Advancing Early Childhood Mental Health Consultation: Evaluating Traditional and AI-Enhanced Approaches to Support Children and Teachers

Abstract

Early Childhood Mental Health Consultation (ECMHC) promotes children’s social–emotional development and reduces challenging behaviors in early care and education (ECE) centers, yet implementation barriers increase teacher stress and reduce confidence. Scalable, efficient, and accessible approaches are needed to meet ECE center demands. This quasi-experimental match-controlled study evaluated two ECMHC programs in promoting children’s social–emotional development and improving teachers’ skills/attitudes compared to an attention control condition in 22 ECE centers in lower-resourced areas of BLINDED. We compared Jump Start (JS; traditional human consultation model), Jump Start Go (JS Go; AI-enhanced consultation model), and Healthy Caregivers–Healthy Children (HC2; obesity-prevention consultation model). Child social–emotional development, teacher workplace stress/confidence, and classroom practices were assessed at pre-and post-intervention. Children in JS and JS Go interventions demonstrated significant social–emotional gains (F = 13.55, p < 0.001), with the largest reductions in internalizing problems observed in children who received JS Go (−2.91 points; F = 9.65, p < 0.001). JS Go classrooms also showed greater improvements in prosocial behavior (F = 5.05, p = 0.012) and resiliency (F = 8.95, p < 0.001) than HC2 classrooms. Findings suggest that both traditional and AI-enhanced ECMHC approaches can promote teachers’ capacity to support children’s social–emotional development.

1. Introduction

Early care and education (ECE) centers are key components in the development of young children, particularly those growing up in under-resourced communities (Harding & Paulsell, 2018; Hartman et al., 2017; Saitadze & Lalayants, 2021). Children from low-income backgrounds tend to exhibit more behavioral challenges within ECE settings than their more advantaged peers (Hartman et al., 2017). These disparities are especially prevalent among boys from racially and ethnically minoritized groups (U.S. Education Department, 2021). If not addressed, these challenging behaviors place children at risk of suspension or expulsion from their ECE center (Sciamanna, 2020). Unfortunately, research shows that not addressing challenging behaviors early on can result in long-term consequences including overall social and emotional development impairments (Stegelin, 2018). Recent evidence from a large-scale evaluation of Early Childhood Mental Health Consultation (ECMHC) in New York City ECE centers found that one year of consultation led to significant improvements in both classroom practices and children’s social–emotional outcomes (Kadik et al., 2025).

The current study aimed to evaluate the effectiveness of two ECMHC programs, Jump Start (JS) and Jump Start on the Go (JS Go), in promoting children’s social–emotional development and enhancing teacher confidence and beliefs towards classroom practices compared to an attention control program, Healthy Caregivers–Healthy Children (HC2). By comparing a traditional, relationship-based ECMHC model with an innovative Artificial Intelligence (AI)-enhanced hybrid approach, this study sought to identify scalable and sustainable strategies to strengthen teacher capacity and foster emotionally supportive learning environments in under-resourced ECE centers. To contextualize these programs, it is important to understand the foundations of ECMHC, the model from which JS and its adaptations were derived.

1.1. Importance of Social–Emotional Development

During the preschool years, young children have the capacity to develop foundational skills to understand and regulate their emotions as well as manage their attention and behaviors (Brown et al., 2012). The development of these social–emotional competencies supports positive adjustment, promotes educational attainment, and reduces the risk of behavioral challenges through adolescence (Bodovski & Youn, 2011; Schreier & Chen, 2013). Researchers have identified a socioeconomic gap in such development (Schreier & Chen, 2013); exposure to poverty-related adversities, such as chronic stressors and unpredictable conditions, has been found to impede social–emotional development in early childhood by overwhelming children’s developing stress response systems (Blair & Raver, 2015). Given that young children spend up to 70% of their waking hours in ECE centers (Harding & Paulsell, 2018; Hartman et al., 2017), they present an unparalleled opportunity to boost social–emotional development within under-resourced communities (Bierman et al., 2025).

