Investigating active ingredients of the GenerationPMTO intervention: Predictors of post-intervention change trajectories in parenting practices

Abstract

Empirically determining the active ingredients of evidence-based parenting interventions is a promising means for strengthening interventions and enhancing their public health impact. This study aimed to determine which distinct ingredients of the GenerationPMTO (GenPMTO) intervention were associated with subsequent changes in parenting practices. Using a sample of 153 participants randomly assigned to the GenPMTO condition, we employed multilevel modeling to identify intervention ingredients empirically linked with change trajectories in parenting practices observed across the two years following intervention exposure. Coercive parenting and positive parenting outcomes were examined. Study results indicated that emotion regulation, effective communication, problem solving, and monitoring each demonstrated a significant pattern of findings for coercive parenting. Differential exposure to each of these ingredients significantly predicted the level of coercive parenting immediately post-intervention and/or trajectories of change in coercive parenting across the subsequent two-year period, controlling for coercive parenting at baseline. No significant predictors were found for positive parenting trajectories. Our findings suggest four components as active ingredients of the GenPMTO intervention for coercive parenting. Identification of these active ingredients may lead to strengthening future iterations of GenPMTO by expanding the set of core components specified in the model, which may further improve public health benefits. Implications for further understanding change stemming from evidence-based parenting interventions are also discussed.

Recognizing the meaningful influence of caregivers in the lives of children, parenting interventions seek to prevent parent-child relational problems and improve child behavior outcomes by promoting effective parenting practices (Holtrop et al., 2020; Sandler et al., 2011). Parenting interventions are a critical psychotherapy approach given the substantial public health burden posed by child behavior disorders—affecting an estimated 10 to 20% of youth worldwide (Kieling et al., 2011). Fortunately, many parenting interventions have demonstrated strong empirical support for preventing and treating a variety of child behavior problems (Kaminski & Claussen, 2017; Weisz & Kazdin, 2017b). However, a number of barriers have impeded the widespread implementation of evidence-based parenting interventions, and there remains a great need to expand their public health impact (Forgatch et al., 2013; Weisenmuller & Hilton, 2020).

One critical opportunity for addressing this gap is to better determine the active ingredients of these interventions (Abry et al., 2015; Blase & Fixsen, 2013). This can promote the development of more efficient, potent, and cost-effective interventions to be implemented in real-world settings and have a meaningful impact on improving public health (Blase & Fixsen, 2013; Collins et al., 2005; Spoth et al., 2013). In response, the guiding aim of this study was to investigate the active ingredients of the well-established GenerationPMTO (GenPMTO) parenting intervention (Forgatch & Gewirtz, 2017; Patterson, 2016).

Literature Review

Determining Active Ingredients of an Intervention

The classical approach to intervention development and testing involves building a multicomponent program from basic research findings and evaluating the overall treatment package in randomized trials (Guastaferro & Collins, 2019). While useful for determining effectiveness, an exclusive focus on condition assignment and intervention outcomes conceals much of the change process (Nelson et al., 2012). As such, it is difficult to discern the processes taking place during the intervention and which specific components produce change (Guastaferro & Collins, 2019; Nelson et al., 2012). To move science forward, it is necessary to advance beyond investigating if an intervention is effective toward more concerted efforts to examine how interventions achieve positive change (Blase & Fixsen, 2013; Weisz & Kazdin, 2017a).

Determining the active ingredients of an intervention is a crucial, yet understudied, aspect of this research agenda (Abry et al., 2015; Blase & Fixsen, 2013). Active ingredients are the specific elements of an intervention responsible for achieving change (Abry et al., 2015; Nock, 2007). More precisely, active ingredients refer to the intervention components that have been empirically linked to changes in outcomes – an important next step beyond hypothesizing components of change based solely on theory (Abry et al., 2015). Identifying active intervention ingredients is a crucial activity to advance intervention research, theory development, and implementation outcomes (Abry et al., 2015; Blase & Fixsen, 2013; Century & Cassata, 2016).

In practice, discerning causal factors is a complex and challenging endeavor (Blase & Fixsen, 2013; Gottfredson et al., 2015; Weisz & Kazdin, 2017a). For example, to examine linkages between intervention components and treatment outcomes during efficacy trials, there must be variation in the delivery of the intervention components; however, intervention trials are often highly standardized in ways that restrict such variation (Blase & Fixsen, 2013). In an alternative approach, researchers have sought to identify active ingredients by conducting entire trials for this purpose (e.g., dismantling designs), but such component treatment studies have been critiqued as resource-intensive and largely unsuccessful (Bell et al., 2013). In this study, we leverage existing, unused video data from a landmark prevention trial of the GenPMTO intervention (see Forgatch & DeGarmo, 1999; Patterson et al., 2010) to empirically examine the association between each observed intervention ingredient and parenting practices over time.

Focal Intervention: GenerationPMTO

GenPMTO, previously known as Parent Management Training—Oregon model (PMTO), is an evidence-based parenting intervention supported by over 50 years of research (Forgatch & Gewirtz, 2017; Forgatch & Patterson, 2010). It is based on coercion theory and the Social Interaction Learning Model, which provide a developmental perspective on the emergence of early problem behaviors and how they are shaped over time (e.g., Dishion et al., 2016; Patterson et al., 2010). GenPMTO is characterized by five theoretically-derived core components—skill encouragement, limit setting, problem solving, monitoring, and positive involvement—which operate as mediators of child outcomes (e.g., DeGarmo et al., 2004; Patterson et al., 2010). The intervention has been shown to prevent and treat a variety of behavior problems for youth from early childhood to adolescence, such as externalizing behaviors, delinquency, and arrests (Dishion et al., 2016; see also Akin et al., 2019). GenPMTO can also benefit parents and families. For example, it has shown positive effects for maternal depression (e.g., DeGarmo et al., 2004), standard of living outcomes (e.g., Patterson et al., 2010), risk of arrest (Patterson et al., 2010), and rates of family reunification following foster care placement (Akin & McDonald, 2018). GenPMTO has been implemented successfully with diverse populations in the US and around the world (Askeland et al., 2017; Forgatch & Gewirtz, 2017; Sigmarsdóttir et al., 2019).

GenPMTO Research.

Research on GenPMTO is particularly well-established for discerning mechanisms of change post-intervention (Forehand et al., 2014). Prior research has shown that assignment to the GenPMTO condition leads to a number of positive outcomes from baseline to nine-year follow-up (Forgatch et al., 2009; Patterson et al., 2010). Longitudinal models have demonstrated that these effects were mediated by changes in observed parenting. Specifically, GenPMTO mothers showed a significant reduction in coercion from baseline to one-year follow-up while control mothers showed an increase. This reduction in coercion mediated intervention benefits to growth in positive parenting from one to three years, which then served as the more proximal parenting mediator of the intervention benefits to delinquency, arrests, and socio-economic status (Forgatch et al., 2009; Patterson et al., 2010). These studies document a cascade of effects taking place following the intervention in which coercive and positive parenting practices are identified as the key mediators accounting for subsequent changes in child behavior (Forehand et al., 2014). The present study extends findings from this sample by evaluating the extent to which exposure to intervention ingredients, beyond just the previously identified core components, is associated with post-intervention changes in parenting.

