What works for whom? Moderators in parental reflective functioning intervention

ABSTRACT

The DUET program (a group intervention) aims to enhance parental reflective functioning (PRF). We examined whether pretreatment levels of key outcomes as well as individual parental and family-environment characteristics predicted improvement after intervention with the DUET program. One hundred forty-two parents (native Israelis; mean age = 34.84 years) of preschool children (n = 107; mean child age = 4.3 years; 53% female) were assessed before, after, and 6 months following the intervention. Results indicated significant associations between lower levels of key outcomes at the pretreatment assessment and greater improvement after treatment in PRF, child self-regulation, and child self-distraction strategies. Furthermore, two subgroups of families were uncovered based on individual parental and family-environment characteristics: low-distress and high-distress parents. Following intervention, the high-distress group showed greater improvement in parental sensitivity and child problem behavior, whereas the low-distress group reported greater improvement in locus of control. Clinical and future directions are discussed.

KEYWORDS:

• Reflective functioning
• mentalization
• parenting
• behavioral problems
• parent-child interaction

Parental reflective functioning (PRF) is a core component of the parent-child relationship; it is a parent’s capacity to hold, regulate, and fully experience one’s own emotions as well as the emotions of others in a nondefensive way, without becoming overwhelmed or shutting down (Slade, 2005). PRF is considered to be the manifestation of a parent’s mentalizing capacity (Fonagy et al., 1991), and it seems to represent the parent’s ability to understand and interpret his/her child’s behavior in terms of mental states: feelings, thoughts, motivations, and beliefs of the child and himself/herself. In other words, PRF is a parent’s ability to treat his/her child as a psychological agent (Slade, 2003). Enhanced PRF strengthens the parent-child relationship, which leads to children who have better socioemotional health overall. Thus, PRF-based approaches, aimed at enhancing PRF to achieve favorable treatment outcomes (Steele & Steele, 2018), have become more prevalent. However, not all parents benefit from intervention to the same extent. Individual differences in parental and family-environment characteristics (such as parental distress levels) may moderate parents’ and children’s outcomes following intervention. For example, parents may vary on the extent of perceived distress, such that parental distress may cause resource depletion, rendering the parent less available for the therapeutic process. Our research moves beyond the question of intervention outcomes to a more mechanism-focused examination addressing the question of what works for whom when employing a PRF-based intervention.

PRF theory and measurement

PRF can be viewed and measured in a variety of ways (Schiborr et al., 2013), each emphasizing a different component of reflective functioning. The two most central ways focus on measuring 1) complexity and 2) spontaneity and accuracy. Complexity refers to the ability to process multiple perspectives and to recognize causal relations between behavior and mental state and the parent’s and the child’s inner worlds, especially in the presence of difficult and sensitive situations and emotions (Slade et al., 2013). This is commonly measured using the Parent Development Interview (PDI, Slade et al., 2003). Spontaneity and accuracy refer to a parent’s ability to use reflective language in real-time interactions and in a way that aligns with the child’s cues. This is commonly measured using the mind-mindedness (MM) observation tool (Meins & Fernyhough, 2010).

The hypothesis that PRF is important to the quality of the parent-child relationship is based on attachment theory. Attachment theory suggests that the quality of parent-child interactions contributes to the type of attachment bond formed, which in turn influences child development (Bowlby, 1958). PRF is thought to enhance parents’ ability to respond sensitively and appropriately to their children’s attachment-related needs, thereby creating attachment security. A higher PRF capacity has been associated with more secure child attachment, more sensitive parent-child interactions, and healthier child development, including child state-regulation capacity, social skills, and symbolic play (Fonagy, 2018).

PRF-Based interventions

Over the last decade, PRF-based therapy programs have gradually become more common, taking the theory and clinical base of RF and using it to create a framework for clinical work with children and their families, for example, GABI (Steele, Murphy, & Steele, 2010), SMART (Fearon et al., 2008), Minding the Baby (Sadler et al., 2013), and Mothering from the Inside Out (MIO; Suchman et al., 2016). These intervention programs aim to enhance PRF as a central element to achieve favorable treatment outcomes (Steele & Steele, 2018) and are based on the idea that PRF needs to be strengthened by being identified, validated, and developed (for a more thorough review, see Camoirano, 2017). Empirical studies have shown the effectiveness of PRF-based interventions for the parents, the children, and the parent-child relationship. Specifically, PRF abilities improved after PRF-based intervention (Author citation; Sadler et al., 2013). In addition, parent-child dyadic outcomes were found to improve after PRF-based intervention (Sadler et al., 2013). Finally, children whose parents attended PRF-based intervention exhibited fewer externalizing behavioral problems (Author citation; Ordway et al., 2014) and more self-regulation after intervention, compared to children whose parents did not participate in such an intervention (Suchman et al., 2016). Taken together, these studies indicate that PRF-based interventions are effective overall. However, because parents and children vary in the extent of their improvement, our study aimed to better understand “what works for whom” in PRF-based interventions. Specifically, we investigated how different pretreatment levels of key outcomes as well as parental and family-environment characteristics would relate to different responses to the PRF intervention.

Moran et al. (2004) were the first to go beyond the question of effectiveness (“what works?”) in parenting programs and refine the question to “what works for whom,” recognizing the fact that one size does not fit all in parenting programs. In their review, they addressed different components of parental intervention programs, such as recruitment and retention, improvement among parents, and improvement among children (Moran et al., 2004). Since then, an expanding body of literature (e.g. Lundahl et al., 2006) has made it clear that when it comes to explaining “what works for whom,” there is a need to consider parental, child, and family-environment characteristics. Identifying and understanding these specific and salient characteristics is crucial for the development of tailored parenting interventions, thereby improving the efficacy of treatment.

Moderating variables in intervention

Over the last two decades, there has been a vast interest in variables that might influence intervention outcomes. These variables are known as moderating variables (MV), and they refer to characteristics of the participant (e.g. pretreatment levels of key outcomes, sociodemographic status, emotional well-being, family support, etc.) that might influence intervention outcomes (Hinshaw, 2007). Uncovering MV can help identify subgroups of patients who might benefit from intervention. Identifying these subgroups based on individual characteristics aids in matching a specific intervention to the participants, which in turn increases the effectiveness of the intervention and results in greater efficacy overall. Furthermore, when considering the effectiveness of parenting interventions, MV might be related to pretreatment levels of key outcomes as well as to individual parental and family-environment characteristics (e.g. Gardner et al., 2010; Lundahl et al., 2006).

Moderating variables in PRF intervention

The examination of MV in PRF intervention studies is scant. The few findings available indicate that, in PRF-based intervention, child age acted as a moderator but parental gender and role (mothers/fathers) and child behaviors (clinical vs. nonclinical) did not (Lo & Wong, 2020). However, to the best of our knowledge, our study is the first to comprehensively investigate─in a single study─pretreatment levels of key outcomes and parental and family-environment characteristics as MV in PRF intervention.