1.2. Role of Early Care and Education Centers

In addition to cognitive, language, and academic benefits, attendance at a high-quality ECE center is associated with significant improvements in social–emotional development, with effects even more prominent for children within lower-resourced communities (Donoghue & Council on Early Childhood, 2017; Draper et al., 2024). However, researchers have identified a relationship between neighborhood poverty and lower levels of preschool quality as well as an indirect relationship between neighborhood poverty and social–emotional development via preschool quality (McCoy et al., 2015). Child behavior problems are often cited as a top burden to teachers at ECE centers, especially within under-resourced communities (Hoffman & Kuvalanka, 2019). Prior research suggests that most teachers report feeling overwhelmed when trying to sustain high-quality learning environments while concurrently managing challenging classroom behaviors (Natale et al., 2023b). In turn, teacher stress can result in classroom instability and staff attrition (Stein et al., 2024).

1.3. Infant and Early Childhood Mental Health Consultation Model

ECMHC is an evidence-based service in which mental health consultants work with center directors and teachers through a reflective stance to strengthen their capacity to support children’s social–emotional development and mitigate challenging classroom behaviors (Drake-Croft et al., 2025; Natale et al., 2025c). While researchers have found that access to ECMHC improves prosocial behaviors (Conners Edge et al., 2021; Miles et al., 2021), the lack of infrastructure and resources within ECE centers has been identified as barriers in the implementation of ECMHC (Sandstrom & Dwyer, 2021).

To address these barriers, our team has led a programmatic line of research developing and refining the JS Early Childhood Consultation program, an ECMHC-informed model designed to improve teachers’ confidence and skills for creating emotionally supportive, inclusive classrooms (Futterer et al., 2024; Natale et al., 2022b, 2024, 2025b). Our prior studies have advanced the ECMHC literature by: (a) establishing the Health Environment Rating Scale-Classroom (HERS-C) as a valid measure of teacher practices (Futterer et al., 2024), (b) demonstrating that participation in JS significantly improves children’s protective factors, as measured by the Devereux Early Childhood Assessment (DECA), and enhances teachers’ use of safety and behavior management strategies (Natale et al., 2024, 2025b), and (c) identifying key implementation barriers that limit long-term sustainability. These findings established JS as a promising, evidence-informed model while highlighting the need for innovations that enhance scalability and ongoing teacher support.

1.4. Theoretical Framework

The JS program is based on the Georgetown Model of ECMHC. This model is grounded in an ecological, relationship-based theoretical framework that situates children’s social–emotional development within interconnected systems of caregivers, classrooms, programs, and communities (Center for Early Childhood Mental Health Consultation, 2013). This framework views young children’s development as embedded within interconnected systems of caregivers, classrooms, programs, and communities. Drawing on attachment theory (Bowlby, 1988), developmental psychopathology (Cicchetti & Lynch, 1993), and ecological systems theory (Bronfenbrenner, 1992), the model emphasizes that strengthening the relationships surrounding a young child, particularly the caregiver–child dyad, is the primary pathway to improving social–emotional wellbeing and reducing challenging behavior. Consultants serve as capacity-builders rather than direct service providers by engaging teachers in reflective practice, coaching, and collaborative problem-solving aligned with adult learning theory (Knowles et al., 2015) and the parallel process central to reflective supervision (Heffron & Murch, 2010). The model also incorporates principles of equity-focused mental health promotion, cultural humility, and family engagement (Hunter et al., 2016a), acknowledging the systemic stressors experienced by ECE providers. Together, this framework positions ECMHC as a preventative, multilevel intervention that enhances protective factors across children, caregivers, classrooms, and programs to reduce suspension and expulsion and promote sustainable mental health in early childhood settings. Despite the strong theoretical grounding and demonstrated promise of the ECMHC model, less is known about how ECMHC-informed programs can be implemented in ways that are feasible, scalable, and responsive to the everyday constraints faced by ECE centers.

1.5. Rationale for Current Study

ECE centers are an essential setting for promoting children’s social–emotional development and long-term success (Natale et al., 2023b; Silver et al., 2023; Stein et al., 2024). Embedding ECMHC-informed approaches within ECE centers benefits not only children but also teachers by enhancing their confidence, reflective capacity, and stress management (Sandstrom & Dwyer, 2021). In addition, mental health consultants play a critical role in helping teachers engage in reflective practices and gain access to resources that support all children (Sandstrom & Dwyer, 2021). However, teachers often encounter barriers to implementing programs such as JS, including staffing shortages, limited time, and lack of awareness of available services (Cipriano et al., 2020; Collie, 2021). These barriers contribute to teaching-related stress and lower professional confidence (Gu & Day, 2007), underscoring the need for a scalable, multi-level approach that is accessible, efficient, and responsive to the demands of ECE centers.

