was for dyads randomized to the Interpretation bias modification treatment. Because of the need for time to pass after the intervention (Stoddard et al., 2016) order could not be counterbalanced.

Method

We tested these interventions in separate parent-child and couple studies with medically at-risk populations. Study procedures were approved by the university institutional review board.

Participants

Study participants were (a) parent-child dyads in which the child, a 1.5- to 3-year-old, was at elevated risk for early childhood caries (ECC) or (b) couples in which at least one adult had, or was at risk for, type II diabetes (T2D). Recruitment was done through several channels: (a) in-person at (1) New York University (NYU) College of Dentistry adult and pediatric dental clinics, (2) the Bellevue Hospital pediatric dental clinic, and (3) the NYU Lutheran Medical Center; (b) referrals from other studies and previous participants; (c) electronic health records; (d) advertisements in public places; and (e) a paid recruitment service (i.e., ClinEdge).

Eligibility for the parent-child substudy was as follows: parent at least 18-years-of-age; the presence of, or risk factors for, ECC in the child or the family; and an elevated level of behavior problems. Seventy-nine caregiver/child dyads (N = 158 individuals) participated. Parents’ age range was 18–47 years (M = 33.11 years, SD = 5.66). Family income range was $0–$250,000 (median = $60,000; IQR = $24,950-100,000). Most parents were married (53.2%), with 28.6% single, 13% living with a partner, and 5.2% in a non-cohabiting relationship. Most parents were employed (28.6% full-time, 20.8% part-time), with 19.5% unemployed, 26% identifying as a homemaker, and 5.2% as student, disabled, or on a leave of absence. In terms of race/ethnicity, 1.4% of parents identified as American Indian or Alaska Native, 8.3% as Asian, 38.9% as African American or Black, 34.7% as white, 8.3% as multiracial and 8.3% as “other;” 34.2% identified as Hispanic or Latinx of any race. Children’s were 55.5% male and 44.5% female; their ages ranged from 18–39 months (M = 26.73 months, SD = 5.58). In terms of race/ethnicity, 8.1% were identified by their guardian as Asian, 44.6% as African American or Black, 29.7% as white, 12.2% as multiracial, and 4.1% as “other;” 35.9% of the children as Hispanic or Latinx of any race.

Eligibility for the couples substudy was as follows: at least 18-years-of-age; presence of, or risk for, T2D in at least one member; and clinical levels of relationship distress reported by at least one member. Participants (N = 71 couples, 142 individuals) ranged from 27–85 years-of-age (M = 52.89 years, SD = 12.52). Family income ranged from $3,120 to $290,000 (median = $40,000; IQR = $20,000-80,000). Most couples were married (66.7%), with 33.3% living together. Most couples were opposite-gender (87.3%); 12.7% same-gender couples participated. About a third of participants were employed (23.6% full-time, 11.1% part-time), with 25% disabled, 23.6% unemployed, 8.2% homemaker, 6.3% retired, and 2.1% students. In terms of race/ethnicity, 4.4% as Asian, 46% as African American or Black, 32.8% as White, 3.6% as multiracial, 13.1% as “other;” 22.7% identified as Hispanic or Latinx of any race.

Measures

See Table S1 of the online supplement for the list of dependent variables analyzed.

Eligibility

Child Behavior Screener.

Parents completed the 9-item Multidimensional Assessment of Preschool Disruptive Behavior (MAP-DB; Wakschlag et al., 2010) to assess child behavior problems. Responses are rated 0 (never) to 5 (many times a day). Scores of 12 or higher were considered eligible for the study.

Child ECC Screener.

Parents answered questions from the American Academy of Pediatric Dentistry’s (AAPD) Caries-Risk Assessment Form for 0–3-year-olds (2014), assessing the presence of early caries (or risk) in participating children or their siblings. Individuals with scores above 0 were considered to be at risk and eligible.

Couple T2D Screener.

Partners indicated whether either had been diagnosed with T2D. Risk for T2D was determined by self-reported presence of gestational diabetes, family history, high blood pressure, lack of physical activity, or high body mass index (Heikes et al., 2008).

Couple Distress Screener.

Couples were screened for eligibility with the 4-item Couple Satisfaction Index-4 (CSI; Funk & Rogge, 2007) and two verbal aggression items (shouted, yelled) from the Revised Conflict Tactics Scale (CTS-2; Straus et al., 1996) psychological aggression subscale. Individuals with CSI scores of 14 or lower who endorsed either of the verbal aggression items happening at least “sometimes” were eligible.

Observational Coding

Observers coded the parent-child and couples conflict interactions using several operationalizations. The coding and scoring are described more fully in the online supplement.

Parent-Child Coercion Coding.