Pretreatment levels of key outcomes as Moderating variables

Children’s and parents’ pretreatment capacities have been found to affect the efficiency of intervention. For example, in a randomized control trial (RCT), childrens’ pretreatment language skills and pretreatment social-interaction abilities were found to moderate the impact of interventions for children diagnosed with autism spectrum disorder (ASD; Begeer et al., 2015; Siller et al., 2013). Pretreatment levels of oppositional defiant disorder were also found to be a significant predictor of treatment outcome for children diagnosed with attention-deficit/hyperactivity (ADHD) when examining children before and after intervention using the Incredible Years (IY) intervention (Webster-Stratton et al., 2013). In addition, using data from a RCT study, researchers found that children’s self-worth and coping style, as well as a secure parent-child relationship moderated the outcome of parental group intervention (Scholten et al., 2015). In a recent systematic review of the role of parental characteristics in parent-mediated interventions for children with ASD, researchers found that parental stress and socioeconomic status were related to children’s outcomes, with varying effects depending on the specific treatment and outcome examined (Shalev et al., 2020).

However, much of the research on parent and family characteristics as treatment moderators has been conducted in more behavioral and non-PRF interventions. Therefore, a deeper understanding of the relationship between parental and family variables and child outcomes in PRF treatment is needed to uncover key mechanisms of therapeutic change.

Parental characteristics as MV

Parental mental distress seems to play a major role in intervention, particularly parental depressive symptoms, stress, and regulation abilities. Parental depression is perhaps the most studied MV and is considered a predictor of less optimal parental intervention response when comparing pre- and postintervention (e.g. Webster-Stratton & Hammond, 1990); reduced effects in parenting outcomes are seen in parents with higher levels of depressive symptoms (McTaggart & Sanders, 2007). Moreover, researchers have found RF and depression to be related. Several studies indicate that depressed patients showed lower levels of RF compared to healthy individuals (e.g. Murri et al., 2017). Due to its link to RF and its role as a MV in other parenting interventions, we propose that parental depression will act as a MV in PRF intervention.

Another proposed MV is parental stress. Parental stress is related to multiple facets of parenting interventions, such as participation, cooperation, and outcome of parental intervention. Indeed, using RCT data, research found that adding an additional component aimed at reducing parental stress to an evidence-based therapy enhanced therapeutic change for children and parents by reducing the barriers that parents experienced during treatment (Kazdin & Whitley, 2003). Parental stress also predicted lower levels of effectiveness in parental interventions (Kazdin, 1995). Furthermore, researchers have found parental stress to be related to PRF and specifically to prementalization, meaning that parents with higher levels of parental stress were less able to enter their child’s internal subjective world, compared to parents with lower levels of parental stress (Luyten et al., 2017). We therefore examined the role of parental stress as a MV in PRF intervention.

Finally, researchers have found emotion regulation to be related to PRF. Specifically, mothers with higher tendencies to suppress their emotions and more difficulties with emotion regulation had higher levels of nonmentalizing engagement with their child. Mothers with poorer emotional awareness also showed less interest and curiosity in their child’s mental states. Likewise, mothers who reported greater difficulty in setting goals also showed a reduced capacity to recognize that their infant’s mental states were not directly observable (Schultheis et al., 2019). Emotion regulation has been previously studied as a mechanism of change in intervention studies (Gratz et al., 2015). However, little is known about the role of emotion regulation in a parenting intervention or its role as a MV. Due to its central role in PRF, we propose that difficulties in emotion regulation among parents may be related to lower effectiveness in PRF intervention programs.

Family-environment characteristics as MV

There are several family characteristics that may be of relevance to parental intervention: marital satisfaction and household chaos seem to be two salient variables. Low levels of marital satisfaction seem to be a cause of distress in parenthood; likewise, low levels of marital satisfaction have been associated with more negative parenting behaviors and elevated levels of maternal depression and dysfunctional parent-child relationships (Robinson & Neece, 2015). In a study examining the relation between RF and marital quality during the transition to parenthood, researchers found that higher levels of RF were associated with higher levels of positive marital and coparenting interactions. Specifically, women who were better able to reflect on their early experiences with their parents had marital interactions that were more positive and supportive and less conflicted and undermining (Jessee et al., 2018). Following Bronfenbrenner’s ecological model (Bronfenbrenner, 1992) proposing that the distal environment may affect the more proximal environment, we examined the role of marital satisfaction as a MV in PRF intervention programs. This examination is based on the premise that marital satisfaction has an important role in parenthood (Fincham & Hall, 2005), and particularly in PRF, therefore, it is possible that when parents experience more distress in their marital relationship, they may be less available to be invested in intervention-affecting parenting.

Chaotic home environment is another prominent distress in parenthood and it has been associated with diverse adverse outcomes, such as parental stress and emotional distress, parenting difficulties, less effective parental discipline, reduced ability to understand and respond to social cues in children, reduced accuracy and efficiency in a cooperative parent-child interactional tasks, and child problem behavior (Farrington & Loeber, 1999). A chaotic home environment acted as a risk factor that moderated the impact of parenting on child problem behavior. In particular, a highly chaotic home environment was found to be an exacerbating factor for the relationship between parenting (warmth and hostility) and child problem behavior (Coldwell et al., 2006). With respect to PRF, a chaotic home environment also acted as a risk factor that moderated the impact of PRF on parent-child and triadic interactions, such that PRF during mother-infant interactions was related to both maternal sensitivity and the quality of family triadic interactions under low cumulative stressful contexts, but not under high cumulative stressful contexts (e.g. premature birth and elevated levels of household chaos; Author citation). Therefore, the role of a chaotic home environment was also investigated as a MV for PRF intervention programs.

Does parental and family-environment distress necessarily predict poorer or alternatively better intervention outcome?

Research is inconsistent regarding the effect of distress on intervention. On the one hand, there is research to suggest that parental and family-environment distress predicts poorer intervention outcomes, implying that parenting interventions generally are less successful for parents with higher levels of distress. Parental and family-environment distress levels are believed to undermine the efficacy of parent-training interventions by disrupting parent-training processes and implementation of recommendations (Lundahl et al., 2006). Elevated levels of personal and family-environment distress may cause resource depletion, rendering the parent less available for the therapeutic process. Moreover, distress affects a parent’s mentalization process, which becomes less controlled and more automatic, as suggested by the dual-system model of social cognition and mentalization (Luyten & Fonagy, 2015). In contrast, however, some large trials found no adverse effects of parental distress on child-intervention outcomes, in both community preventive and clinic-referred samples. Indeed, some studies found that individual parental distress, such as elevated levels of parental depression and stress, were related to favorable intervention outcomes (Gardner et al., 2010). One possible explanation for these finding might be that distressed parents tend to rate their children’s behavioral problems more severely than nondepressed, nonstressed parents (Webster-Stratton & Hammond, 1990), which could account for the greater reported benefit for their children, particularly if the parenting intervention leads to a decrease in parental depression or stress (Gardner et al., 2010). Another possible explanation is that parents with higher levels of mental distress are less available to their children and therefore their children show more behavioral problems. Following an intervention program, parents have learned to read their children’s signals and accordingly to provide their children with a more accurate holding environment. Thus, we examined two competing hypotheses in this study: would parental and family-environment distress predict poorer or more favorable intervention outcomes