Building on this foundation, the present study represents the next phase in our team’s ongoing work to evaluate the impact of scalable, technology-enhanced consultation models for ECE centers. Our previous findings demonstrated that traditional ECMHC programs like JS are effective but resource-intensive, requiring substantial consultant time and on-site presence. To address this limitation, our team recently developed Jump Start on the Go (JS Go), an AI-enhanced extension of the JS program. JS Go integrated live consultation with app-based tools designed to sustain reflective practice and promote teacher resiliency (Natale et al., 2025a, 2025c). Preliminary pilot data showed that JS Go can enhance teachers’ classroom practices for improving behavioral challenges.

Despite these promising findings, no study has directly compared JS, JS Go and an active control condition to determine the contributions of traditional and AI-enhanced consultation models. Therefore, the current quasi-experimental study evaluates the effectiveness of JS and JS Go in promoting children’s social–emotional development and improving teacher practices and attitudes, compared with an obesity-prevention program (HC2). By examining these two consultation models side-by-side, this study extends our prior research on ECMHC implementation and contributes new insights into how technology-enhanced consultation can improve the sustainability, accessibility, and scalability of mental health programs within ECE settings. We aimed to answer the following questions:

Research Question 1: Will children in the JS and JS Go intervention conditions exhibit greater improvements in prosocial behaviors compared to children in the attention control condition (HC2)?

Research Question 2: Will children in the JS and JS Go intervention conditions exhibit greater reductions in behavior problems compared to children in the attention control condition (HC2)?

Research Question 3: Will teachers in the JS and JS Go intervention conditions demonstrate improved skills (i.e., classroom practices, teacher stress, and confidence in working with children with challenging behaviors, teacher coping) compared to teachers in the attention control condition (HC2)?

2. Materials and Methods

2.1. Participants

First, 434 ECE centers in South Florida were screened for study participation using the following eligibility criteria: (1) more than 30 children aged 1.5 to 5 years old; (2) serve low-income families; (3) reflect the ethnic diversity of the Miami-Dade County school district (either >=60% Hispanic or >=60% Black/African American); and (4) teachers agree to participate. Of these programs, 45 met eligibility criteria, all of which agreed to participate. Of those that agreed to participate, only 22 programs were included in the analysis after applying matching procedures (described in Procedures).

Teachers and parents from the selected ECE centers were asked to participate if they met the following criteria: (1) spoke English or Spanish and (2) taught children aged 1.5 to 5 years and had a child aged 1.5 to 5 years enrolled in the selected ECE center. Teachers who agreed to participate completed a written informed consent process, and parents provided written informed consent for the child enrolled in the study. The University of Miami’s Institutional Review Board approved this study (IRB #s 20220115 and 20231026), and it is currently registered with ClinicalTrials.gov (NCT05445518 and NCT06374550).

2.2. Procedures

A quasi-experimental match-controlled design was implemented to assess the efficacy of the traditional human consultation model, JS, compared to the AI-enhanced application, JS Go, and compared to an obesity-prevention attention control, HC2. Participants were evaluated at baseline (pre-intervention) and post-intervention. They received $20 for completing baseline assessments and $30 for completing post-intervention assessments. Teachers provided information about themselves and their students, while caregivers provided their child’s sociodemographic information. Teacher measures were completed through Research Electronic Data Capture (REDCap), a secure, web-based platform specifically designed to support research data collection and management (Harris et al., 2009). Parents also completed their measures in REDCap via email invitations and public links, and all study data were stored in REDCap. To promote baseline comparability across intervention conditions, we implemented a nearest-neighbor, one-to-one matching procedure without replacement. Children were matched based on the following criteria: (1) gender, (2) age (within one year), and (3) DECA Total Protective Factors T-score (within one standard deviation). This resulted in a total sample of 153 children, with 51 children in each intervention condition (also referred to as treatment group).