Observers coded the content and intensity of parental or child behaviors every 5-sec for a variety of positive/negative physical behaviors, positive/negative verbal behaviors, and commands. Sequential scores were then computed for the negative reinforcement of child and parent negativity escalation versus de-escalation.

Couple Coercion Coding.

Observers used an expanded version of the validated Rapid Marital Interaction Coding System, Version 2 (Heyman et al., 2015), coding couple behavior in 5-sec intervals on a continuum from high positivity to high negativity. Sequential scores were then computed for the relative negative reinforcement of each partner’s negativity escalation versus de-escalation.

Global Coercion Coding.

We designed global codes for parent-child and couple interactions. Coders watched an entire interaction and made one set of Likert-scale ratings (e.g., “How much did the [child/partner] escalate the conflict?”). Two scores were operationalized for parent-child dyads: (1) either party wins (i.e., the extent to which the child or parent prevails in conflict bouts), and (2) child escalation wins (i.e., the extent to which negativity escalation wins conflict bouts). Three scores were operationalized for couples: (1) escalation-de-escalation works ratio (i.e., the extent to which negativity escalation vs. de-escalation ends the conflict, softens the conversation, or the person gets something out of it), (2) escalation wins (i.e., the extent to which conflicts are won via negative escalation), and (3) couple level coercion-stepdown (i.e., the extent to which conflict is characterized by coercion [mutual escalation until someone backs off, changes the topic, or gives up] vs. mutual stepdown [mutual, gradual steps down the anger continuum to get back to a lower level of upset]).

Self-Report

Self-reported coercion.

These 9-item questionnaires assess the coercive process in couple (Couple Coercive Process Scale; CCPS) and parent-child interactions (Parent-Child Coercive Process Scale; PCCPS). Seven of the items (e.g., “When I get into a conflict with my child/partner, we go back and forth taking it up a notch until things get too heated and one of us gives up”) are rated from 0 (never) to 5 (always), and two items (e.g., “When my child/partner gets hostile or combative, I often give in to what s/he wants”) from 0 (strongly disagree) to 5 (strongly agree). Both measures are unifactorial; have evidence of reliability, especially at higher levels of coercive process; and demonstrate concurrent validity with constructs in their nomological networks, with medium to large effect sizes (Mitnick et al., 2021). Cronbach’s αs = .88 (Parent-child) and .90 (Couple).

Participant-Reported Intervention Use and Impact

Participants who were assigned to attributional reframing or implementation intention intervention were asked to complete a short questionnaire after the subsequent interaction with their partner/child. The questionnaire assessed intervention helpfulness, previous knowledge, participant use during their subsequent interaction with a child or partner, and their perception of its impact, in a series of yes/no questions (e.g., Did you use these tips during the interactions with your child? If yes, did these tips make the interactions with your child any different? If yes, were the interactions more positive than usual?). Participants randomized to the computerized interventions were not asked these questions because the effects of these programs are mostly outside of conscious awareness.

Intervention Manipulation Checks

Parenting/Partner Cognition Scale.

(PCS; Snarr et al., 2009). This 11-item scale assesses the degree to which parents or partners endorse dysfunctional child- or partner-responsible attributions and more parent- or self-causal attributions for undesirable child or partner behavior. This scale was administered after the Attribution Bias intervention and interaction with the partner/child as a manipulation check to assess whether attributional reframing had occurred.

Procedure

Dyads were invited to participate in two lab sessions totaling 4-5 hours, during which families were consented, received intervention and control conditions, and completed interactions with their partners/children as well as self-report measures. Dyads were randomly assigned to one of the four intervention types—attribution retraining (n = 21 [parent-child substudy]; n = 22 (couples study), implementation intentions (n = 20 [parent-child substudy]; n = 19 (couples study), evaluative conditioning (n = 20 [parent-child substudy]; n = 14 (couples study), or interpretation bias modification (n = 18 [parent-child substudy]; n = 16 (couples study). Order of intervention and control was counterbalanced for dyads in the cognitive, behavioral, or evaluative conditioning intervention, such that 50% of dyads were observed interacting with their children/partners after first receiving an intervention, and 50% were first observed after the control condition. Because of the need to have at least 24 hours between the interpretation bias modification intervention and the assessment of coercion (Stoddard et al., 2016), all dyads randomized to the interpretation bias condition received the intervention condition during the first visit and no intervention during the second visit.

Interventions

Attribution Reframing (AR).

Adapted from Bugental et al. (2002), the facilitator asked the participant to describe a problem behavior (of the child’s or partner’s), and list potential causes of that problem with repeated inquiry until a benign or non-blame-oriented cause was generated. The facilitator then asked about potential ways of solving the problem until the participant generated a strategy and had them practice that strategy through imaginal rehearsal.