DUET parenting model: a PRF group intervention

The DUET parenting model (Author citation) is based on the Reflective Parenting Program (Grienenberger et al., 2004) and has been adapted and further developed for Israeli parents. It is a 12-week program in which parents of preschool children meet in a group setting for a weekly, 90-minute session. Each meeting is delivered by two facilitators using a structured curriculum. The program encourages parental engagement in an in-depth experiential learning process designed to enhance critical parenting skills. Each of the workshops is organized around a central parenting topic, such as temperament, distress, separation, play, discipline, anger, or trauma. Each meeting includes a short topic introduction (i.e. psychoeducation), group discussion, and exercises to enhance parental use of reflective thinking while inviting parents to relate to issues in their own families. A special emphasis is given to the notion that children’s feelings and thoughts are important and require attention and validation. The meetings are based on the premise that parents need to connect to their own feelings and thoughts before they can relate to their child’s inner world. Parents are encouraged to discover new ways of thinking about the links between behaviors while learning strategies and techniques designed to enhance PRF. Parents are inspired – through role play and brainstorming – to think about their own thoughts and feelings and their children’s thoughts and feelings, in daily situations. The group enables parents to consider multiple emotions and thoughts that might emerge in a given situation. Parents are encouraged to use these ideas to explore, in a real-life situation, possible solutions for conflicts in the relationship with their children and to choose a suitable reaction instead of an automatic reaction. Emphasis is placed on parents recognizing how their capacity for reflective thinking can serve as a powerful tool that enhances their emotional availability and increases their attunement and sensitivity in their relationship with their children.

The present study

Our study investigated the impact of pretreatment levels as well as parental and family-environment distress as MV in PRF-based intervention. More specifically, we examined the role of pretreatment capacities (i.e. PRF, parental sense of control and competence, parent-child interactions, and child outcomes), parental distress (i.e. parental depression, parental stress, and emotion regulation), and family-environment distress (i.e. marital dissatisfaction and home chaos) as moderators when looking at the improvement in PRF and parental sense of control and competence, parent-child interactions, and child outcomes following the DUET intervention program. We examined whether pretreatment capacities, individual parental characteristics, and family environment would predict change in outcome measures. More specifically: A) Do pretreatment levels predict change? B) Do different subgroups (classes) exist based on individual parental characteristics (depression, anxiety, stress, emotion regulation) and family environment (marital dissatisfaction, home chaos)? C) Do these classes differ in the extent of change in the outcome measures? and D) Do the pretreatment scores of the outcome measures differ between these classes? And if so, are these differences predictive of change?

Methods

Procedure

Families were recruited from a community clinic as well as from the general population through kindergartens and advertisements in the community. Parents were offered an opportunity to participate in a parenting-intervention group where they could reflect upon their child, their parenting, and their relationship with their child while sharing everyday situations. All parents reported parent-child relationship challenges and concerns and were interested in better understanding their child’s behaviors as well as working on their relationship with their child. Children were aged 2–6 years old, in accordance with the DUET program for early childhood (see above for a more detailed description of the intervention). Inclusion criteria were children with typical development and parents’ ability to speak Hebrew. All families who met the inclusion criteria were invited to participate in the study. The study received Institutional Review Board approval. Interested parents signed a consent form and were assigned to a group intervention.

Interested families were visited at home for approximately 1.5 hours. Two researchers conducted home visits for study data collection, during which structured parent-child interactions were videotaped for 15 minutes, children were assessed, and parents were interviewed and asked to complete questionnaires. A total of three home visits took place at the start of the intervention (i.e. before), at the end of intervention (i.e. four months after start of intervention; M time between T1 and T2 = 120.24 days; SD = 59.51 days), and 6 months after the end of intervention (i.e. follow-up; M time between T2 and T3 = 192.28 days; SD = 46.91 days).

Intervention

Each DUET group was led by two trained psychologists or social workers. The training included a two-day introduction to the DUET reflective parenting program and was followed by weekly supervised meetings, where the facilitators received the DUET reflective parenting manual (Author citation) and met weekly for 1 hour of supervised discussion on reflective practices. The supervision included the review of both discussion of the protocol (i.e. in-session protocol implementation, review of the next week manual content) and discussion of the group work and the reflective process of the group as a whole as well as the individuals (e.g. discussion of specific examples from the group, working on the facilitators’ countertransference). The trained facilitators conducted the group sessions as described above, according to the structured manual; each group included, on average, six participants. In addition to the weekly supervised group meeting, fidelity measures were taken by an external examiner. All sessions were recorded, allowing an assessment of fidelity to the manual; 20% is most commonly used to measure fidelity. Therefore, two group meetings (out of the twelve group meetings in the program) were observed by trained research assistants and 91% of fidelity checklist items were performed.

Sample

Our study included 107 Jewish, Hebrew-speaking families living in the southern Israel. Of the 107 families that were included in the current study, 98 mothers agreed to participate (91.6% participation rate for mothers; 8.4% nonparticipation rate for mothers) and 44 fathers agreed to participate (41.0% participation rate for fathers; 59.0% nonparticipation rate for fathers). In other words, for 35 families both mothers and fathers participated; for 63 families only the mother participated; and for 9 families only the father participated, thus a total of 142 participants were included in this study. Concerning attrition, from the 98 mothers that participated in the study (data for at least one study timepoint), 0% had full missing questionnaire/interview data at the preintervention timepoint; 34.3% had full missing questionnaire/interview data at the postintervention timepoint; and 41.4% had full missing questionnaire/interview data at the follow-up timepoint. Of the 44 fathers that participated in the study (data for at least one study timepoint), 2.3% had full missing questionnaire/interview data at the preintervention timepoint; 46.5% had full missing questionnaire/interview data at the postintervention timepoint; and 46.5% had full missing questionnaire/interview data at the follow-up timepoint. Further information about missing data is provided in the Results section.

Measures

Parental and family measures

PRF was assessed in two ways: an interview and an observation. The Parent Development Interview-Revised (PDI-R; Slade, Aber, Bresgi, Berger, & Kaplan, 2004) is an interview assessing PRF levels. Parents were interviewed about their experience as parents, about their child, and about their relationship with their child, using the Hebrew version of the PDI-R. Interviews were transcribed and coded using the Reflective Functioning Scoring Manual (Fonagy et al., 1998) for the PDI (the PDI-RF addendum to RF scoring manual; Slade et al., 2007). Scoring was based on a thorough reading of verbatim transcripts made from audiotapes of the PDI. Each answer was given a score ranging from “−1” (negative or bizarre PRF) or “0” (disorganized disavowal PRF) to “9” (full or exceptional PRF) based on whether it reflected that parents were aware of the nature of mental states, made active attempts to account for their child’s behavior in terms of underlying mental states, described transactions between their own and their child’s mental states and behaviors, or took a developmental perspective on mental states and their regulation. Overall, the scoring distinguishes between Negative to Limited RF (−1 to 3) and Moderate to High RF (5 to 9). Scores under 5 indicate either negative, absent, or low PRF, whereas scores of 5 and above indicate clear evidence of mentalizing capacities. An overall score was determined for each interview, based on the highest score across the questions (for more information see Slade et al., 2005). Coding was conducted by two trained coders who were blinded to any information regarding the participants and study hypotheses. Interrater agreement calculated on 15% of the interviews, coded by the two coders, was extremely high (.96). In this study, two subscales of the PDI-RF were calculated: parent-related RF and child-related RF. For the parent-related RF scale, mean score was calculated using all the questions that were parent related (e.g. “How have you changed since you became a parent?”), Cronbach’s α = .84.And for the child-related RF scale, mean score was calculated using all the questions that were child related (e.g. “Has your child ever felt rejected?”), Cronbach’s α = .75.Higher scores indicate a higher reflective capacity.