2.2.1. Intervention Condition 1: Traditional JS

Informed by the Georgetown ECMHC framework, our team developed the traditional JS intervention, which delivers a standardized implementation toolkit curriculum centered around four pillars: safety, behavioral support, communication, and resiliency coping (Hunter et al., 2016b; Natale et al., 2022b). The Toolkit contained 24 infographics corresponding to the four program pillars, along with supplemental classroom materials (e.g., handwashing timers, “No Yell” bells, See My Feelings mirrors). Each infographic guided the consultation using a standardized sequence: (1) Reflect, (2) Inform, and (3) Practice. Teachers chose to receive their weekly consultation sessions delivered by licensed mental health consultants via a virtual telepresence robot or via in-person consultation. Sessions were delivered over 14 weeks, and each session lasted approximately 60 min. Additional details about the JS intervention can be found elsewhere (Natale et al., 2023a).

2.2.2. Intervention Condition 2: JS Go

Centered around the same four pillars as traditional JS and informed by the Georgetown ECMHC framework, JS Go incorporated an AI-enhanced mobile app that utilizes a hybrid approach to provide mental health consultations within ECE centers. It was designed as a multi-level, multi-modal intervention consisting of: (1) simplified, multilingual video content in English and Spanish available within the JS Go app; (2) personalized, AI-driven guidance in the JS Go curriculum; and (3) secure, bidirectional communication between teachers and consultants. The program was developed using a community-based participatory research approach to ensure cultural and linguistic relevance across diverse ECEs (Natale et al., 2025a).

In contrast to traditional JS, JS Go teachers received weekly, 30-min individualized or paired live consultations over 14 weeks centered around the four JS pillars, supplemented by ongoing app-based resources and asynchronous AI-based support. This was a lower dosage than traditional Jump Start. A licensed and trained mental health consultant directed the consultations, which included modeling how the teacher could use the JS Go app for classroom management support. Throughout the 14 weeks, teachers had access to the intervention toolkit within the JS Go app, which included infographics with key points, reflective questions, practical tips aligned with the four JS program pillars, instructional videos demonstrating strategies to implement each pillar in their classroom, and retrieval augmented AI-driven chatbot guidance on implementing JS pillars. The mental health consultant developed and used an individualized action plan within each consultation to set weekly goals based on the targeted pillar, introduce strategies using the infographics and videos within the JS Go app, and support teachers’ practice of these strategies with real-time feedback. Each pillar was implemented over approximately three weeks before moving on to the next pillar. The sequence of pillars was individualized based on participants’ own self-assessment and subsequent discussions to agree upon a plan. Consultants were instructed to follow up with teachers during each live consultation to troubleshoot any access issues with the JS Go app. However, actual rates of access barriers were not formally assessed within the study. Additional details about the app interface can be found elsewhere (Natale et al., 2025a).

JS Go Chatbot Model: Data Flow, Storage, and Safeguards for Teacher Input

The JS Go chatbot operates as a retrieval-augmented generation model trained on the Jump Start Go curriculum and constrained via boundary-controlled prompt governance to respond only to content indexed within the authorized knowledge repository. Teachers interact with the AI chatbot through text-based input submitted either as free-text queries or by selecting predefined prompts within the interface. Queries falling outside the model’s retrieval domain or involving clinical risk content are programmatically deflected to maintain scope fidelity and prevent unsupervised clinical guidance outside intended use. Specifically, the chatbot will refer teachers back to their mental health consultant for support involving clinical risk and/or guidance related to topics beyond the scope of the knowledge repository.

No personally identifying student information is requested or required for system use, and the platform is structured to maintain user anonymity. User accounts retain only minimal identifiers (username and email) necessary for login management. All data transmissions occur within a secure cloud environment hosted on Amazon Web Services. Text inputs are encrypted using Advanced Encryption Standard and Base64 (Base64 format is a way of turning the encrypted data into plain text characters so it can be stored or transmitted safely and consistently) during transfer and at rest. Access to stored information is role-restricted (only teachers can access their own chat history), and account access is automatically disabled after 90 days of inactivity unless re-authenticated. Conversations between teachers and the AI system are not accessible to mental health consultants, research staff, or system administrators, preserving confidentiality and ensuring that all AI interactions remain anonymized. Teachers may delete conversation logs at any time, after which records are immediately anonymized, retained for 30 days for system integrity purposes, and subsequently purged permanently, with associated backups removed after six months. Continuous monitoring protocols provide 24/7 detection of unauthorized access attempts and infrastructure anomalies, with automated response safeguards in place.