Fidelity.

Interventions were recorded and coded with an observer impressions coding system. Coder agreement ranged from 83 to 100% on all items except whether rehearsal had occurred, which was agreed on 50% of the time. Ratings indicated that a minimally sufficient intervention was delivered 95% of the time, and almost always (98%) identified an attribution, prompted toward a benign attribution, and selectively reinforced the benign attribution.

Manipulation Checks.

About 80% of both caregivers (n = 17 out of 21 valid responses) and couples (n = 36 out of 44 valid responses) reported using attributional reframing in their interactions with their children/partners. Caregivers who received the intervention reported more parent-causal attributions for their children’s negative behaviors during the intervention visit as compared with the control visit (t(19) = 3.24, p < .01). Couples who received the intervention reported no significant changes in partner-responsible cognitions (t(39) = 0.47, p = .64).

Implementation Intentions (II).

Adapted from Gollwitzer and Oettingen (2011), participants were encouraged to identify a way to not perpetuate conflict—not “taking a turn”—and a cue that would prompt them to use the strategy. Then they practiced this through imaginal exposure.

Fidelity.

Interventions were recorded and coded with an observer impressions coding system. Coder agreement ranged from 88 to 100% on all items. Fidelity ratings indicate that interventionists delivered a minimally sufficient intervention 98% of the time. They always identified negative behavior and how not to take a turn. Facilitating participant rehearsal of their strategy was less consistent (57-60%).

Manipulation Checks.

About 80% of both caregivers (n = 15 out of 18 valid responses) and partners (n = 31 out of 38 valid responses) reported using reported using the intervention in their interactions with their children/partners.

Evaluative conditioning (EC).

This computer-based intervention presents participants with pictures of ambiguous child or adult faces (conditioned stimuli) and pairs them with positive word descriptors (unconditioned stimuli; e.g., sweet, cooperative). Participants were provided the following instructions before the task began: “Now, I’m going to have you complete the image viewing task. In this task, you’ll be presented with images and/or words. All you will need to do is focus on the plus sign in the center of the screen. The images and/or words will flash quickly, so your job is to mentally note when you see them. If you happen to look away from the center of the screen, please re-focus your attention on the plus sign. Any questions?”

Fidelity.

The EC program was completed in its entirety without interruption in 91% of study trials. The other 9% included administration issues (e.g., ran original version instead of custom) or technical issues (e.g., computer froze mid-program).

Interpretation Bias Modification (IBM) Treatment.

This computer-based intervention presents participants with morphed facial expressions and has them determine whether the face is happy or angry. Across 15 morphs, participants are asked to make a simple forced-choice judgment (i.e., “happy” or “angry”) about how they think the individual is feeling. There are three blocks: 1) assessment; 2) training; 3) test. During the assessment block, the “balance point” of each participant (i.e., the degree to which they have a hostile interpretation bias for ambiguous expressions) is established, and no feedback is given for participant responses. During the training block feedback, based on the participant’s “balance point” established during the assessment block, is provided. Participants receive positive feedback for a “correct” happy classification. The program shifts what is classified as “happy” to two morphed images toward the angry side of the continuum from the participant-established balance point. They receive negative feedback for incorrect classification based on their shifted set-point. During the test block, participants again rate faces as “happy” or “angry” and no feedback is provided. Critically important to the design of this study, IBM developers state that the intervention takes 24 hrs to take effect. As such, the participants’ second visit had no intervention component, but was the “test” interaction for the intervention that was always administered in this condition in the first session.

Fidelity.

The interpretation bias program was completed in its entirety without interruption in 94% of trials in this study. The other 6% included stopping prematurely due to interruption or participant request.

Manipulation Checks.

The mean balance point was compared for the pre- and post-test phases using generalized estimating equation analyses. Suggesting successful experimental manipulations, compared to the pre-test phase, the mean balance point was significantly higher in the post-test phase in parents (B = 1.146, SE = 0.187, p < .001, d = 1.498) and couples (B = 1.374, SE = 0.183, p < .001, d = 0.964). This indicates that the intervention shifted the point in the morphed faces where participants rated a face as happy vs. angry.

Control.

During the control conditions for Attribution Reframing and Implementation Intentions, participants were asked to describe recent conflicts with their partner or child. Facilitators prompted as necessary to keep participants talking for approximately 10 minutes, to match the amount of time spent completing micro-interventions. This same procedure was used during the second interaction for participants in the IBM condition. Note that for the IBM condition, the interaction expected to have fewer treatment effects was the first interaction, which was the one in which IBM treatment was delivered. During the control condition for Evaluative Conditioning, participants completed an almost-identical computerized program as the intervention, but without any word descriptors.

Interactions

Parent-Child Interaction.