As it currently stands, we know that intervention-induced changes in parenting are instrumental in bringing about better outcomes for families; however, there remains a pressing need to examine what takes place during the GenPMTO intervention that triggers these changes. The few existing studies in this area focus on linking intervention fidelity to parenting outcomes. Forgatch and colleagues (2005) and Forgatch and DeGarmo (2011) determined that higher levels of competent adherence to the delivery of two core components (i.e., skill encouragement, limit setting) predicted improvements in effective parenting. This research supports the important role of treatment fidelity in bringing about changes in parenting but does not fully open the black box to examine the role of exposure to the whole array of GenPMTO ingredients. Qualitative work with parents completing GenPMTO suggests that ingredients not designated as core components (e.g., clear directions, emotion regulation) were highly influential to parenting practices (Holtrop et al., 2014). In response, the goal of this study was to investigate the extent to which exposure to eight different GenPMTO ingredients was associated with changes in parenting.

GenPMTO Ingredients.

The core components of GenPMTO include five parenting practices (Forgatch et al., 2017). Skill encouragement uses contingent positive reinforcement and scaffolding strategies to strengthen child prosocial behaviors. Limit setting involves setting clear rules and applying consistent, mild consequences for misbehavior. Problem solving supports families in resolving disagreements and achieving shared goals. Monitoring entails supervising children to keep them safe and out of trouble. Positive involvement emphasizes the importance of love and affection in the parent-child relationship. In addition, three auxiliary skills are integral to the GenPMTO model. Clear directions provide a foundational set of skills for giving calm, specific, and effective directions to elicit child cooperation. Emotion regulation helps caregivers to identify emotions and apply emotion regulation strategies. Effective communication focuses on active listening and speaking skills to enhance interpersonal interaction. While these auxiliary skills have yet to be empirically validated, they are considered important supporting elements of the intervention (e.g., Forgatch & Gewirtz, 2017; Rains et al., 2021). Together, these eight ingredients were considered as potential active ingredients of GenPMTO in this study.

Current Study

The purpose of the current study was to determine the extent to which exposure to each of eight distinct ingredients of the GenPMTO intervention was associated with subsequent changes in the theoretically-based parenting constructs found to mediate positive intervention outcomes, i.e., coercive parenting and positive parenting (Forgatch et al., 2008; Patterson et al., 2010). To accomplish this aim, we rated video recordings of parent group sessions from a prior GenPMTO prevention trial targeting mothers with school-aged sons (grades 1–3). We examined two primary research questions: (1) Which GenPMTO ingredients are empirically linked with changes in coercive parenting practices?; and (2) Which GenPMTO ingredients are empirically linked with changes in positive parenting practices? The parenting practice outcomes were based on observations of parent-child interactions assessed four times over the two years post-intervention. We hypothesized that the amount of exposure to specific active ingredients would demonstrate a significant, independent association with changes in parenting outcomes post-intervention and therefore be implicated as core components of GenPMTO.

Method

Participants

A total of 238 recently separated mothers and their school-aged sons (grades 1–3) participated in the original study. Families were recruited through media advertisements, flyers distributed in the local community, and divorce court records. The mean age for mothers was 34.8 years (SD = 5.4). Their reported race/ethnicity was 88% Caucasian, 5% Hispanic, 3% Native American, 1% Asian/Pacific Islander, 1% African American, and 3% classified as “Other.” Most (76%) had completed some academic or vocational training beyond high school, although only 17% had a 4-year college degree. At baseline, mothers had been separated an average of 9.2 months. For the boys in the study, the average age was 7.8 years (SD = .93). Their reported race/ethnicity was 86% Caucasian, 2% Hispanic, 2% Native American, 1% African American, and 10% “Other.” This racial/ethnic distribution was reflective of the community in the Pacific Northwest at the time of the original study (Forgatch & DeGarmo, 1999). The average annual family income was $14,900 and 76% of families received public assistance.

A subset of 153 families were randomly assigned to the GenPMTO group. Mothers in this group had a mean age of 35.0 years (SD = 5.5) and had been separated an average of 9.8 months. The racial/ethnic composition for these mothers was slightly more Caucasian than the overall sample (92% vs. 88%). Boys in the GenPMTO group had a mean age of 7.6 years (SD = 0.93). The racial/ethnic composition for the boys in this group was also slightly more Caucasian than the overall sample (86% vs. 80%). Intervention group families had an average of 2.4 children living in the home (SD = 1.0) and an average annual family income of $15,400.

Data Source

This study used archival data from a NIH-funded trial examining the effectiveness of the GenPMTO intervention for recently separated mothers and their school-aged sons over a nine-year period (see Forgatch & DeGarmo, 1999; Forgatch et al., 2009). Families experiencing separation were considered an appropriate target for this prevention trial because of an increased risk for negative outcomes, including disruptions in positive parenting and child adjustment (Forgatch & DeGarmo, 1999). Families were randomly assigned to receive GenPMTO (n = 153) or a no intervention control (n = 85) using an unequal allocation strategy to ensure adequate sample size in the experimental condition. Multi-informant, multi-method assessments were conducted at several time points, including laboratory observations of mother-child interactions collected at baseline and for two years post-intervention. In addition, each GenPMTO intervention session was video recorded. For the current study, we measured the delivery of eight GenPMTO ingredients using the intervention session video data and then examined relationships between each ingredient and the parenting practice observed during the mother-child laboratory assessments. All study procedures were approved by the appropriate institutional review boards.

Intervention

Caregivers in the intervention group were invited to attend the GenPMTO program known as Parenting through Change (PTC; Forgatch, 1994). PTC is a group-based version of GenPMTO developed to support mothers recently experiencing separation or divorce (Forgatch & DeGarmo, 1999). In this trial, a total of 13 PTC groups were conducted, each lasting between 14 and 16 weeks. PTC followed the manualized curriculum (Forgatch, 1994). Weekly sessions lasted approximately 90 minutes and included reviewing and troubleshooting prior material, introducing and applying a new topic, and providing a home practice assignment. Session topics were: (1) working through change, (2) encouraging cooperation, (3) teaching through encouragement, (4) setting limits, (5) following through, (6) promoting school success, (7) communicating with children, (8) observing emotions, (9) managing/regulating emotions, (10) problem solving, (11) managing conflict, (12) building skills, (13) monitoring children’s activities, and (14) balancing work and play. If mothers missed a session, they were encouraged to visit another PTC group or participate in a make-up session with an interventionist.

Each PTC group was co-led by two interventionists, with eight in total delivering the intervention. All interventionists were female and varied in educational background and experience. To support effective program delivery, those without prior experience participated in a 2-to-4-month structured training program. The weekly PTC group sessions were video recorded for fidelity monitoring and supervision purposes. Further details about the original PTC trial are reported elsewhere (e.g., Forgatch & DeGarmo, 1999; Forgatch & DeGarmo, 2002).