2.2.3. Matched-Control Condition: HC2

Teachers within the ECE centers that were assigned to the matched-control condition completed an obesity-prevention curriculum, HC2 (Natale et al., 2022a). A doctoral-level study member with expertise in child nutrition and exercise trained research assistants to deliver the intervention. The HC2 program focused on four pillars: physical activity, snack, beverage, and screen time. Each teacher’s classroom received lesson plans that outlined how to integrate each pillar into daily classroom activities, including the objective, preparation steps, relevant props, activities, language and vocabulary, and culturally and linguistically appropriate service associations and enrichment questions (Natale et al., 2022a). The HC2 program (similar to intervention conditions) was delivered over 14 weeks, and the 60-min sessions occurred in-person or virtually.

2.2.4. Consultant Training

The JS and JS Go mental health consultants were trained early childhood mental health professionals; most had endorsements from a state infant mental health association, and they all received ongoing reflective supervision from a licensed clinical psychologist. All mental health consultants completed extensive training in the JS program, which included five virtual onboarding sessions: (1) introduction to Georgetown University’s ECMHC and JS models, (2) program-level observation training, (3) classroom-level observation training, (4) curriculum infographic review, and (5) exemplar consultation video analysis. The JS Go mental health consultant completed a sixth session on how to integrate the JS Go app into consultation sessions with ECE teachers.

2.3. Measures

2.3.1. Demographics

At pre-intervention, teachers were asked to provide their own age, gender, race, ethnicity, preferred language, level of education, and years of experience in the ECE setting; parents were asked to provide their child’s age, gender, race, ethnicity, primary language spoken in the home, and English proficiency.

2.3.2. Child Measures

Prosocial behaviors that support resilience in young children were assessed using two versions of the DECA: the DECA for Infants and Toddlers (Powell et al., 2007) and the DECA for Preschoolers, Second Edition (LeBuffe & Naglieri, 2012). Both versions are validated, reliable, standardized, and norm-referenced teacher-report measures for children ages 1 month to 5 years old (LeBuffe & Naglieri, 1999; Naglieri et al., 2013). The DECA for Infants and Toddlers includes 36 items, and the DECA for Preschoolers includes 38 items. Teachers are prompted to rate each item on a 5-point Likert scale ranging from “Never” to “Very Frequently”. Both instruments yield three subscales (Initiative, Self-Regulation, and Attachment/Relationships) as well as a Total Protective Factors (TPF) score. Items begin with the prompt “During the past 4 weeks, how often did the toddler/child…” followed by specific behaviors (e.g., “show affection for familiar adults” and “calm herself/himself down”). Both versions of the DECA have demonstrated adequate internal consistency across English- and Spanish-speaking, low-income, and racially and ethnically diverse samples (Crane et al., 2011).

Similar to the DECA, the Strengths and Difficulties Questionnaire (SDQ) was administered at baseline and post-intervention to gauge teachers’ perceptions of child challenging behaviors (Goodman, 1997). The SDQ is a 25-item behavioral screening instrument for children and adolescents aged 2 to 17 years old, with different forms for ages 2 to 3 and ages 4 to 17. Items are rated on a 3-point Likert scale ranging from “Not True” to “Certainly True” and describe specific child behaviors, such as “Many worries or often seems worried,” “Easily distracted, concentration wanders,” and “Often offers to help others.” The Externalizing, Internalizing, and Total Problems scales were used in this study. The SDQ has established validity and reliability (Goodman, 1997) and demonstrated satisfactory internal consistency with preschool-aged children (Croft et al., 2015).