Caregiver-child dyads were observed in a series of behaviorally challenging structured tasks in a controlled laboratory setting. Caregivers were instructed to have their children clean up toys they had not had the opportunity to play with (≤10 min), play with a small set of far less attractive toys while their caregivers were on the phone answering developmental history questions (10 min; developmental history data will not be collected for analysis), and then lie quietly on a mat while the caregiver completed questionnaires (10 min). Attractive objects were placed around the room within reach of the child but were off-limits during all three tasks.

Couple Interaction.

Couples engaged in four 10-minute conflict discussions (two per lab visit) based on the most important topics for the couple. To generate topics for the discussions, each participant completed a questionnaire about things they had unsuccessfully tried to get their partners to do, do differently, or change during the preceding year. Participants created a list of possible desired changes, indicated whether they had engaged in discussions about each change over the past year, and rated the importance of each change to them. If more than one topic was rated as being of greatest importance, one was chosen via a random number generator. The four most important change(s) that had been discussed in the past year were selected. Participants were not told which topic had been selected until immediately before the conversation.

Analytic Strategy

Regression models were estimated using Mplus version 8 (L. K. Muthén & B. Muthén, 2017). The nesting of participants within dyads was handled via the pseudo maximum likelihood (ML) and weighted least squares (WLSMV) estimation methods (“type = complex” in the Mplus analysis specifications). These methods use sandwich estimators to adjust parameters’ SEs for the non-independence of nested observations and allows for the inclusion of cases with missing data¹ (Asparouhov & Muthén, 2005). Given the relatively small sample sizes and several nonnormal DV distributions, inferences were based on bias-corrected bootstrapped CIs (2,000 replicates). Several variables were winsorized (Wilcox, 2005) to reduce the influence of outliers. All analyses were conducted with standardized dependent variables; thus, raw regression coefficients (B) can be interpreted in the Cohen’s d metric.

Parent-Child study analyses were estimated at the level of the dyad and visit and used the WLSMV estimator due to the ordinal structure of the either party wins DV. The first set of analyses addressed the combined main effect of three intervention conditions: AR, II, and EC. IBM main effects were analyzed independently. In each analysis set, the DVs (Sequential negative reinforcement of the child and parent, either party wins, child escalation wins, and self-reported coercion) were simultaneously regressed on intervention, coded 1 (intervention) vs. 0 (control). To evaluate Intervention × Condition interactions, the DVs were simultaneously regressed on intervention, and two dummy variables (1/0) for contrasts among AR, II, and EC conditions, and two Intervention × Condition variables—one per contrast.

Couple study analyses were estimated at the level of the person and visit and used the ML estimator. Main effects and Intervention × Condition were evaluated following the Parent-Child study model. The DVs were sequential negative reinforcement, coercion-stepdown, escalation-deescalation works ratio, escalation wins, and self-reported coercion. The Intervention × Gender interaction, combining across the AR, II, and EC conditions, was estimated by simultaneously regressing the DVs on intervention, gender (1 = female; 0 = male), Intervention × Gender, and a same-gender couple covariate (same gender = 1; opposite gender = 2). Significant interactions were decomposed by calculating the relevant simple slopes per Preacher et al. (2006).

IBM effects were analyzed separately. As described above, IBM lacked a true control condition; active training occurred in the first visit, with the effects not expected to be apparent until the second visit. Given the lack of a control visit truly parallel to the other interventions, analyses of IBM could not be combined with the other conditions. Instead, the means for each coercion variable from the AR, II, and EC conditions’ control visit were used as the comparator against which the means from the IBM condition’s second visit were evaluated. This was accomplished by subtracting the AR, II, and EC conditions’ control visit means from each individual coercion variable score from the IBM condition’s second visit. Following the logic of a single sample t-test, the “control-adjusted” means from the IBM condition were tested against zero. To accomplish this test, a saturated covariance matrix was estimated, with the CIs for each coercion variable examined; the CIs’ exclusion of zero would indicate a significant effect.

Power

N was planned based on the minimum necessary n to achieve adequate power to detect medium size main effects within each condition. A priori power analyses were estimated with G*Power (Faul et al., 2007), with α =.05 two-sided significance tests. The minimum necessary n to achieve 80% power to detect medium size main effects (d = .50) within each condition was n = 34, indicating a target overall N per study of 136.²

Results

Participant-Reported Intervention Use and Impact

Participants generally felt that the interventions were helpful (100% caregivers; >95% couples), and many believed it was new information (40% caregivers; 47% couples) or strategies they had not tried before (40% caregivers; 47%) couples. They thought the strategy impacted their child’s/partner’s behavior (67% caregivers; 84% couples) in largely positive ways. Nearly 85% of caregivers reported using the intervention in their lab interactions, and 65% of couples reported using the intervention in their lab interactions.