Mind-mindedness (MM; Meins & Fernyhough, 2010b) is an observation tool used to assess parents’ ability to reflect their children’s mental states during real-time, ongoing parent-child interactions. Each observation is transcribed and thereafter coded using the Interactional Mind-Mindedness Coding System (Meins & Fernyhough, 2010b). Parents’ speech and child’s speech were first transcribed and the total number of utterances each parent and child made was counted. All comments in which parents or children used mental-state-language regarding their own mind and each other’s minds (i.e. sentences that included words that referred to cognitive or affective states, such as like, want, happy, sad, angry, etc.) and comments in which parents talked on behalf of the children (as if they were giving words to the child’s cognitions and feelings) were marked as mind-related comments. Finally, coders classified each mind-related comment as “appropriate,” “nonattuned,” or “self-related.” A “parent appropriate” MM comment was coded when it reflected a plausible interpretation of a mental state behind the child’s observed behavior; for example, when a child tried to grab a ball and the mother said: “I see that you want to hold the ball.” A “parent nonattuned” MM comment was coded when it did not seem to match the current mental state, as interpreted by the observed child’s behaviors; for example, when the child eagerly played with the ball and the mother said: “Are you bored. Do you want another toy?” A “self-related” MM comment was coded when the parent made a comment regarding his/her own mental state (e.g. “I want to play with the ball” or “I am sad”). Similarly, a “child appropriate” MM comment was coded when it reflected the child’s interpretation of a mental state behind the parent’s observed behavior (e.g. when the mother smiled and the child said: “Mummy are you happy?”). A “child nonattuned” MM comment was coded when the child misinterpreted the parent’s mind (e.g. when the mother smiled and the child said: “Mummy why are you mad?”). To control for verbosity, MM scores were calculated as proportion scores out of the total number of utterances each parent and each child made during the interaction (regardless of whether they were mind-related). For children, the nonattuned-comment score was equal to zero, therefore, we had four final scores: parent-appropriate comments (higher score = higher use of appropriate comments), parent-nonattuned comments (higher score = higher use of non-attuned comments), parent self-related comments (higher score = higher use of self-related comments), and child-appropriate comments (higher score = higher use of child appropriate comments). Eighteen percent of the total number of videos was coded by all three coders, blinded to any information regarding the participants, group belonging, and study hypotheses. Intraclass correlation coefficients were high (.79 to .97).

Parental efficacy was measured using two questionnaires: The Parental Locus of Control (PLOC; Campis et al., 1986), a-27 items, reflecting parental responsibility, child’s control of parents’ life, and parents’ control of child’s behavior (α = .84); and the Parenting sense of competence (PSOC; Johnston & Mash, 1989), a 17-item questionnaire, reflecting parental satisfaction and efficacy (α = .83).

Parental depression was assessed using the Center for Epidemiological Studies Depression Scale (Radloff, 1977), a 20-item inventory of depression symptoms (α =.86). A higher score means an elevated level of depression.

Parental anxiety was assessed using the 20-item state anxiety scale (α = .89 to .93), from the State-Trait Anxiety Inventory (Spielberger, 1970). A higher score means an elevated level of anxiety.

Parental stress was assessed using the mean of the 36-item Parenting Stress Index; short-form (PSI-SF; Abidin, 1990), which includes items assessing stress related to tasks and responsibilities of the parental role and stress related to child temperament and behavior (α = .82 to .86). A higher score means an elevated level of stress.

Emotion regulation was assessed using the total score of the Difficulties in Emotion Regulation Scale (DERS; Gratz & Roemer, 2004), which is the mean score of a 36-item measure that assesses emotional awareness, acceptance, perceived ability to regulate emotion, and the degree to which emotion dysregulation interferes with goal-directed activity (α = .92 to .93). A higher score means greater difficulty in emotion regulation.

Parental marital dissatisfaction was assessed using the Dyadic Adjustment Scale (DAS; Spanier, 1976b), which is the mean score of a 48-item measure. The DAS is an instrument widely used to measure overall marital satisfaction (e.g. “In general, how often do you think that things between you and your partner are going well?;” α = .88). A higher score means less marital satisfaction.

Home chaos was assessed using the mean score of the short version (6 items) of the Confusion, Hubbub, and Order Scale (CHAOS; Matheny et al., 1995) questionnaire (e.g. “You can’t hear yourself think in our home;” α = .66). A higher score means a more chaotic home.

Interaction measures

Parental sensitivity

The free-play interactions were coded using the sensitivity scale from the Emotional Availability Scales (EAS; 4th edition; Biringen, 2008). Three trained research assistants coded the interactions, blinded to any information regarding the participants, group belonging, and study hypotheses. Interrater agreement calculated on 20% of the dyads coded by the three coders was high (ranging between .86 and .94). A higher score means more sensitive parenting.

Child behavior

Child Behavior Checklist (CBCL-Preschool version; Achenbach & Edelbrock, 1979) is a 113-item questionnaire measuring parental reports of child behavioral problems. The total behavioral problem scale (total CBCL) was used in this study, reflecting overall severity of child dysfunction and adjustment difficulties as reported by parents (α = .96). For each child, scores of mothers and fathers were averaged to create a total score. A higher score means elevated levels of child problem behaviors.

Social Skills Rating System (SSRS: Gresham & Elliot, 1990) is a 31-item measure assessing child’s self-control, assertion and initiative, and cooperation (α = .83). For each child, scores of mothers and fathers were averaged to create a total score. A higher score means better child social skills.

Child self-regulation was measured using the snack-delay task (Kochanska et al., 2000), in which children were asked to withhold from eating or touching a piece of candy placed in front of them (under a clear, plastic cup) until the experimenter rang a bell. There was a practice trial followed by five real, timed trials (20 s, 30 s, 40 s, 10 s, and 60 s). The self-regulation score was based on three scales: 1) Follows directions: did the child follow the directions; 2) Waits for bell: did the child wait for the bell to ring before eating the candy; and 3) Expectation behavior: to what extent did the child exhibit attentiveness or impatience for the bell ring and the subsequent approval to eat the snack. Three trained research assistants coded the interactions, blinded to any information regarding the participants, group belonging, and study hypotheses, achieving agreement of .95 on 20% of the sample. Principal component analysis revealed good internal consistency between all three scales, with 43% of the variance accounting for child task (loadings from 0.61 to 0.94). Therefore, all three scales were averaged into a single “self-regulation” score. A higher score means better child self-regulation.

Child self-distraction strategies were measured using the snack-delay task (Kochanska et al., 2000). The self-distraction score was based on three scales: 1) Fidgeting time: time until first fidget; 2) Other’s encouragement: the experimenter had to remind the child to eat the candy; and 3) Diversion time: time until the child’s first diversion activity. Three trained research assistants coded the interactions, blinded to any information regarding the participants, group belonging, and study hypotheses, achieving agreement of .95 on 20% of the sample. Principal component analysis revealed good internal consistency between all three scales, with 20% of the variance accounting for child task (loadings from 0.55 to 0.66). Therefore, all three scales were averaged into a single “self-distraction” score. A higher score means more self-distracting strategies.