2.3.3. Teacher Measures

Classroom practices were assessed using the HERS-C, a 30-min observational assessment developed by the study authors, following an observation protocol described previously (Futterer et al., 2024; Natale et al., 2023a). HERS-C evaluates safety, behavioral support, communication, and resiliency coping, which correspond to national early childhood education program health and safety standards as well as the four pillars of JS and JS Go (Futterer et al., 2024; National Resource Center for Health and Safety in Child Care and Early Education, 2016). Trained consultants observed classrooms either in person or via a virtual telepresence robot and rated core classroom practices on a 7-point Likert scale ranging from “Little or No Implementation” to “Excellent Implementation”. HERS-C includes items such as “Teacher shares behavior expectations and classroom rules using positive language, praise, and redirection when needed” and “Classroom has safety guidelines and procedures for responding to crisis situations and adheres to Center for Disease Control/Department of Children and Families guidelines”. This measure demonstrated good internal consistency = 0.835 for the current sample of teachers.

The Childcare Worker Job Stress Inventory was used to measure teachers’ beliefs about their job by assessing three domains of workplace stress: Job Demands, Job Resources, and Job Control (Curbow et al., 2000). Each subscale comprises 17 items with response options provided on a 5-point Likert scale ranging from “Very Little” to “Very Much.” For example, the Childcare Worker Job Stress Inventory includes “I feel like I have to be a parent and a teacher to the children” (Job Demands), “I get praise from the parents for the work that I do” (Job Resources), and “How much control do you have over the number of children you care for?” (Job Control). Higher scores indicate greater perceived job-related demands, resources, and control. The measure has demonstrated convergent validity (Curbow et al., 2000) and strong internal consistency (Gray, 2015). This tool was selected to assess teacher stress and coping because it is a standardized measure consistently used across multiple iterations of the JS program, allowing for longitudinal comparison across models and cohorts. In addition, it has been widely used in the ECMHC literature to evaluate teacher work-related stress (Natale et al., 2025b, 2025c).

The Teacher Opinion Survey was used to evaluate teachers’ confidence in managing challenging child behaviors (Geller & Lynch, 1999). The measure consists of 12 items and implements a 5-point Likert scale ranging from “Strongly Disagree” to “Strongly Agree”, such that higher scores indicate greater teacher confidence. For example, the Teacher Opinion Survey includes “If I keep trying, I can find some way to reach even the most challenging child,” and “If a student in my class became disruptive and noisy, I feel pretty sure that I’d know how to respond effectively.” The total score demonstrated acceptable internal consistency (Alkon et al., 2003; Shamblin et al., 2016).

The Brief Resilient Coping Scale was used to assess teachers’ coping behaviors (Sinclair & Wallston, 2004). This measure comprises 4 items and includes a 5-point Likert scale ranging from “Does not describe me at all” to “Describes me very well”, such that higher scores indicate greater teacher resiliency. The measure’s internal consistency is Cronbach’s alpha = 0.69, and the test–retest reliability correlation = 0.68 (Sinclair & Wallston, 2004).

2.4. Statistical Analysis

We first examined baseline child and teacher characteristics by intervention condition (JS, JS Go, and HC2) using means and standard deviations (SD) for continuous variables and counts (percentages) for categorical variables. The study used a matched-cohort allocation at the child level to promote baseline comparability across groups, matching on children’s age, gender, and prosocial skills. Group differences in continuous variables were assessed with one-way analysis of variance (ANOVA); when Levene’s test indicated unequal variances, Welch’s ANOVA was used. Categorical variables were compared using χ² tests, with Fisher’s exact test applied when expected cell counts were <5.

To further document comparability following matching, we calculated standardized mean differences for the three matched child-level variables (age, gender, and DECA Total Protective Factors). Standardized mean differences were computed for JS Go vs. HC2 and JS vs. HC2 using standard formulas for continuous and binary variables.

We fit separate linear mixed-effects models for each child outcome (DECA TPF, Attachment/Relationships, Self-Regulation, and Initiative; SDQ Total, Externalizing, and Internalizing Problems) to evaluate change from baseline (T0) to post-intervention (T1) across the three intervention conditions. Models included fixed effects for time, intervention condition, and the time-by-intervention interaction, and were adjusted for baseline score of the outcome and pre-specified child covariates (age, gender, race, ethnicity, and primary language). Adjusted least-squares means with 95% confidence intervals were estimated for each group and timepoint.