Intervention Effects on Parent-Child Coercion

Attributional Reframing, Implementation Intentions, Evaluative Conditioning Effects

Intervention produced a significant reduction in sequential negative reinforcement of the child (B = −0.385, CI: −0.723, −0.076), but not in sequential negative reinforcement of the parent (B = 0.223, CI: −0.147, 0.585), child escalation wins (B = 0.018, CI: −0.284, 0.288), either party wins (B = −0.194, CI: −0.600, 0.216), or parent-reported coercion (B = −0.141, CI: −0.417, 0.118). Intervention × Condition interactions were nonsignificant for all dependent variables (Table 1).

Table 1.

Intervention × Condition Interactions (Parent-Child)

		95% CI				95% CI
Predictor	B	Low	High	Predictor	B	Low	High
DV: Sequential negative reinforcement of child
Intervention a	−0.207	−0.716	0.289	Intervention a	−0.468	−1.175	0.039
Imp. Intent. b	0.017	−0.661	0.687	Imp. Intent. c	−0.060	−0.783	0.625
Att. Reframing b	0.077	−0.532	0.655	Eval. Cond. c	−0.077	−0.656	0.532
Int. × Imp. Int.	−0.269	−1.021	0.508	Int. × Imp. Int.	−0.008	−0.774	0.859
Int. × Att. Ref.	−0.261	−1.066	0.486	Int. × Eval. Cond.	0.261	−0.495	1.065
DV: Sequential negative reinforcement of parent
Intervention a	0.073	−0.444	0.565	Intervention a	0.367	−0.284	1.055
Imp. Intent. b	−0.220	−0.833	0.422	Imp. Intent. c	−0.019	−0.732	0.721
Att. Reframing b	−0.200	−0.860	0.504	Eval. Cond. c	0.200	−0.508	0.854
Int. × Imp. Int.	0.148	−0.746	1.009	Int. × Imp. Int.	−0.146	−1.189	0.821
Int. × Att. Ref.	0.294	−0.579	1.175	Int. × Eval. Cond.	−0.294	−1.176	0.578
DV: Either party wins
Intervention a	−0.548	−1.379	0.136	Intervention a	−0.088	−1.008	0.921
Imp. Intent. b	−0.686	−1.497	0.130	Imp. Intent. c	−0.372	−1.174	0.422
Att. Reframing b	−0.314	−1.176	0.511	Eval. Cond. c	0.314	−0.512	1.174
Int. × Imp. Int.	0.546	−0.315	1.448	Int. × Imp. Int.	0.088	−1.045	1.125
Int. × Att. Ref.	0.458	−0.716	1.695	Int. × Eval. Cond.	−0.458	−1.699	0.707
DV: Child escalation wins
Intervention a	0.180	−0.423	0.781	Intervention a	0.005	−0.521	0.538
Imp. Intent. b	−0.300	−0.889	0.350	Imp. Intent. c	−0.019	−0.673	0.607
Att. Reframing b	−0.281	−0.946	0.342	Eval. Cond. c	0.281	−0.347	0.941
Int. × Imp. Int.	−0.310	−1.019	0.369	Int. × Imp. Int.	−0.135	−0.775	0.449
Int. × Att. Ref.	−0.175	−0.983	0.594	Int. × Eval. Cond.	0.175	−0.595	0.982
DV: Self-reported coercion
Intervention a	0.085	−0.607	0.606	Intervention a	−0.414	−0.842	0.029
Imp. Intent. b	−0.468	−1.203	0.147	Imp. Intent. c	−0.461	−1.056	0.147
Att. Reframing b	−0.007	−0.776	0.786	Eval. Cond. c	0.007	−0.791	0.775
Int. × Imp. Int.	−0.185	−0.858	0.569	Int. × Imp. Int.	0.312	−0.281	0.869
Int. × Att. Ref.	−0.499	−1.188	0.312	Int. × Eval. Cond.	0.499	−0.321	1.184

Note. Bias corrected bootstrapped CIs;

^acoded 1 (intervention) vs. 0 (control);

^breference category is evaluative Conditioning;

^creference category is attribution reframing.

Interpretation Bias Modification Effects

Intervention significantly reduced either party wins (B = −0.478, CI:−0.992, −0.118). In contrast main effects were nonsignificant for sequential negative reinforcement of the child (B = 0.191, CI:−0.224, 0.693) and parent (B = 0.104, CI:−0.574, 0.600), child escalation wins (B = 0.038, CI:−0.229, 0.294), and parent-reported coercion (B = −0.18, CI:−0.503, 0.184).