Data analytic approach

All analyses included both mothers and fathers. Latent class analysis (LCA) was used to identify latent classes of parents with comparable patterns of observed individual characteristics (depression, anxiety, parental stress, emotion regulation) and family environment (marital dissatisfaction, home chaos). When conducting LCA, a number of models are fit to the data, stepwise, beginning with a one-class model (the most parsimonious solution) and, subsequently, the number of latent classes increases. More specifically, we investigated a total of five models (one- to five-class solutions). The best-fitting model (class-solution) was chosen based on theoretical interpretability (Nylund et al., 2007), in addition to the following fit indices: Bayesian Information Criteria (BIC; smaller values indicate better model fit), Akaike Information Criterion (AIC; Akaike, 1987); Lo-Mendell-Rubin Likelihood Ratio Test (LMRL; significant values indicate a better model fit than the k-1 class solution [Lo et al., 2001]); and entropy (values closer to 1 represent better discrimination among latent classes; Vermunt & Magidson, 2003). The Parametric Bootstrap Likelihood Ratio Test (BLRT) was not considered due to potential bias for bootstrapped analyses with clustering (Nylund-Gibson & Choi, 2018). In addition, “classification probability of most likely latent class membership” was taken into account (probabilities >0.80 were considered adequate [Geiser, 2013]) as well as class size (classes including < 5% of all participants were rejected for sake of parsimony). In line with standard recommendations, variances were fixed across classes (Collins & Lanza, 2009). Once the number of classes was determined, continuous posterior probabilities of class membership were computed for each participant based on Bayes’ theorem and subsequently used as a predictor in further analyses.

For main analyses, a set of latent growth models (LGMs; Duncan et al., 2013) was run. First, an unconditional LGM was run for each outcome variable separately, to estimate overall change as well as to examine whether pretreatment scores (i.e. intercept) were predictive of change (research question A). Subsequently, probability of LCA class membership was added to the model to assess its associations with intercept (research question D) and/or slope (research question C). As per common practice, potential dependence of data was accounted for through clustering by family (Hanley et al., 2003) as well as through adding a dummy covariate for each intervention group (1 = member; 0 = no member; Robinson, 2003) that indicated whether an individual participated in the intervention in this specific group or in a different group. In addition, recruitment method (community clinic versus general population), timing of data collection (days between T1 and T2 data collection; days between T2 and T3 data collection) and other demographic characteristics (child age, child gender, child birth order, parent age, parent gender, parent place of birth, parent education, family constellation) were controlled for in all analyses; however, to limit power-related bias, covariates were removed from the model if they did not significantly contribute to it (i.e. p > .05).

For all LGMs, unstandardized results were reported and model fit was assessed by means of the following indices: Chi-square goodness of fit (nonsignificant values indicate good model fit), comparative fit index (CFI; values > 0.80 indicate good model fit), root mean square error of approximation (RMSEA; values < 0.08 indicate good model fit), and the standardized root mean square residual (SRMR; values < 0.08 indicate good model fit; [see Akaike, 1987; Byrne, 2012]). All LGMs included raw scores as assessed at three timepoints: start of intervention (i.e. before), end of intervention (i.e. four months after start of intervention), and six months after end of intervention (i.e. follow-up).

For all latent class and growth models, missing data were handled through “full information maximum likelihood” (FIML; Graham et al., 2007) estimation, which has been shown to be reliable even in cases with up to 70% of missing data (Graham et al., 2007), without risk of loss of power or requirement of increased sample size. FIML can be used under the assumption of “missing data completely at random” (MCAR), which was tested by means of Little’s MCAR test, and for which p > .05 is indicative of MCAR (Little, 1988). To avoid local solutions/maxima, the number of random starts and final stage optimalizations was set to 5,000, and model results were only interpreted on the condition that the best log-likelihood value was replicated (Hipp & Bauer, 2006). All analyses were conducted in MPLUS software version 8 (Muthén & Muthén, 1998–2012). Simulation studies have indicated that sample sizes of 100 yield sufficient statistical power for LCA (Collins & Lanza, 2009) as well as LGM (Shi et al., 2021).

Results

Across all study variables, preintervention data was missing for 19%−29% of participants; postintervention data for 41%-43%, and follow-up data for 47%−49%. Little’s MCAR test supported the “missing completely at random” assumption (χ²[1617] = 1643; p = .673) and, as such, in all further analyses, missing data was handled by means of FIML.

For each outcome variable separately, we first ran an unconditional growth model (see Table 1a). Results indicated good fit for all models, with insignificant χ², CFI > 0.80, RMSEA < 0.08, and SRMR < 0.08. As reported in Table 1a, for most of the outcome variables, unconditional growth models identified significant slopes (for two variables slopes neared significance) in expected directions, indicating change across the three timepoints. However, for MM parent-appropriate comments, no such change was seen.

Table 1. Unconditional and conditional latent growth models for each outcome variable separately.

				a								b				c
				Unconditional Latent Growth Model								Conditional Latent Growth Model Including Probability of Membership in the “High Individual and Family Environment Difficulties” Class as Predictor
				△	SE	p		Intercept on △ (B)	SE	p		LCA class on Intercept (B)	SE	p			LCA class on △ (B)	SE	p
Parent
	Model 1	PDI- Self		0.09	0.05	0.050		−0.36	0.06	<0.001		−0.05	0.16	0.74			−0.02	0.09	0.844
	Model 2	PDI- Child		0.09	0.05	0.063		0.06	0.42	0.889		−0.13	0.15	0.40			0.03	0.09	0.719
	Model 3	MM Parent- Self-Related Comments		0.01	0.00	0.049		−0.24	0.22	0.284		0.02	0.01	0.14			−0.01	0.01	0.613
	Model 4	MM Parent Child-Appropriate Comments		0.01	0.01	0.322		−0.22	0.28	0.424		−0.00	0.02	0.80			−0.01	0.01	0.190
	Model 5	MM Parent-Nonattuned Comments		−0.00	0.00	0.052		−0.58	0.08	<0.001		0.00	0.00	0.84			−0.00	0.00	0.601
	Model 6	Parental Locus of Control		0.07	0.02	<0.001		0.27	0.36	0.454		−0.10	0.07	0.17			−0.10	0.04	0.018
	Model 7	Parental Sense of Competence		0.06	0.03	0.021		0.00	0.20	0.998		−0.52	0.12	<0.001			0.02	0.06	0.762
Child
	Model 8	Child Behavioral and Emotional Problems		−0.04	0.01	0.002		−0.05	0.15	0.742		0.20	0.05	<0.001			−0.05	0.02	0.041
	Model 9	Child Social Skills		0.07	0.01	<0.001		0.01	0.14	0.945		−0.22	0.06	<0.001			0.04	0.02	0.139
	Model 10	Child Self-Regulation Strategies		0.04	0.01	0.001		−0.48	0.05	<0.001		0.00	0.05	1.00			0.02	0.02	0.370
	Model 11	Child Self-Distraction Strategies		0.49	0.18	0.006		−0.76	0.13	<0.001		0.09	0.61	0.89			0.53	0.34	0.116
Parent-Child Interaction
	Model 12	Parental Sensitivity		0.13	0.06	0.032		0.10	0.37	0.794		−0.36	0.26	0.15			0.25	0.12	0.040

Note. PDI = Parent Development Interview; MM = Mind-Mindedness.