Measures and Construct Scores

Parenting Practices

Laboratory observations were conducted at five time points in the original study: baseline (T0), post-intervention (T1), 6-months post-intervention (T2), 12-months post-intervention (T3), and 24-months post-intervention (T4). The baseline (T0) and post-intervention (T1) assessments took place six months apart, during which time the intervention was delivered. Mother-child dyads participated in a 45-minute structured interaction task, which included four problem solving discussions (20 minutes total), a teaching task (10 minutes), an unstructured activity (10 minutes), and a refreshment time (5 minutes). These interactions were video recorded and scored using two observational coding systems. The Interpersonal Process Code (IPC; Rusby et al., 1991) was used to code microsocial behaviors, including data about respondent and recipient, sequence, content, affect, activity/context, and duration. Behaviors were coded as positive if they were positive in content and delivered with positive or neutral affect (e.g., endearment expressed using a happy tone) or neutral in content and delivered with positive affect. Behaviors were coded as aversive if they were negative in content or affect (e.g., negative talk or hostile tone). A global rating system (Forgatch et al., 1992) was also applied to the interaction tasks to rate coder impressions of mother-child processes. Example items include “mom broke down tasks as necessary” and “family worked together as a team.” Approximately 15% of the interactions were scored by two randomly selected coders to assess for intercoder agreement. Average Cohen’s kappa has been reported as .78 (range = .77–.80) for IPC content codes and .70 (range = .67–.76) for affect across all five timepoints (e.g., DeGarmo et al., 2004; Martinez & Forgatch, 2001).

Instead of attempting to recreate the parenting practice outcomes used in prior studies (e.g., Forgatch & DeGarmo, 2002; Patterson et al., 2010), we derived two parenting practice constructs, positive parenting and coercive parenting, by subjecting the observed parenting practice variables to exploratory factor analysis (EFA). This allowed us to construct variables with the strongest possible factor structure specific to the sample and timepoints relevant to the current study. The EFA models were tested at T1, T2, T3, and T4 and consistently reported two distinct factors with identical indicators strongly loading within the same factor at each timepoint. In response to these results, we did not include ineffective discipline as an indicator of coercive parenting—a common practice in prior GenPMTO studies; instead, we rescaled each individual item to reflect effective limit setting and included this indicator in the positive parenting construct (see also DeGarmo et al., 2004). The resulting positive and coercive parenting factors remain theoretically consistent with coercion theory and the Social Interaction Learning Model underpinning the GenPMTO intervention (Forgatch & Patterson, 2010; Patterson et al, 2009). Since each parenting construct was comprised of multiple variables, we created mean construct scores to model change following a similar procedure to past GenPMTO studies (e.g., DeGarmo et al., 2004), which used methods outlined by Stoolmiller (1995).

Coercive Parenting.

Coercive parenting was operationalized as a combination of negative reinforcement and negative reciprocity. Both of these variables were measured based on behaviors taking place throughout the entire 45-minute structured interaction task. These laboratory observations were coded with the IPC coding system (Rusby et al., 1991).

Negative Reinforcement.

The negative reinforcement variable was based on instances of mother-child conflict. Coders observed when the mother initiated a conflictual interaction with an aversive behavior (e.g., negative talk, physical interaction) and scored this as an instance of negative reinforcement if the child responded with an aversive behavior that ended the interaction without another aversive behavior by the mother or child in the next 12 seconds. This was considered negative reinforcement because the child’s aversive behavior was reinforced. This variable was a frequency count of this specific form of conflictual interaction.

Negative Reciprocity.

Negative reciprocity was also assessed from instances of mother-child conflict. Coders observed how mothers responded to instances when the child displayed an aversive behavior. If the mother responded with an aversive behavior, this was considered an instance of negative reciprocity. These data were used to derive a Haberman binomial z-score (Gottman & Roy, 1990; see DeGarmo et al., 2004), reflecting the conditional probability that the mother would reciprocate the child’s aversive behavior.

Positive Parenting.

This outcome variable was derived from five indicators of positive parenting constituting the core components of the GenPMTO model: skill encouragement, limit setting, problem solving, monitoring, and positive involvement. Each positive parenting indicator was rescaled to range from 0 to 4 and then averaged to create the construct score.

Skill Encouragement.

The skill encouragement score was based on global ratings of mother-child interactions during the 10-minute teaching task. It was calculated as the mean of 11 items rating mothers’ use of contingent positive reinforcement and scaffolding strategies with her child. Cronbach’s alphas (n = 153) were .76, .80, .83, .65, and .72 from T0 to T4, respectively.

Limit Setting.

Limit setting was assessed via observations of maternal discipline practices during the entire 45-minute set of structured interaction tasks. The score was calculated as the mean of 13 items, four indicating positive discipline practices and nine indicating inept discipline practices (reverse-scored), with higher scores indicating greater levels of positive limit setting. Cronbach’s alphas (n = 153) were .92, .92, .92, .93, and .92 from T0 to T4, respectively.

Problem Solving.

This score was determined by coder ratings of three family problem solving discussions on topics chosen by the mother. For each task, nine items were rated. Items were then averaged, across each of the three tasks, to derive the final problem-solving score. Cronbach’s alphas (n = 153) were .93, .93, .93, .95, and .94 from T0 to T4, respectively.

Monitoring.

Monitoring was measured via three items from a structured caregiver interview (i.e., the Parent Interviewer Impressions form [PIIMP]; Capaldi & Patterson, 1989) and two items from laboratory observations. After reverse scoring relevant items, we calculated an average score, with higher scores indicating greater levels of monitoring. Cronbach’s alphas (n = 153) were .80, .75, .69, .72, and .60 from T0 to T4, respectively.

Positive Involvement.

This variable was measured by 38 items characterizing the mother-child relationship. These items combined observations of five structured interaction tasks with two global ratings. These seven indicators were then averaged to arrive at a positive involvement score. Cronbach’s alphas (n = 153) were .96, .97, .97, .96, and .97 from T0 to T4, respectively.

Intervention Ingredients

GenPMTO ingredients were measured using the Component Level Implementation Fidelity Rating System (CLIFRS), an innovative assessment tool allowing researchers to quantify the extent to which distinct intervention ingredients are delivered with fidelity (Holtrop et al., 2021). The CLIFRS is a 76-item measure intended to assess eight GenPMTO ingredients: clear directions, skill encouragement, emotion regulation, limit setting, effective communication, problem solving, monitoring, and positive involvement. Rigorous developmental work has demonstrated support for the CLIFRS in terms of item performance, reliability, and validity across seven component scales (Holtrop et al., 2021).

Using video data from a total of 170 GenPMTO intervention sessions, coders rated the extensiveness with which each item took place using a 7-point Likert scale (0 = not at all to 6 = extensively). For example, items assessed the extent to which participants were exposed to verbal teaching about the characteristics of clear directions, demonstrations of the time out sequence, and opportunities to practice a problem solving procedure. Interrater reliability was calculated on a subsample of 20% (n = 35) of sessions. Two raters provided data for each reliability check and rating pairs varied session-by-session. Intraclass correlation coefficients (ICC) were calculated using IBM SPSS Statistics version 24 based on a one-way random effects model, using a stringent standard of single-rating, absolute agreement (Shrout & Fleiss, 1979). The average ICC across all items was .77 (SD = .22), indicating good reliability (Koo & Li, 2016).