Similarly, linear mixed effects models for each teacher outcomes; HERS classroom practices; Brief Resilient Coping Scale; Teacher Opinion Survey; and Childcare Worker Job Demands, Resources, and Control) were estimated using teacher-level repeated measures. Random intercepts were included for program and for teacher nested within program, with repeated measures at the child level (compound-symmetry covariance). Degrees of freedom were estimated by the Kenward-Roger method, and significance was set at a 2-sided α of 0.05. To evaluate post-intervention differences, least-squares means were compared pairwise between each intervention condition (JS and JS Go) and HC2 using Tukey–Kramer adjusted p values.

In addition, as a robustness check, we conducted a sensitivity analysis by re-estimating all child-level mixed-effects models with program (center) fixed effects to account for residual demographic imbalance and site-level heterogeneity. To help interpret precision given differences in the number of teachers per arm, we also estimated intraclass correlations for the primary child outcome (SDQ Internalizing) using variance components from the mixed-effects model. Program- and teacher-level random intercept variances were used to compute a combined intraclass correlation and design effect, which we then used to approximate the effective child sample size. Analyses were performed using SAS version 9.4 (SAS Institute, 2019).

3. Results

3.1. Baseline Characteristics

Children and teachers across the three conditions were generally comparable at baseline, although some demographic differences were observed. Child characteristics by intervention condition are summarized in

Table 1. Child Baseline Demographic Characteristics by Intervention Condition (n = 153).

Characteristic	JS Go (n = 51)	JS (n = 51)	HC2 (n = 51)	Total (n = 153)
Child age (years), mean (SD)	3.66 (0.72)	3.67 (0.81)	3.49 (0.76)	3.61 (0.77)
Child gender, n (%)
Female	22 (43.1)	22 (43.1)	22 (43.1)	66 (43.1)
Male	29 (56.9)	29 (56.9)	29 (56.9)	87 (56.9)
Child race, n (%)
White	43 (89.6)	40 (87.0)	24 (52.2)	107 (76.4)
Black	0 (0.0)	3 (6.5)	18 (39.1)	21 (15.0)
Native American	1 (2.1)	0 (0.0)	0 (0.0)	1 (0.7)
Multiracial	4 (8.3)	0 (0.0)	2 (4.4)	6 (4.3)
Other	0 (0.0)	3 (6.5)	2 (4.4)	5 (3.6)
Child ethnicity, n (%)
Hispanic	45 (93.8)	41 (89.1)	28 (59.6)	114 (80.9)
Non-Hisp White	2 (4.2)	3 (6.5)	1 (2.1)	6 (4.3)
Non-Hisp Black	0 (0.0)	0 (0.0)	13 (27.7)	13 (9.2)
Haitian	0 (0.0)	2 (4.3)	5 (10.6)	7 (5.0)
Other	1 (2.1)	0 (0.0)	0 (0.0)	1 (0.7)
Primary language, n (%)
English	9 (18.8)	6 (13.0)	29 (61.7)	44 (31.2)
Spanish	39 (81.3)	40 (87.0)	17 (36.2)	96 (68.1)
Creole	0 (0.0)	0 (0.0)	1 (2.1)	1 (0.7)

Fisher’s exact test used due to sparse cells; p-value is Fisher’s two-sided. p-value compares JS Go, JS, and HC2.; From General Linear Models/ANOVA (Welch test F = 0.88, p = 0.42); chi-square with Fisher’s exact confirmation; and Fisher p shown when <5 expected counts.

, and teacher characteristics are summarized in

Table 2. Teacher Baseline Demographic Characteristics by Intervention Group (n = 46).