Intervention Effects on Couple Coercion

Attributional Reframing, Implementation Intentions, Evaluative Conditioning Effects

Intervention produced a significant reduction in escalation wins (B = −0.174, CI: −0.336, −0.011), but had nonsignificant effects on sequential negative reinforcement (B = −0.132, CI: −0.405, 0.143), coercion-stepdown (B = −0.205, CI: −0.537, 0.042), escalation-deescalation works ratio (B = −0.198, CI: −0.509, 0.155), and self-reported coercion (B = 0.088, CI: −0.147, 0.313).

Intervention × Condition interactions were nonsignificant for all dependent variables. (Table 2). However, significant Intervention × Gender interactions were found for escalation wins and self-reported coercion, but not sequential negative reinforcement, coercion-stepdown, or escalation-deescalation works ratio (Table 3). Simple slopes analysis revealed that the intervention significantly reduced escalation wins in females (SS = −0.393, CI: −0.619, −0.159), but not males (SS = 0.023, CI: −0.22, 0.292). Despite the significant interaction, simple slopes for self-reported coercion were nonsignificant for both females (SS = 0.258, CI: −0.049, 0.562) and males (SS = −0.066, CI: −0.328, 0.194).

Table 2.

Intervention × Condition Interactions (Couples)

		95% CI				95% CI
Predictor	B	Low	High	Predictor	B	Low	High
DV: Sequential negative reinforcement
Intervention a	0.001	−0.426	0.521	Intervention a	−0.340	−0.938	0.041
Att. Reframing b	0.091	−0.285	0.542	Att. Reframing c	0.038	−0.433	0.505
Eval. Cond. b	0.053	−0.359	0.523	Imp. Intent. c	−0.053	−0.524	0.358
Int. × Att. Ref.	−0.118	−0.800	0.494	Int. × Att. Ref.	0.223	−0.377	0.929
Int. × Eval. Cond.	−0.341	−1.055	0.255	Int. × Imp. Int.	0.341	−0.255	1.055
DV: Coercion-stepdown
Intervention a	−0.274	−0.537	0.014	Intervention a	−0.269	−0.647	0.137
Att. Reframing b	0.052	−0.668	0.759	Att. Reframing c	0.051	−0.635	0.694
Eval. Cond. b	0.002	−0.616	0.563	Imp. Intent. c	−0.002	−0.565	0.615
Int. × Att. Ref.	0.049	−0.609	0.790	Int. × Att. Ref.	0.044	−0.732	0.790
Int. × Eval. Cond.	0.005	−0.452	0.469	Int. × Imp. Int.	−0.005	−0.469	0.449
DV: Escalation-deescalation works ratio
Intervention a	−0.478	−0.830	−0.201	Intervention a	−0.160	−0.750	0.211
Att. Reframing b	−0.055	−0.682	0.622	Att. Reframing c	0.072	−0.635	0.758
Eval. Cond. b	−0.126	−0.760	0.549	Imp. Intent. c	0.126	−0.552	0.759
Int. × Att. Ref.	0.488	−0.221	1.351	Int. × Att. Ref.	0.170	−0.605	1.051
Int. × Eval. Cond.	0.318	−0.285	0.830	Int. × Imp. Int.	−0.318	−0.832	0.279
DV: Escalation wins
Intervention a	−0.170	−0.360	0.024	Intervention a	−0.119	−0.385	0.123
Att. Reframing b	0.079	−0.374	0.502	Att. Reframing c	0.059	−0.381	0.531
Eval. Cond. b	0.020	−0.397	0.413	Imp. Intent. c	−0.020	−0.414	0.393
Int. × Att. Ref.	−0.053	−0.436	0.346	Int. × Att. Ref.	−0.104	−0.547	0.315
Int. × Eval. Cond.	0.051	−0.269	0.355	Int. × Imp. Int.	−0.051	−0.356	0.266
DV: Self-reported coercion
Intervention a	−0.020	−0.255	0.238	Intervention a	0.168	−0.483	0.569
Att. Reframing b	0.101	−0.412	0.543	Att. Reframing c	0.387	−0.204	0.880
Eval. Cond. b	−0.286	−0.762	0.320	Imp. Intent. c	0.286	−0.323	0.761
Int. × Att. Ref.	0.136	−0.360	0.635	Int. × Att. Ref.	−0.052	−0.661	0.646
Int. × Eval. Cond.	0.188	−0.447	0.687	Int. × Imp. Int.	−0.188	−0.695	0.446

Note. Bias corrected bootstrapped CIs;

^acoded 1 (intervention) vs. 0 (control);

^breference category is Implementation Intentions;

^creference category is Evaluative Conditioning.

Table 3.