To assess exposure to intervention ingredients, rating scores were aggregated at the participant level. We summed the items for each ingredient across all intervention sessions the participant attended. Eight scores were calculated for each participant, representing their total exposure to each GenPMTO ingredient. Higher scores indicated greater exposure. More detailed information about each CLIFRS component is provided in Table 1.

Table 1.

Descriptive Statistics for CLIFRS Component Scales

Component	# items	M	SD	Range
Clear Directions. Developmentally appropriate expectations for child cooperation; use of calm and specific statements; role play and debriefing Example item: Facilitating in-session practice of clear directions	8	48.9	28.5	0–88
Skill Encouragement. Contingent positive reinforcement; dividing complex behaviors into small steps; incentive charts and token systems Example item: Demonstrating how to explain incentive chart to child	9	67.2	43.4	0–125
Emotion Regulation. Identifying and regulating emotions; attention to body language, voice tone; separating own emotions from those of child Example item: Assigning home practice of emotion regulation strategies	10	53.7	40.0	0–125
Limit Setting. Calm application of clear rules and consistent, non-punitive consequences; use of time out and privilege removal Example item: Teaching instructions for using time out strategy	10	51.4	41.5	0–115
Effective Communication. Active listening and speaking skills; respond vs. react; supports interpersonal interaction and information exchange Example item: Facilitating in-session practice of communication activity	8	28.1	25.6	0–81
Problem Solving. Structured procedure for solving problems and achieving goals; input from all family members; use of family meetings Example item: Assigning home practice to hold a family meeting	9	34.8	33.3	0–83
Monitoring. Supervising activities and peers; screening childcare providers; use of skill-building activities to decrease unsupervised time Example item: Demonstrating how to question a potential childcare provider	10	19.4	19.1	0–56
Positive Involvement. Showing affection; spending enjoyable time together; supporting school activities; integrated throughout all topics Example item: Teaching the importance of spending time with child	3	5.1	4.6	0–15

Note. N = 153. A reduced number of items were used to measure limit setting (items 1–6, 8, 9, 10) and positive involvement (items 8, 9, 10) based on the results of psychometric testing during development of CLIFRS (Holtrop et al., 2021). For interested readers, additional information is available regarding the CLIFRS measure (Holtrop et al., 2021) and the GenPMTO intervention ingredients (e.g., Forgatch et al., 2007; Forgatch & Domenech Rodríguez, 2016; Forgatch & Patterson, 2010).

Analytic Strategy

Our study included intervention data from multiple time points clustered within mothers in parenting groups. Given the importance of modeling nested data using multilievel modeling (MLM) procedures to obtain accurate estimates (Raudenbush & Bryk, 2002), study analyses were conducted using two-level MLM growth curve analyses in HLM 7 (Raudenbush et al., 2011). Level 1 estimated the within-subject trajectories of change using the initial level (intercept) and rate of change (slope) in parenting practices for each participant. Instead of assuming a linear rate of change, we tested if quadratic and cubic functions improved model fit, as this may better capture the dynamic and complex process of change over time. Level 2 allowed for testing between-subject differences for time-invariant predictors. In other words, predictors were added at Level 2 to predict the intercept and rates of change in Level 1.

In this study, the equation for coercive parenting (CoercPar) trajectories at Level 1 was

$${\mathit{\text{ CoercPar }}}_{ti}={\pi }_{0i}+{\pi }_{1i}*\left({\mathit{\text{ TIME }}}_{ti}\right)+{\pi }_{2i}*\left({{\mathit{\text{ TIME }}}^2}_{ti}\right)+{\pi }_{3i}*\left({{\mathit{\text{ TIME }}}^3}_{ti}\right)+e_{ti}$$

where CoercPar_ti is the outcome variable for individual i at Time t; π_0i is the intercept of individual i (the initial level of coercive parenting at T1); π_1i is the linear rate of change in coercive parenting from T1 through T4 for individual i at time t; π_2i is the quadratic rate of change; π_3i is the cubic rate of change; and e_ti is the residual variance in repeated measures for individual i at time t. We assume that the errors e_ti are independent and normally distributed with a common variance (Raudenbush & Bryk, 2002). Time was scaled in years. We assessed the average intercept and slopes (linear, quadratic, and cubic rates of change), the variation in these parameters, and then determined if these values differed significantly from zero.

We conducted our analyses using the sample of 153 mothers who were assigned to the GenPMTO condition. To examine the GenPMTO ingredients associated with trajectories of coercive parenting over time (RQ1), we entered the eight GenPMTO ingredients (i.e., clear directions [CD], skill encouragement [SE], emotion regulation [ER], limit setting [LS], effective communication [EC], problem solving [PS], monitoring [MO], and positive involvement [PI]) as predictors in the model. These predictors were standardized and entered simultaneously at Level 2. Entering these eight ingredients together into the model allowed us to test the unique effects of each ingredient while accounting for the others, since they are intended to operate in an integrated manner in the intervention. We also included coercive parenting at baseline (CoercParT0) in the model to control for initial levels of this parenting practice. The Level 2 equations were as follows:

$${\pi }_{\mathit{0}i}={\beta }_{\mathit{00}}+{\beta }_{\mathit{01}}*\left(C{D}_i\right)+{\beta }_{\mathit{02}}*\left(S{E}_i\right)+{\beta }_{\mathit{03}}*\left(E{R}_i\right)+{\beta }_{\mathit{04}}*\left(L{S}_i\right)+{\beta }_{\mathit{05}}*\left(E{C}_i\right)+{\beta }_{\mathit{06}}*\left(P{S}_i\right)+{\beta }_{\mathit{07}}*\left(M{O}_i\right)+{\beta }_{\mathit{08}}*\left({\mathit{\text{PI}}}_i\right)+{\beta }_{\mathit{09}}*(\mathit{\text{CoercParT}}{\mathit{0}}_i)+r_{\mathit{0}i}$$

$${\pi }_{\mathit{1}i}={\beta }_{\mathit{10}}+{\beta }_{\mathit{11}}*\left(C{D}_i\right)+{\beta }_{\mathit{12}}*\left(S{E}_i\right)+{\beta }_{\mathit{13}}*\left(E{R}_i\right)+{\beta }_{\mathit{14}}*\left(L{S}_i\right)+{\beta }_{\mathit{15}}*\left(E{C}_i\right)+{\beta }_{\mathit{16}}*\left(P{S}_i\right)+{\beta }_{\mathit{17}}*\left(M{O}_i\right)+{\beta }_{\mathit{18}}*\left({\mathit{\text{PI}}}_i\right)+{\beta }_{\mathit{19}}*\left(\mathit{\text{CoercParT}}{\mathit{0}}_i\right)+r_{\mathit{1}i}$$