Research question A: associations between intercept (pretreatment levels) and slope (change)

For the majority of the outcome variables, intercepts were unrelated to slopes, or in other words, pretreatment strengths and difficulties were not predictive of change. Nevertheless, a significant association between intercept and slope was observed for PDI-self and MM parent-nonattuned comments, child self-regulation strategies, and child self-distraction strategies. More specifically, for these four variables, regression coefficients indicated that lower levels of pretreatment capacities were predictive of greater change. Follow-up analyses were separately performed for these four outcome variables, each time dividing the sample into two groups of participants: one with intercepts below the median and one with intercepts above the median, and subsequently running the unconditional LGM in both samples separately. Importantly, the same results were seen for all four outcome variables, in that change was only significant in the group of participants with lower levels of pretreatment capacities (see Figure 1). Additional follow-up correlation analyses were run between dummy indicators of these four variables (i.e. 0 = below the median pretreatment value;1 = above or equal to the median pretreatment value), in order to understand whether families who had stronger pretreatment capacities on one variable were also more likely to have stronger pretreatment capacities on other variables. Specifically, non-significant correlations between various dummy indicators would suggest that participants with ceiling effects on some outcomes do still have room for change on other variables. Results indicated only one significant correlation between low pretreatment scores for MM parent-non attuned comments and high pretreatment scores for child distraction strategies (r = −0.47; p = .009). As such, our data are of little support to the idea that families who showed ceiling effects for one outcome variable were also more likely to show ceiling effects for other variables.

Figure 1. Posthoc analyses for significant associations between latent growth model intercept and change.

Are parents’ individual characteristics and family environment predictive of change?

Research question B: latent class analysis

To uncover the different subgroups (profiles/classes) in our study sample, based on individual parental characteristics and family-environment characteristics, LCA with FIML was run, including six continuous parent-report measures: depression, state anxiety, parental stress, emotion regulation; marital dissatisfaction, home chaos. Latent class models with one to five classes were estimated and fit indices are presented in Table 2. The LMRL fit index indicated that compared to the two-class solution, the three-class solution had a significantly better fit (p < .10), with no significant differences between the other class solutions (ps > .128). In addition, the biggest decrease in BIC and AIC was observed from two classes to three classes. At 0.916, entropy was high for the three-class solution. Because fit indices were, overall, in favor of the three-class solution, we further explored this option for theoretical interpretability.

Table 2. Latent class analysis: identifying the optimal number of latent classes.

	Fit Indices				Class Size: Proportion Participants in Each Class					Classification Probability: Most Likely Latent Class Membership
	Entropy	BIC	AIC	LMRL	1	2	3	4	5	1	2	3	4	5
Class solution
1 class	/	1112	1080	/	1.00					1.00
2 classes	0.83	1010	959	−528	0.32	0.68				0.93	0.96
3 classes	0.92	957	887	−460*	0.01	0.72	0.27			0.97	0.98	0.92
4 classes	0.91	955	866	−417	0.29	0.01	0.09	0.62		0.92	0.97	0.92	0.96
5 classes	0.93	956	848	−400	0.29	0.03	0.61	0.06	0.01	0.94	0.98	0.97	0.89	0.97

Note. * p < .01; “Fit indices” indicate how well each class solution fits the data; “Classification probability: Most likely latent class membership” indicates, per class solution, for each class, the probability of class membership, thereby identifying homogeneity within each class. BIC = Bayesian Information Criterion; AIC = Akaike Information Criterion; LMRL = Lo-Mendell-Rubin Likelihood Ratio Test.

Table 3 presents descriptive data for the two main classes separately. Importantly, the third class included one single participant, who relative to the other participants, reported high levels of parental stress, high levels of emotion regulatory problems, high levels of home chaos, low levels of depression, low levels of state anxiety, and low levels of marital dissatisfaction. As such, this person was likely to experience specific difficulties concerning regulatory functions. Regardless of its theoretical meaningfulness and importance, this class was excluded from further analyses due to its limited sample size. Figure 2 presents the standardized scores for the first two classes only, allowing for a more detailed scaling of the Y-axis and a better understanding of specific differences between the classes. The first class comprised 73.4% of the LCA sample (80 parents; 23 fathers, 57 mothers) and represented low levels of self-reported difficulties across all domains (i.e. “Low Individual and Family Environment Difficulties” class). The second class comprised 25.7% of the LCA sample (28 parents; 8 fathers, 20 mothers) and represented high levels of self-reported difficulties across all domains (i.e. “High Individual and Family Environment Difficulties” class). Independent sample t-tests indicated that compared to parents in the low-level class, parents in the high-level class had significantly higher scores on all measures (ps > .001).

Figure 2. Three-class solution (class 1 and class 2, without class 3): Z-scores for individual characteristics and family environment measures per class and change.

Table 3. Descriptive data for the two main classes separately.

			“clinical” cut-off	First Class Mean (low distress)	Second Class Mean (high distress)	Difference Statistic
Variables Included in Latent Class Analysis
	Parental Depression		0.8 and above^1	0.53 (0.29)	1.20 (0.34)	t(103)=-9.96; p < 0.001
	Parental State Anxiety		2 and above^2	1.56 (0.40)	2.16 (0.51)	t(103)=-6.37; p < 0.001
	Parental Stress		2.80 and above (85th percentile)^3	2.12 (0.20)	2.77 (0.51)	t(101)=-6.68; p < 0.001
	Parental Emotion Regulation		2.22 and above^4	1.78 (0.36)	2.36 (0.49)	t(105)=-6.58; p < 0.001
	Marital Dissatisfaction		3.50 and below^5	2.09 (0.40)	3.07 (0.75)	t(28.6)=-6.23; p < 0.001
	Home Chaos		2.5 and above (75th percentile)^6	1.91 (0.58)	2.53 (0.46)	t(103)=-5.12; p < 0.001
Demographic Data
	Child’s Age (Years)			3.98 (1.71)	4.04 (1.46)	t(87)=-0.12; p = 0.905
	Mother’s Age (Years)			33.8 (4.52)	34.9 (4.28)	t(99)=-1.04; p = 0.299
	Father’s Age (Years)			36.0 (4.88)	36.1 (3.15)	t(90)=-1.09; p = 0.914
	Days Between T1 and T2 Data Collection			143.9 (97.78)	128.2 (64.97)	t(19)=-0.33; p = 0.743
	Days Between T2 and T3 Data Collection			184.3 (42.20)	179.5 (18.63)	t(18)=-0.22; p = 0.829
	Child’s Gender					χ2(1) = 0.000; p = 0.993
		Boy		48.1%	48.0%
		Girl		51.9%	52.0%
	Recruitment Type					χ2(1) = 0.686; p = 0.408
		Clinic		38.5%	29.2%
		General Population		61.5%	70.8%
	Mother’s Place of Birth					χ2(1) = 0.129; p = 0.720
		Israel		78.5%	75.0%
		Other		21.5%	25.0%
	Father’s Place of Birth					χ2(1) = 3.064; p = 0.080
		Israel		77.9%	95.0%
		Other		22.1%	5.0%
	Mother’s Education					χ2(3) = 6.281; p = 0.099
		8–12 Years of Education		11.5%	25.0%
		High School Education		12.8%	12.5%
		Higher Education		75.6%	58.3%
		Other		0.0%	4.2%
	Father’s Education					χ2(2) = 4.176; p = 0.124
		8–12 Years of Education		16.9%	35.0%
		High School Education		16.9%	5.0%
		Higher Education		66.2%	60.0%
		Other		0.0%	0.0%
	Family Constellation					χ2(2) = 6.373; p = 0.041
		Married		98.7%	88.0%
		Divorced		0.0%	4.0%
		Single		1.3%	8.0%
	Birth Order					χ2(3) = 4.052; p = 0.256
		First child		67.1%	63.2%
		Second child		20.5%	36.8%
		Third child		8.2%	0.0%
		Fourth child		4.1%	0.0%
	Family Religion
		Jewish		100.0%	100.0%	NA
		Other		0.0%	0.0%

Note. Mean score above or below (respectively) cut-off point. ¹(Thomas et al., 2001); ²(Vigneau & Cormier, 2008); ³(Abidin, 1990); ⁴(Haynos et al., 2015); ⁵(Spanier, 1976a); ⁶None.