Intervention × Gender Interactions (Couples)

		95% CI
	B	Low	High
DV: Sequential negative reinforcement
Same gender couple a	0.040	−0.588	0.474
Intervention b	−0.007	−0.472	0.498
Gender c	0.267	−0.056	0.600
Intervention × Gender	−0.271	−0.848	0.268
DV: Coercion-stepdown
Same gender couple a	0.045	−0.728	0.772
Intervention b	−0.236	−0.523	0.054
Gender c	−0.098	−0.421	0.098
Intervention × Gender	−0.026	−0.201	0.187
DV: Escalation-deescalation works ratio
Same gender couple a	−0.216	−0.609	0.145
Intervention b	−0.089	−0.433	0.312
Gender c	−0.004	−0.315	0.298
Intervention × Gender	−0.230	−0.541	0.073
DV: Escalation wins
Same gender couple a	−0.061	−0.482	0.314
Intervention b	0.023	−0.220	0.292
Gender c	−0.035	−0.426	0.353
Intervention × Gender	−0.416	−0.743	−0.078
DV: Self-reported coercion
Same gender couple a	0.079	−0.455	0.663
Intervention b	−0.066	−0.328	0.194
Gender c	−0.046	−0.327	0.292
Intervention × Gender	0.324	0.015	0.674

Note. Bias corrected bootstrapped CIs;

^acoded (1 = same gender; 0 = opposite gender);

^bcoded (1 = intervention; 0 = control);

^ccoded (1 = female; 0 = male).

Interpretation Bias Modification Intervention Effects

Intervention produced a significant decrease in coercion-stepdown (B = −0.41, CI:−0.786, −0.019) and escalation wins (B = −0.293, CI:−0.613, −0.014). Nonsignificant effects were found for sequential negative reinforcement (B = −0.15, CI:−0.59, 0.182), self-reported coercion (B = −0.24, CI:−0.641, 0.222), and escalation-deescalation works ratio (B = −0.063, CI:−0.501, 0.413).

Discussion

Our hypotheses regarding the interventions’ impact on coercion received mixed support. Three micro-interventions (i.e., attributional reframing, implementation intentions, evaluative conditioning) jointly reduced coercion as assessed by gold-standard sequential coding of child coercion among parent-toddler dyads but did not result in a significant overall reduction in sequentially-coded parent coercion, child escalation, or either party winning via escalation from the global coding, or parent-reported coercion. With couples, micro-interventions produced an joint overall main effect in reducing the rating of the effectiveness of escalation but did not result in an overall reduction in other measures (i.e., globally coded, sequentially coded, or self-reported coercion). Interpretation bias modification appeared to decreased either party winning via escalation among parent-child dyads and escalation winning among couples, as well as couples being coded as less coercive overall, but produced nonsignificant changes on other measures.

Follow-up analyses identified relatively few reliable interactions and contrasts. Among couples, women were found to be more impacted by the interventions than men were based on self-report and globally coded coercion indices. Across parent-toddler dyads and couples, there is little indication of differential effectiveness of the micro-interventions. The disparate effects regarding observationally coded and self-reported coercive conflict provide further evidence of the importance of measuring behavior directly rather than relying on self-report. It may be that self-reports are less sensitive to change in actual behavior than are outsiders’ observations. In fact, none of the micro-interventions resulted in significant changes to self-reported coercion.

Our findings indicate that micro-interventions have promise in reducing coercive conflict. Among both couples and parent-toddler dyads, the main effect across the micro-interventions reduced at least one index of observed coercive conflict compared to the control/comparison condition. Given the relatively small sample sizes, the limited observation times that can contribute to floor effects (Heyman et al., 2001), and the rigorous experimental controls (counterbalancing the order of intervention and control observations, which could lead to carryover effects when the intervention observation precedes the control observation), that the micro-interventions showed signs of improving observed coercion is encouraging. Furthermore, All four of the micro-interventions were implementable with fidelity via computers or paraprofessionals. Parents and couples perceived both implementation intentions and attributional reframing as helpful.

This is the first study to test if evaluative conditioning changes observed behavior. Previous studies (Milner et al., 2017) indicated effects on parental attitudes, behavioral intentions, and behavioral analogue tests but did not assess parent-child interactions. Our study is also the first to test evaluative conditioning within adult relationships. Given our encouraging findings, in conjunction with the ease of dissemination of a brief computer-based tool, it would be worthwhile to research how to optimize the impact of this approach within, and potentially across. sessions. Milner and colleagues have conducted some research varying trials within a session (e.g., Milner et al., 2017; Wagner, 2018), but we are unaware of research examining (a) the decay of intervention impact or (b) the effects of repeating the intervention.