$${\pi }_{\mathit{2}i}={\beta }_{\mathit{20}}+{\beta }_{\mathit{21}}*\left(C{D}_i\right)+{\beta }_{\mathit{22}}*\left(S{E}_i\right)+{\beta }_{\mathit{23}}*\left(E{R}_i\right)+{\beta }_{\mathit{24}}*\left(L{S}_i\right)+{\beta }_{\mathit{25}}*\left(E{C}_i\right)+{\beta }_{\mathit{26}}*\left(P{S}_i\right)+{\beta }_{\mathit{27}}*\left(M{O}_i\right)+{\beta }_{\mathit{28}}*\left({\mathit{\text{PI}}}_i\right)+{\beta }_{\mathit{29}}*(\mathit{\text{CoercParT}}{\mathit{0}}_i)+r_{\mathit{2}i}$$

$${\pi }_{\mathit{3}i}={\beta }_{\mathit{30}}+{\beta }_{\mathit{31}}*\left(C{D}_i\right)+{\beta }_{\mathit{32}}*\left(S{E}_i\right)+{\beta }_{\mathit{33}}*\left(E{R}_i\right)+{\beta }_{\mathit{34}}*\left(L{S}_i\right)+{\beta }_{\mathit{35}}*\left(E{C}_i\right)+{\beta }_{\mathit{36}}*\left(P{S}_i\right)+{\beta }_{\mathit{37}}*\left(M{O}_i\right)+{\beta }_{\mathit{38}}*\left(P{I}_i\right)+{\beta }_{\mathit{39}}*(\mathit{\text{CoercParT}}{\mathit{0}}_i)+r_{\mathit{3}i}$$

These models included the T1, T2, T3, and T4 time points, spanning the two-year period after the intervention ended (i.e., the intervention ingredients were not expected to predict parenting behavior prior to treatment [T0]). Restricted likelihood estimation was used in all MLM analyses. We also computed effect sizes for these predictors, using effect size r to indicate the relative influence of each predictor compared to the others (r = sqrt [t²/(t² + df)]; see Lawrence et al., 2008). We ran these analyses first with coercive parenting, and then followed the same pattern to test the positive parenting outcome.

Results

Preliminary Analyses

We began by examining the extent to which participants in the GenPMTO condition had been exposed to each of the intervention components to investigate if there was variation among our predictor variables. Table 1 displays the mean, standard deviation, and range for each ingredient. On average, participants received the greatest exposure to the topics presented in the first half of the intervention curriculum: skill encouragement (M = 67.2), emotion regulation (M = 53.7), limit setting (M = 51.4), and clear directions (M = 48.9). Each ingredient had a comparatively large standard deviation and range, relative to its mean. These descriptive data confirmed there was sufficient variation in the predictors to proceed with the modeling.

Next, we examined if the GenPMTO group demonstrated more optimal trajectories of coercive and positive parenting than the control group. Similar analyses have been described in prior studies with this sample (e.g., Forgatch & DeGarmo, 2002; Martinez & Forgatch, 2001). Since we operationalized these constructs in a slightly different manner, we repeated the analyses here to confirm that similar intervention effects were observed in our focal parenting outcomes (i.e., coercive parenting, positive parenting). We first tested unconditional growth models using MLM procedures to examine the average trajectory of change in coercive and positive parenting from baseline (T0) to 24-months post-treatment (T4). We then entered condition assignment (i.e., GenPMTO or control group) as a dichotomous predictor.

Although GenPMTO participants and control group participants did not differ in coercive parenting scores immediately following the intervention (T1), GenPMTO participants showed significantly better trajectories of coercive parenting behavior over time, as expected. That is, control group participants, on average, experienced a significant increase in coercive parenting from T0 through T4 (b = .46, p = .029, r = .19), whereas GenPMTO participants decreased in coercive parenting at a significant rate compared to control participants (b = −.79, p = .003, r = .25). On positive parenting outcomes, GenPMTO participants obtained marginally higher scores immediately following the intervention (b = .10, p = .096, r = .14) relative to the control group. For rate of change, control group participants declined .17 units per year in positive parenting, which was a significantly steeper decline than no change at all (p < .001). In comparison, GenPMTO participants experienced marginally less steep linear declines in positive parenting (b = .11, p = .089, r = .15). Overall, these analyses confirmed more favorable trajectories of coercive and positive parenting for the GenPMTO participants. We could now proceed with our primary analyses examining which specific GenPMTO ingredients were associated with changes in coercive and positive parenting following intervention exposure.

RQ1: GenPMTO Ingredients Predicting Trajectories of Coercive Parenting

We first investigated intervention ingredients associated with reductions in coercive parenting. For these analyses we were concerned with examining patterns of change following intervention exposure, from T1 (immediately post-intervention) to T4 (24-months post-intervention). The fully unconditional model (i.e., model with no predictors) indicated an overall average score of coercive parenting from T1 to T4 was 2.72, and there was significant variation in these scores (p < .001). The intraclass correlation (ICC) was .20, representing the proportion of variance in coercive parenting occurring between-subjects (i.e., 20% of the variation was between mothers, whereas 80% was due to changes within mothers across time). We then proceeded to model these changes in coercive parenting trajectories with a linear slope (deviance = 1943.94, df = 4), then with a quadratic slope added (deviance = 1926.70, df = 7), and then with a cubic slope added (1908.13, df = 11). Deviance tests indicated that adding a quadratic slope significantly improved the model fit beyond just a linear slope (Δχ² (3) = 17.24, p < .001), and adding the cubic slope significantly improved model fit even further (Δχ² (4) = 18.57, p < .001). Thus, the most complex model was the best representation of how mothers changed in coercive parenting for two years post-intervention. Within this unconditional growth model, there was significant variation to be explored in the intercept, linear slope, quadratic slope, and cubic slope.

We then created the conditional model by entering the eight GenPMTO ingredients as predictors of the intercept, linear slope, quadratic slope, and cubic slope in the two-year post-intervention assessments of coercion. We also entered coercive parenting at baseline into the model as a control variable. In this model, the average level of coercive parenting immediately following the intervention (i.e., intercept) was 3.10, and the average GenPMTO mother declined significantly and linearly in coercive parenting across the next two years at a rate of −3.68 units per year (p < .001, r = .37). The shape of this change in average trajectory was augmented by significant (p < .001) quadratic (4.69, r = .32) and cubic (−1.45, r = .28) functions in that it declined from T1 to T2 and then increased after T3 and declined somewhat towards T4. Figure 1a depicts the average trajectory of change in coercive parenting.

Figure 1.

Trajectories of Change in Coercive Parenting across 2 Years Post-Intervention (T1 to T4) for Parents in the GenPMTO condition (N = 153)

Note. The solid line corresponds with the average trajectory of change in observed coercive parenting across two years post-intervention among the sample of 153 mothers in the GenPMTO condition. The solid line is identical in each figure. The dashed line indicates the effect of a one standard deviation unit increase in exposure to that intervention ingredient on the predicted rate of change in coercive parenting over time. These trajectories include betas from the prediction of the intercept, linear, quadratic, and cubic slopes, while controlling for the other 7 components and baseline level of coercive parenting. Immediately following intervention (T1) was coded as 0 on the x-axis, 6 months post-intervention (T2) was coded as 0.5 on the x-axis, 1 year post-intervention (T3) was coded as 1 on x-axis, and 2 years post-intervention (T4) was coded as 2 on the x-axis. The y-axis indicates observed coercive parenting scores, thus higher scores indicate worse parenting practice outcomes.