Notably, as shown in Table 3, for all the variables, with the exception of home chaos, mean level in the “High Individual and Family Environment Difficulties” class was higher than the cut-off point, whereas mean level in the “Low Individual and Family Environment Difficulties” class was within the normal range. The distinction between these two classes can be supported theoretically and aligns with previous studies that have revealed comparable classes in general population families (e.g. Rosellini et al., 2021). As such, and in line with model fit indices and theory, we retained these two classes for further analyses.

Research question C: associations between LCA class membership and change in outcome measures

For each outcome variable, a conditional LGM was run separately, regressing probability of membership in the “High Individual and Family Environment Difficulties” class on slope (see Table 1c). For the majority of outcome variables, LCA class membership was unrelated to slope, suggesting that both classes of participants benefitted from the intervention. Nevertheless, for the outcome variables “Child Behavioral and Emotional Problems,” “Parental Locus of Control,” and “Parental Sensitivity” differential intervention effects were seen. More specifically, for “Child Behavioral and Emotional Problems,” both classes experienced significant change, although greater change was observed for the “High Individual and Family Environment Difficulties” class (see Figure 3a). For “Parental Locus of Control,” significant change was specific to individuals in the “Low Individual and Family Environment Difficulties” class (Figure 3b), whereas for “Parental Sensitivity,” significant change was unique to individuals in the “High Individual and Family Environment Difficulties” class (Figure 3c).

Figure 3. Posthoc analyses for significant associations between LCA class membership and change.

Research question D: associations between LCA class membership and pretreatment scores of outcome variables

For each separate outcome variable, the above discussed conditional LGMs also included the regression of probability of membership in the “High Individual and Family Environment Difficulties” class on intercept (see Table 1b). Results indicated significant associations between higher probability of membership in the “High Individual and Family Environment Difficulties” class and lower pretreatment levels of “Parenting Sense of Competence” as well as lower pretreatment levels of “Child Social Skills” and higher pretreatment levels of “Child Behavioral and Emotional Problems.” Interestingly, as outlined above, none of these three pretreatment scores were significantly associated with change in these variables, suggesting that none of the class-related differences in change were due to differences in pretreatment values.

Discussion

The goal of our study was to investigate whether pretreatment capacities as well as differences in individual parental and family-environment distress variables were related to improvement in parent outcomes, parent-child interaction outcomes, and child outcomes following the DUET intervention program.

What works for whom? The associations between pretreatment scores and intervention improvement

Aiming to uncover the importance of pretreatment scores for intervention outcomes, we examined, for each outcome variable, whether the starting point (pretreatment level) was related to the extent (slope) of improvement following the DUET intervention. Results indicated that parents who were less reflective about their own internal world and parents who, during real-time interactions, tended to misinterpret their child’s internal world had greater improvement in these capacities following intervention, compared to parents who had better pretreatment capacities. Moreover, results indicated that children who were less capable of regulating themselves and using beneficial distraction strategies had greater improvement in self-regulation and self-distraction strategies, respectively, following their parents’ participation in the DUET intervention. These findings support previous findings (e.g. Lundahl et al., 2006) indicating that children with more problem behavior tended to show better outcomes after parental intervention. These results might indicate ceiling effects. Indeed, among those parents who showed interest in the intervention due to worries about the relationship with their children, some might have shown actual great objective strengths in parental reflective functioning, parent-child interaction, and child outcomes. For these families, the intervention may not have had much additional value in these domains. If this were true, one could advise preintervention screenings and offer alternative interventions (such as anxiety-focused interventions) for families concerned about their parent-child relationship but with actual objective strengths in PRF, parent-child interaction, and child outcomes.

However, two sources of additional data suggest a different picture. First, ceiling effects were only seen for four out of 12 variables, meaning that for all other variables, significant intervention main effects were relevant for all parents, regardless of their starting capacities. Specifically, such main effects were seen for increased parental reflective abilities when thinking of their child’s inner world, as well as their own mind, as seen in self-related comments; increased parental sense of control and competence; decreased child behavioral and emotional problems; increased child social skills; and increased parental sensitivity. Second, an additional set of analyses indicated only one significant correlation among the four variables for which ceiling effects were observed. Specifically, ceiling effects for child distraction strategies were significantly correlated with ceiling effects for parents’ limited use of comments reflecting misinterpretation of their child’s internal world during real-time parent-child interactions. Nevertheless, other ceiling effects were not significantly correlated, meaning that even if a family showed ceiling effects for one or (maximum) two domains, they would still benefit from the intervention for all other domains. Thus, overall, our results indicate that although some families might have strong starting capacities in one or two domains that would prevent them from benefitting from the intervention within those domains, they would still benefit from the intervention in the other 10–11 domains. As such, our results do not support the practice of excluding families from PRF-based intervention based on high starting capacities within one or even a few subfields of PRF, parent-child interaction, or child outcomes.

Identifying subgroups based on parental and family-environment distress

The examination of parental and family-environment distress characteristics uncovered two subgroups of families: A high-level distress group (i.e. higher levels of parental depression, anxiety, stress, marital dissatisfaction, and home chaos, and lower levels of emotion regulation), and a low-level distress group (i.e. parents experiencing low levels of distress in all the above variables). Moreover, our results indicated that the mean-level scores in the parental and family-environment distress variables for the low-level distress group were within the normal range, whereas for the high-level distress group, the mean-level scores were higher than the cut-off point, for all the parental and family-environment distress variables (except for home chaos). This finding strengthens the idea that the two subgroups are clinically distinct from each other and supports previous studies that have demonstrated that elevated levels of parental depression, anxiety, stress, and marital dissatisfaction have often been found to coexist) (Odinka et al., 2018).

What works for whom? The associations between subgroup membership and intervention improvement

For the majority of outcome variables subgroup membership was unrelated to intervention outcome and all families benefitted from the DUET intervention by experiencing growth in their capacity of reflecting on their own inner world or their child’s inner world. Nevertheless, our results did reveal some interesting patterns of intervention effects that were different between subgroups. Firstly, and most importantly, children in the two groups showed a significant decrease in behavioral and emotional problems after their parents’ participation in the group intervention; however, the decrease was greater among children from high-distress families. As for the parents, all parents benefitted from the intervention, but in different domains. Specifically, parents in the low-distress subgroup reported significant improvement in parental locus of control, whereas parents in the high-distress subgroup did not. Additionally, parents in the high-distress group showed an increase in parental sensitivity during parent-child interactions after the intervention, whereas parents in the low-distress group did not show any significant change in sensitivity after intervention.