Attributional reframing has been found to produce behavior change in prior short-term experimental studies (e.g., Fincham & Bradbury, 1988; Slep & O’Leary, 1998) and as a component in a long-term intervention with repeated trials (Bugental et al., 2002). Although this intervention is not as easily disseminated as one that is entirely computer-based, the strength of the evidence suggests it would be useful to study how to optimize the impact of this intervention while minimizing the delivery burden. It could be that many people can learn how to identify benign attributions fairly quickly (i.e., with a handful of trials with an interventionist). Then, perhaps, technology-assisted prompts (Westwood et al., 2021) could cue the benign attributions, gradually achieving prominence in participants’ overall attributional repertoires.

An extensive literature supports the use of implementation intentions (e.g., Gollwitzer & Sheeran, 2006; Toli et al., 2016). However, this is the first test of using implementation intentions to reduce coercive conflict. In this study, we engaged participants to form implementation intentions to “not take a turn” in a conflict. This seems to have promise in producing at least short-term reductions in coercion. This is another micro-intervention that, after being initially delivered by a person in a health care setting, could be made more durable via technology-assisted prompts and reminders.

Interpretation bias modification treatment was tested with somewhat different methods within the context of this study, because within-subject counterbalancing was not an option given the nature of the intervention, which took 24 hours to take effect. This micro-intervention was tested with a between-subjects approach. It resulted in a similar pattern of findings however, resulting in change in one observational measure of coercion in parent-child dyads and two observational measures of coercion among couples. Interpretation bias modification treatment has been found to reduce self-reported anger and hostility (e.g., Dillon et al., 2020), but this is the first experimental test of its effect on observed behavior. As is the case with evaluative conditioning, this computer-based brief intervention is highly disseminable and holds promise for application in a wide range of contexts.

The dyadic nature of coercive conflict is apparent from our findings. By making a small shift in one person’s behavior (e,g., the parent but not the child), both members of the dyad are affected. Because of this, the power of interventions that target interpersonal processes is extended. If dyadic micro-interventions of the type tested here are further developed and optimized, it may be possible to intervene on one person and obtain a cascade of secondary effects benefiting not only the target person, but also their families.

It is worth highlighting that the current study is the only one of which we are aware that measured coercion in parallel ways in samples of both couples and parent-child dyads. Furthermore, the micro-interventions tested seemed to have comparable impacts in both dyads. As noted in the introduction, Coercion Theory (Patterson, 1982) was initially developed and tested in samples of parents and children. However, theories about escalating conflict in couples have drawn heavily on Coercion Theory without measuring it with the precision we have here (e.g., Gottman, 2005). On contribution of our study is to document that the same measurement strategies and intervention approaches that are used in the parent-child literature are applicable to couples as well.

Scientists and practitioners focused on reducing coercive conflict tend to build large, multicomponent interventions rather than isolate one intervention component and then try to optimize its efficacy. Given the challenges of engaging people in lengthy family interventions (Chako et al., 2016), and given the potential impact of small improvements to coercive conflict among the wider population, greater efforts to optimize micro-interventions as described above could significantly enhance public health. The COVID-19 pandemic has highlighted the critical need for a public health approach to family stress as families are under unprecedented strains leading to marked increases in dysfunction (Abrams et al., 2022). One can imagine a different approach to delivering behavioral intervention: rather than have a therapist work with couples or parents for an hour each week for months, perhaps many different types of health care providers could be able to include a brief intervention, perhaps even one that is computer-administered to all families identified at high risk, whether the provider is seeing the child (e.g., pediatricians, early interventionists) or the parent (e.g., OB/GYNs, primary care providers).

This study has several important limitations. The first is that the sample size is smaller than optimal. The COVID-19 pandemic halted enrollment. Thus, our findings especially need replication. As one of the aims of our research was to develop multiple measures of coercion, we have multiple dependent variables that do not provide a unified picture of our effects. Had our sample size been sufficient, the planned latent modeling would have been preferable to our eventual analytic approach. An additional challenge in micro-intervention research is that researchers must consistently evoke conflict at a relatively high level to have any ability to observe a reduction in conflict. Floor effects are a common problem, and one we encountered. Of course, our sample is limited to volunteers who were selected for being at risk for health conditions and thus results cannot be assumed to be the same if these interventions were more widely deployed without additional research.

Despite these limitations, we believe that taken as a whole, this study provides support for the notion that coercive conflict among couples and parent-child dyads can be reduced via very brief, highly disseminable interventions. This is exciting given the corrosive effects of coercion on mental and physical health and the number of families that are affected by it. Optimizing the impact of the types of micro-interventions studied and testing them across settings has substantial promise to improve public health.

Highlights.

• Mutual escalation that ends by one person giving in or giving up is coercive conflict

• Coercive conflict is implicated in aggression and relationship and health problems

• Three specific microinterventions reduced coercive conflict

• We may be able to use microinterventions in healthcare to reduce coercive conflict