Next, we moved on to examine if any intervention ingredients predicted the model intercept – i.e., the level of coercive parenting immediately post-intervention (T1). Two ingredients demonstrated significant results. Greater exposure to effective communication predicted lower levels of coercive parenting at T1 (b = −0.83, p = .040, r = .18). Conversely, participants exposed to more of the monitoring ingredient during GenPMTO demonstrated marginally higher coercive parenting at T1 (b = 0.65, p = .061, r = .16).

We then proceeded to examine change over time by investigating if any intervention ingredients predicted the model slopes. Four intervention ingredients were significant in predicting trajectories of coercive parenting across the 2-year period following intervention (see Table 2): emotion regulation, effective communication, problem solving, and monitoring. To aid in the interpretation of study findings, the trajectories of change (from T1 to T4) are depicted in Figure 1. The solid line corresponds with the average trajectory of change in coercive parenting, with higher scores indicating worse (i.e., more coercive) outcomes; it is the same in each figure. The dashed line indicates the effect of a one standard deviation unit increase in exposure to that intervention ingredient on the predicted rate of change in coercive parenting.

Table 2.

MLM Results Predicting Coercive Parenting Trajectories from T1 to T4 Using Restricted Maximum Likelihood Estimation and Robust Standard Errors with Standardized Predictors (N = 153)

	Initial Level T1			Linear Slope			Quadratic Slope			Cubic Slope
Fixed Effect	b	SE	r	b	SE	r	b	SE	r	b	SE	r
Intercept	3.10**	0.16	.86	−3.68**	0.81	.37	4.69**	1.21	.32	−1.45**	0.43	.28
Clear directions	−0.39	0.42	.08	1.33	2.28	.05	−0.31	3.43	.01	−0.09	1.21	.01
Skill encouragement	−0.27	0.60	.04	1.36	3.18	.04	−1.76	4.50	.03	0.41	1.54	.02
Emotion regulation	0.61	0.39	.14	−5.34**	1.76	.26	6.13*	2.78	.19	−1.77†	1.00	.15
Limit setting	0.35	0.36	.08	−4.04†	2.07	.17	5.95†	3.31	.15	−1.96	1.19	.14
Effective communication	−0.83*	0.40	.18	5.26*	2.34	.19	−7.58*	3.41	.19	2.53*	1.17	.18
Problem solving	−0.04	0.30	.01	4.51**	1.57	.24	−6.73**	2.45	.23	2.29*	0.88	.22
Monitoring	0.65†	0.34	.16	−6.20**	1.99	.26	9.54**	2.85	.28	−3.25**	0.98	.28
Positive involvement	−0.05	0.26	.02	2.05	1.37	.13	−2.92	2.09	.12	0.98	0.74	.12
Coercive parenting (T0)	0.11	0.08	.11	1.06*	0.47	.19	−1.66*	0.75	.19	0.60*	0.27	.19

Random Effect	Variance Component	df	χ 2
Intercept (T1)	2.04**	84	226.76
Linear slope	18.16*	84	111.95
Quadratic slope	35.79†	84	102.40
Cubic slope	4.59†	84	101.81
Level-1 error	1.51

Reliability Estimate at Level 1
Intercept (T1)	.57
Linear slope	.21
Quadratic slope	.19
Cubic slope	.20

Note. b = Unstandardized Beta; SE = Standard Error; r = Effect Size. Effect size r = sqrt [t²/ t² + df)]. T1 = Immediately following intervention; T2 = 6 months post-intervention; T3 = 1 year post-intervention; T4 = 2 years post-intervention.

^†p < .10.

^*p < .05.

^**p < .01 (two-tailed).

Exposure to emotion regulation was associated with significant linear (b = −5.34, p = .003, r = .26), quadratic (b = 6.13, p = .029, r = .19), and [marginally] cubic (b = −1.77, p = .079, r = .15) rates of change. Inspection of Figure 1d reveals that a one standard deviation unit increase in exposure to emotion regulation predicted a greater decrease in coercive parenting early on, followed later by an increase in coercive parenting, before settling near the average level of coercive parenting two years post-intervention. Monitoring was also associated with significant linear (b = −6.20, p = .002, r = .26), quadratic (b = 9.54, p = .001, r = .28), and cubic (b = −3.25, p = .001, r = .28) rates of change, revealing a pattern similar to emotion regulation but with less favorable fluctuations in coercive parenting. Not only did participants exposed to greater monitoring demonstrate a trend toward higher levels of coercive parenting immediately following the intervention (see Figure 1f, intercept), but a one standard deviation unit increase in exposure to monitoring predicted a small, initial decrease in coercive parenting but then much higher later levels of coercive parenting. In contrast, participants exposed to more effective communication demonstrated significantly lower levels of coercive parenting immediately following the intervention (see Figure 1b, intercept). Given the significant linear (b = 5.26, p = .026, r = .19), quadratic (b = −7.58, p = .028, r = .19) and cubic (b = 2.53, p = .033, r = .18) rates of change, inspection of Figure 1b indicates that a one standard deviation unit increase in effective communication predicted early levels of coercive parenting that were close to average and then lower levels of coercive parenting that persisted until two-years post-intervention. The fourth and final component to demonstrate significant linear (b = 4.51, p = .005, r = .24), quadratic (b = −6.73, p = .007, r = .23), and cubic (b = 2.29, p = .010, r = .22) rates of change was problem solving. A one standard deviation unit increase in exposure to problem solving (Figure 1c) predicted elevated levels of coercive parenting early, a slight dip later on, and then a gradual rise toward the two-year mark. Among the remaining four components, limit setting showed a marginally significant trend for linear (b = −4.04, p = .053, r = .17) and quadratic (b = 5.95, p = .075, r = .15) rates of change, while exposure to clear directions, skill encouragement, and positive involvement showed no significant association to changes in coercive parenting over time, after controlling for the other components and baseline levels of coercive parenting. These analyses all controlled for baseline levels of coercive parenting; coercive parenting at T0 was not linked with coercive parenting immediately post-intervention (T1), but was linked with the linear (b = 1.06, p = .025, r = .19), quadratic (b = −1.66, p = .028, r = .19), and cubic (b = 0.60, p = .028, r = .19) rates of change in coercive parenting trajectories.

RQ2: GenPMTO Ingredients Predicting Trajectories of Positive Parenting

Our second research question investigated which GenPMTO ingredients predicted trajectories of the positive parenting outcome measure following intervention exposure, using an analytic approach identical to RQ1 and testing the intercept, linear, quadratic, and cubic rates of change in positive parenting. Although the quadratic and cubic functions did not significantly improve model fit beyond the linear rate of change, there were significant variance components in the quadratic and cubic [marginally] rates of change which indicated adequate variation to be explained by predictors. Further, the inclusion of the quadratic and cubic rates of change increased the reliability estimate for our intercept and rates of change. This was sufficient justification to include the quadratic and cubic rates of change in the positive parenting model.