These findings add to the literature on parenting intervention programs, proposing that subgroup analysis can matter a great deal in intervention theory and research. Rothwell (2005) claims that the importance of subgroup analysis is in identifying how to maximize benefits from treatments. Moreover, moderated effects can help prevention scientists refine theory and tailor intervention to the needs of specific populations (Rothman, 2013). It is therefore important to investigate whether treatment effects vary among subgroups of patients defined by individual characteristics. As demonstrated in our study, subgroup analysis was indeed necessary to trace the different impacts of DUET intervention on the different subgroups. Our findings indicated that there is no group that overall benefits more from the DUET program, but rather, that each group benefitted differently.

Why is it that when parents experienced more distress there was a (greater) improvement in problem child behavior and parental sensitivity after intervention, whereas when parents reported having less distress, they was improvement in their parental locus of control after intervention? To answer this question, the core components of the DUET intervention need to be addressed. One of the main goals of the intervention is to enhance parents’ ability to think about what is going on in their child’s mind as well as in their own mind when considering different behaviors and actions. That is, to reflect upon their own and their child’s behaviors as being driven by feelings, thoughts, and motivations. Throughout the intervention, week after week, parents practiced using different techniques (e.g. role play, group brainstorming). They became more curious about their child’s internal world and were capable of perspective taking. Thus, throughout the intervention, they became aware of and better connected to their child’s emotional world. Because high-distress parents experienced more difficulties in their everyday life, before the DUET intervention they may have tended to pause and reflect less during interaction with their children, due to elevated levels of distress. Therefore, their growing PRF capacity after intervention seems to be related to the more basic connection to their child, by strengthening their ability to think and understand them better. That is, the parent became more aware of the child’s cues and signals during the interaction and thus were better able to interpret and respond to them in a sensitive manner (Author citation). Furthermore, the decrease in child problem behavior suggests that children react to a parent’s enhanced ability to connect to them, perhaps due to the child’s increased sense of emotional confidence in the parent and enhancement of affect regulation and self-control (Ensink & Mayes, 2010; Schultheis et al., 2019). It is possible that for highly distressed parents, thinking about their child’s internal world is repressed by daily stress and that being part of the DUET group intervention enabled them to put aside all daily hassles and focus on their child’s inner world and their relationship with their child. For example, having a short mindfulness exercise each week enabled parents to pause and move to a “reflective space” without distractions, in a structured manner.

As for the low-distress parents, the main improvement following the intervention was related to their own internal locus of control. That is, after the intervention, parents believed that they could better contribute to their children’s adjustment. This finding is related to another component of the DUET intervention, which is the enhancement of parental sense of efficacy. During the group sessions, parents were regarded as the experts of their child’s inner world. Through exercises that promote PRF, they were encouraged to think about what their children needed from them. It seems that during the DUET intervention, parents fully understood that their child’s behaviors were related to their child’s emotions, thoughts, feelings, and motivations, and were not random. Therefore, parents were better able to position themselves next to their child and feel that they mattered. Differently from the high-distress parents, it seems that the low-distress parents were able to better reflect on their role in their child’s behavior and development.

Follow-up analyses indicated that these differences in intervention outcomes for high-distressed versus low-distressed families were unrelated to differences in starting capacities. Taken together, our results regarding moderation of the subgroup belonging indicates that regardless of specific levels of parental distress, parents can benefit from the DUET intervention in different ways.

Study limitations and future directions

This study has several limitations to consider. First, the small sample size limited the analyses, thus a future study with a larger sample would enable the identification of potential additional subgroups, uncovering other processes of change following the DUET intervention. Second, the current study’s large amount of missing data and attrition must be considered as a clear limitation. Missing data and attrition are common in intervention studies (Rioux & Little, 2021). In the current study, FIML was employed to deal with missingness a posteriori, a technique that has been shown to be reliable even for large amounts of missing data, without risk of loss of power or requirement of increased sample size (Graham et al., 2007). Nonetheless, it is important to acknowledge that missingness and attrition might still have influenced the results. Study findings must be interpreted in light of this limitation and future studies should take additional measures to prevent missingness and attrition. Such an approach might focus specifically on increasing the number of participating fathers so that similarities and differences between paternal and maternal processes of change after intervention can be modeled. Finally, our study did not include a control group and future research may include it to control for nonintervention-related change. Although unlikely, various nonintervention-related factors may have potentially explained the current study’s findings of different trajectories of change depending on pretreatment capacities or characteristics. For example, time could affect the current study’s outcome variables differently depending on pretreatment levels or characteristics. In addition, one may wonder whether families with high problems and low baseline capacities might have naturally attracted other types of support that induced positive changes in child symptoms or PRF and sensitivity. Also, families with high problems and low pretreatment capacities might have shown “regression to the mean,” where variables that are extremely higher or extremely lower than average on the first measurement move closer to the average on the second measurement.

Clinical implications

This study emphasizes the importance of working on PRF as a means of change for the parent, the child, and the parent-child interaction. Some clinical implications can be drawn from this study. First, our study found a decrease in behavioral problems after the “Duet” program, as compared to the levels measured before the intervention. Therefore, it is recommended to include this program when working with parents of children having adjustment difficulties. Furthermore, the study’s results support the idea of tailored therapy by illustrating how effective intervention can help different families in different ways. It seems that different parents, having distinct levels of distress, may benefit in a different manner from intervention programs. Clinician should consider the different outcomes for the different parents, depending on their initial stress level, when admitting parents to intervention. For example, our results suggest that when the intervention aims are to increase parental sensitivity, DUET program can be a good program to use in particular for parents who experience higher levels of distress. The program supports their ability to better observe the child and then think what the child’s needs are and what can they offer to support these needs. However, if the intervention aims are to empower parental sense of efficacy, support parents’ awareness of themselves within the parents, the DUET intervention program is recommended in particular for parents who are have lower levels of distress.

Finally, the sample attrition should also be considered from a clinician perspective. It appears that about 30% of the sample did not complete the group meetings. The reasons given by the parents for dropping out are mainly lack of time and difficulty in finding an arrangement (babysitter) for their children during the meetings. Dropout is a major and well-known problem among clinicians and therapy scholars. In a research study carried out by the National Institute of Mental Health (NIMH) in the U.S., it was found that only two third of the participants completed 12 sessions of therapy (Wierzbicki & Pekarik, 1993). A meta-analysis of 125 outpatient therapy studies which examined factors related to attrition, indicated that on average, 47% of clients dropout of therapy. Considering this information, it seems that the dropout from the DUET treatment is relatively low, and yet, need to be addressed to increase the number of full participation. Clinicians should be aware of the risk factors for therapy dropout (see Egan & Kenny, 2005), in particular to facilitate parental participation (e.g. time of the group meetings), to increase parental motivation for therapy, and to support them along the process in order to ensure lower dropout rates.

Acknowledgments

This research was supported by Judge Leon S. Kaplan, who donated funds to Duet center. We would like to thank the group facilitators, the students who collected and coded data, the parents and their children for their ongoing participation in this project.

Disclosure statement

The author(s) have no relevant financial or non-financial interests to disclose.

Data availability statement

The data that support the findings of this study are available on request from the corresponding author.

Additional information

Funding

The work was supported by the This research was supported by Judge Leon S. Kaplan, who donated funds to Duet center.