The results are presented in Table 3. The ICC for positive parenting was .57, indicating 57% of the variance was between mothers while 43% was due to changes in mothers across time. Given that less variance was estimated to result from change across time for positive parenting (43%) compared to coercive parenting (80%), we anticipated more difficulty in predicting positive parenting. In this final unconditional growth model, the average post-intervention level (T1) of positive parenting was 2.96, and the average linear rate of change was a .14 unit increase per year (not a significant average increase from T1 to T4). The quadratic and cubic rates of change were likewise not significantly different from no change.

Table 3.

MLM Results Predicting Positive Parenting Trajectories from T1 to T4 Using Restricted Maximum Likelihood Estimation and Robust Standard Errors with Standardized Predictors (N = 153)

	Initial Level T1			Linear Slope			Quadratic Slope			Cubic Slope
Fixed Effect	b	SE	r	b	SE	r	b	SE	r	b	SE	r
Intercept	2.96**	0.04	.99	0.14	0.18	.06	−0.09	0.27	.03	0.02	0.09	.02
Clear directions	0.02	0.10	.02	−0.03	0.53	.00	−0.19	0.86	.02	0.10	0.30	.03
Skill encouragement	0.18	0.13	.12	−0.06	0.73	.01	−0.13	1.08	.01	0.08	0.36	.02
Emotion regulation	0.02	0.11	.01	−0.32	0.50	.06	0.69	0.80	.07	−0.28	0.28	.09
Limit setting	−0.11	0.10	.09	0.66	0.50	.11	−0.84	0.81	.09	0.26	0.29	.08
Effective communication	0.17	0.11	.13	−0.21	0.50	.04	0.21	0.76	.02	−0.07	0.26	.02
Problem solving	0.00	0.09	.00	−0.14	0.66	.02	0.18	1.01	.02	−0.05	0.35	.01
Monitoring	−0.15	0.09	.14	0.45	0.52	.07	−0.71	0.79	.08	0.26	0.27	.08
Positive involvement	−0.04	0.05	.07	−0.42	0.35	.10	0.78	0.50	.13	−0.28†	0.17	.14
Positive parenting (T0)	0.41**	0.08	.39	0.50	0.42	.10	−0.57	0.68	.07	0.16	0.24	.06

Random Effect	Variance Component	df	χ 2
Intercept (T1)	0.12**	86	231.47
Linear slope	1.23*	86	116.47
Quadratic slope	2.35*	86	111.08
Cubic slope	0.26†	86	108.50
Level-1 error	0.07

Reliability Estimate at Level 1
Intercept (T1)	.64
Linear slope	.28
Quadratic slope	.25
Cubic slope	.24

Note. b = Unstandardized Beta; SE = Standard Error; r = Effect Size. Effect size r = sqrt [t²/ t² + df)]. T1 = Immediately following intervention; T2 = 6 months post-intervention; T3 = 1 year post-intervention; T4 = 2 years post-intervention.

^†p < .10.

^*p < .05.

^**p < .01 (two-tailed).

Once the eight GenPMTO ingredients were entered into the conditional model, while controlling for positive parenting at baseline, exposure to greater positive involvement showed a trend toward predicting the cubic rate of change (b = −0.28, p = .096, r = .14) in positive parenting. However, none of the eight ingredients significantly predicted the linear, quadratic, or cubic rates of change across time from T1 to T4. Additionally, positive parenting at baseline was significantly associated with higher positive parenting immediately post-intervention (b = 0.41, p < .001, r = .39), but was not linked with the rates of change in positive parenting.

Discussion

Despite strong evidence of the effectiveness of parenting interventions, much less is known regarding the specific intervention ingredients linked to changes in parenting. To move science forward, this study investigated the active ingredients of the GenPMTO intervention by examining the extent to which exposure to eight distinct ingredients was associated with trajectories of change in coercive and positive parenting in the two years post-intervention. Growth curve analyses were conducted within a MLM framework. Each ingredient was tested while controlling for the other seven ingredients, and coercive parenting at baseline, to discern its unique contribution to the expected trajectory of change in parenting practices. The results supported our hypothesis that a subset of intervention ingredients would demonstrate significant, independent associations with trajectories of change in coercive parenting (RQ1).

A robust set of significant findings suggested four components as active ingredients in reducing coercive parenting: emotion regulation, effective communication, problem solving, and monitoring. That is, differential exposure to each of these four ingredients was significantly associated with the post-intervention level and/or trajectories of change in coercive parenting across two years post-intervention. These four ingredients demonstrated a pattern of modest, but meaningful effect sizes (M_r = .20) across the coercive parenting outcomes.

Within the context of experimental research with GenPMTO, problem solving and monitoring are acknowledged as core components of the model for their role as mediators of more distal change outcomes; emotion regulation and effective communication are considered only as supporting dimensions (e.g., Forgatch & Gewirtz, 2017; Forgatch & Patterson, 2010). Yet, prior research has signaled their relevance. For example, in a non-intervention longitudinal study, Forgatch and Stoolmiller (1994) found the exchange of contempt, a negative emotion, was associated with disrupted monitoring. In another non-experimental study, negative emotional expression during family problem solving was associated with less effective problem-solving outcomes (Forgatch, 1989). The present study builds on these trials to support the inclusion of emotion regulation and effective communication skills as core components of GenPMTO.

Our search for ingredients tied to changes in positive parenting was less fruitful. None of the ingredients tested demonstrated significant, independent associations with trajectories of change in positive parenting (RQ2). Looking at the extant literature, prior GenPMTO research has examined coercive parenting and positive parenting constructs in relation to several outcomes, including youth delinquency (Forgatch et al., 2008) and maternal standard of living (Patterson et al., 2010). In both studies, assignment to the GenPMTO condition led to favorable changes in coercive parenting and positive parenting. By arranging the parenting dimensions of coercion and positive parenting longitudinally, it was possible to find that early changes in coercive parenting (from baseline to 1-year) fully mediated the improvements to positive parenting (from baseline to 30-months). Based on these findings, it is reasonable to hypothesize that the active ingredients of GenPMTO may most immediately impact changes in coercive parenting. Coercion theory and the Social Interaction Learning Model specify that coercive parenting disrupts positive parenting, leading Patterson and colleagues (2010) to suggest that for positive parenting to grow, first there must be a decrease in coercion. We investigated this idea in post-hoc exploratory analyses using a time-varying covariate model with shifts across time in coercive parenting predicting shifts across time in positive parenting (while controlling for trajectories of positive parenting), and vice-versa (i.e., with positive parenting predicting negative parenting). Both models were significant, but the model with changes in coercive parenting predicting changes in positive parenting fit substantially better (deviance = 410, df = 11) than the alternative (deviance = 1840, df = 11), offering support for this hypothesis.

Active Ingredients and Change Over Time

Our results add to the literature investigating the active ingredients of parenting interventions. Past studies have examined moderators of treatment (e.g., Lundahl et al., 2006) as well as key program components linked to both parenting (e.g., Hill & Owens, 2013; Kaminski et al., 2008) and child (e.g., Kaminski et al., 2008; Leijten et al., 2019) outcomes. For example, in their seminal meta-analytic review, Kaminski and colleagues (2008) found that programs teaching positive parent-child interactions and emotional communication skills content were significantly associated with better parenting outcomes, after controlling for threats to internal validity; programs including problem solving and cognitive/academic skills content were associated with smaller effects. Indeed, there is precedent in the literature for some intervention components to be associated with better parenting outcomes while some seem linked with poorer effects (Hill & Owens, 2013; Kaminski et al., 2008). However, most prior investigations have not examined change over time in the advanced manner that our data allowed.

Future intervention research should continue to move beyond only comparing change at pre- and post-test and relying on models that presuppose simple linear rates of change toward a more nuanced examination of change trajectories, which can yield valuable information about how the change process unfolds. Prior research with this sample found that GenPMTO participants initially worsened in negative reinforcement during the intervention (from BL to post-test) compared to the control group, but then went on to demonstrate significantly better negative reinforcement outcomes six months later (Forgatch & DeGarmo, 1999). This negative quadratic shape, where things get worse before they get better, has also been observed in earlier GenPMTO studies examining parental resistance (e.g., Patterson & Chamberlain, 1994; Stoolmiller et al., 1993). These studies found that on average, parental resistance peaked near the midpoint of therapy and then decreased by termination. Moreover, the absence of this “struggle and work through” process during therapy predicted increased child arrests over the two years post-intervention (Stoolmiller et al., 1993). This clinically meaningful pattern of change would have been missed if just a linear rate of change was modeled with pre- and post-test assessments. Based on this prior work, we modeled change trajectories across four timepoints post-intervention, expecting that the intricate change processes observed earlier in the intervention (e.g., Forgatch & DeGarmo, 1999) would continue across the subsequent time periods.

The longitudinal growth curves in our study revealed complex and varied trajectories of change that included times of clear improvement as well as periods of notable deterioration in coercive parenting (see Figure 1). For instance, note the similar shapes for emotion regulation, limit setting, and monitoring—they improve 6-months post-intervention, then worsen, and finally rebound two years after the intervention. The symmetry in these shapes is interesting given these ingredients were designed to work in tandem in the GenPMTO intervention. The change trajectories suggest that, while holding the other ingredients steady, a one standard deviation unit increase in emotion regulation, limit setting, or monitoring would predict a rather similar trajectory of coercive parenting over time. The change trajectories for effective communication and problem solving, another set of ingredients hypothesized to work together, were also comparable to one another.

Somewhat ironically then, our research touting the need for more concerted efforts to understand how interventions achieve positive change succeeded in highlighting the need for more concerted efforts to understand how change actually unfolds. Further research is needed to examine the long-term effects of these change trajectories on child behavior and other family outcomes. For example, do early and sustained—although more modest—improvements in coercive parenting (e.g., effective communication) lead to better outcomes than more erratic change trajectories (e.g., monitoring)? Or, does more immediate progress during the six months following treatment ignite collateral improvements in family functioning that inoculate the family system against later increases in coercive parenting (e.g., emotion regulation)? Perhaps different ingredients exert different effects over time in the context of child development. That is, as children get older, different parenting strategies may prove more or less helpful in staving off coercion (e.g., monitoring in pre-adolescence). Also, how might other transitions (e.g., parental partnering or unpartnering) affect these trajectories? Determining how interventions achieve change is recognized to be a challenging venture (e.g., Blase & Fixsen, 2013; Gottfredson et al., 2015). With continued research using observational intervention data and advanced analytic approaches, researchers must continue to unravel these complex change processes.

Limitations and Suggestions for Future Research

Interpretation of our findings should be tempered by several factors. First, significant associations between intervention ingredients and change trajectories do not prove causation. Our results can be valuable for informing subsequent research where specific intervention ingredients can be selected and tested in randomized experimental designs (e.g., MOST trials; Collins et al., 2005). Second, our study relied on a relatively small sample size for the analyses employed and should be replicated in larger samples. Furthermore, our research focused on an early iteration of GenPMTO targeted toward a prevention sample of recently separated mothers with school-aged sons. Time and historical factors may have influenced our findings. The group-based delivery of GenPMTO also did not account for different levels of participant engagement during sessions. Future studies should investigate the active ingredients of GenPMTO in more contemporary applications of the intervention across different populations (e.g., two-parent families, mother-daughter dyads, youth of different ages) in clinical and prevention samples and in individual family formats. Moreover, while our study sample was primarily low-income mothers who self-identified as Caucasian, research has suggested that family-based intervention components may be associated with differential effects among minoritized vs. non-minority individuals (Hill & Owens, 2013). The role of coparent involvement over time may also impact change trajectories and would be an important variable to investigate in future research. In addition, it is possible that different ingredients may predict parenting practice trajectories beyond two years post-intervention. For instance, if changes in coercive parenting precede changes in positive parenting, perhaps a longer observation period is needed before the beneficial effects of exposure to skill encouragement, a topic meant to support positive parenting, are manifested. Lastly, the construct of intervention ingredient “exposure” used in this study aggregates different dimensions of GenPMTO delivery. Subsequent research could employ the CLIFRS measure (Holtrop et al., 2021) to examine more nuanced hypotheses regarding intervention ingredient dosage as well as dimensions of intervention delivery.

Looking forward, the advancements in this study pave the way for continued efforts to expand understanding of intervention ingredients. As active ingredients are identified, opportunities emerge to examine the dose-response relationship between ingredients and participant outcomes. Once we determine how much exposure to each ingredient is sufficient to achieve improvements, this knowledge can be used to deliver interventions more efficiently. Future research can also investigate differential effects of exposure to various dimensions of intervention delivery. For example, are participants more impacted by leader demonstrations of parenting behaviors versus opportunities to practice behaviors in group? Does optimal exposure involve a combination of various dimensions? These future directions would provide valuable information for refining the delivery of parenting interventions in real-world contexts.

Conclusion

This study identified four GenPMTO ingredients as the intervention elements linked to change in coercive parenting across two years post-intervention—emotion regulation, effective communication, problem solving, and monitoring—while controlling for each of the other ingredients and coercive parenting at baseline. The change processes were complex and nonlinear, therefore, these results do not immediately suggest that parents will experience better outcomes if interventionists ramp up delivery of these topics. Likewise, it is important to clarify that many GenPMTO components build on other ingredients, so it would be ill-advised to deliver only these four ingredients in an effort to enhance parenting outcomes, simply based on these study results. However, what this study does suggest is that the threads linking exposure to GenPMTO ingredients and coercive parenting outcomes are most strongly tied to these four ingredients. In response, exploring ways in which exposure to these four ingredients could be calibrated to optimize intervention benefits would be a promising next step. Overall, this study underscores the possibility of strengthening future iterations of GenPMTO by expanding the core components specified in the model, which may further improve public health benefits.