Characteristic	JS Go (n = 5)	JS (n = 18)	HC2 (n = 23)	Total (n = 46)	p-Value 1
Teacher age (years), mean (SD)	36.0 (5.6)	42.0 (9.5)	46.4 (12.9)	43.4 (11.0)	0.0016 2
Gender, n (%)
Female	5 (100)	16 (88.9)	22 (95.7)	43 (93.5)	0.70 3
Male	0 (0.0)	2 (11.1)	1 (4.3)	3 (6.5)
Race, n (%)
White	4 (80.0)	13 (72.2)	16 (69.6)	33 (71.7)	0.33 3
Black	0 (0.0)	3 (16.7)	6 (26.1)	9 (19.6)
Multiracial	0 (0.0)	0 (0.0)	1 (4.3)	1 (2.2)
Other	1 (20.0)	2 (11.1)	0 (0.0)	3 (6.5)
Ethnicity, n (%)
Hispanic	5 (100)	16 (88.9)	17 (73.9)	38 (82.6)	0.39 3
Non-Hisp Black	0 (0.0)	0 (0.0)	4 (17.4)	4 (8.7)
Haitian	0 (0.0)	2 (11.1)	2 (8.7)	4 (8.7)
Primary language, n (%)
English	0 (0.0)	1 (5.6)	6 (26.1)	7 (15.2)	0.37 3
Spanish	5 (100)	16 (88.9)	16 (69.6)	37 (80.4)
Creole	0 (0.0)	1 (5.6)	1 (4.3)	2 (4.4)
Highest education, n (%)
Elementary or less	0 (0.0)	1 (5.9)	1 (4.3)	2 (4.4)	0.83 3
Some High School	0 (0.0)	0 (0.0)	1 (4.3)	1 (2.2)
High School/GED	0 (0.0)	4 (23.5)	1 (4.3)	5 (11.1)
Technical training	1 (20.0)	1 (5.9)	2 (8.7)	4 (8.9)
Some College	1 (20.0)	3 (17.6)	4 (17.4)	8 (17.8)
Associate Degree	1 (20.0)	1 (5.9)	2 (8.7)	4 (8.9)
Bachelor’s Degree	2 (40.0)	5 (29.4)	11 (47.8)	18 (40.0)
Graduate Degree	0 (0.0)	2 (11.8)	1 (4.3)	3 (6.7)

Fisher’s exact test used due to sparse cells; p-value is Fisher’s two-sided. ¹ p-value compares JS Go, JS, and HC2. ² Welch’s ANOVA F = 7.71, p = 0.0016 (Levene p = 0.0041 indicates unequal variances; Welch used). ³ Chi-square with Fisher’s exact confirmation; Fisher p shown when <5 expected counts. Note. Although the JS Go arm includes fewer teachers than JS and HC2, the combined program- and teacher-level intraclass correlation for SDQ Internalizing was modest (≈0.04), implying a design effect of ≈1.09 and an effective child sample size of ~140 children.

. Among 153 children (n = 51 per group), mean age was 3.61 years (SD = 0.77) and did not differ by group (F = 0.88, p = 0.42). Gender was also balanced (43.1% female across groups). However, race, ethnicity, and primary language differed significantly: HC2 included a higher proportion of Black children (39% vs. 0–6% in JS and JS Go), more non-Hispanic Black children (27.7% vs. 0%), and more English-speaking households (61.7% vs. 13–19%; all p < 0.001).

To evaluate baseline equivalence on the matched child-level variables, we examined standardized mean differences comparing JS Go and JS to HC2 (Supplemental Table S1). All standardized mean differences were small in magnitude (absolute standardized mean differences ranging from 0.00 to 0.23), indicating that the matching procedure achieved an acceptable balance for child age, gender, and prosocial skills at baseline.

Among 46 teachers (JS: n = 18, JS Go: n = 5, HC2: n = 23), mean age was 43 years (SD = 11), differing significantly across groups (F = 7.71, p = 0.002), with the mean teacher age 42 years in JS, 36 years in JS Go, and 46 years in HC2. Gender, race, ethnicity, primary language, and highest level of completed education showed no significant differences between groups. See Table 2.

3.2. Child Social–Emotional Outcomes

Across the intervention period, children demonstrated improvements in several social–emotional domains (Figure 1), with some variation in trajectories across intervention conditions. Estimated adjusted means and model effects are presented in

Table 3. Estimated Means (SE) for Child Social–Emotional Outcomes by Intervention Group and Time, With Main and Interaction Effects.

Outcome	Time	JS Go Mean (SE)	JS Go 95% CI	JS + Mean (SE)	JS + 95% CI	HC2 Mean (SE)	HC2 95% CI	Time Main Effect (F, p)	Intervention × Time (F, p)
DECA Total Protective Factors	Baseline	49.23 (2.05)	[45.10, 53.36]	49.94 (1.72)	[46.51, 53.37]	48.66 (1.68)	[45.33, 51.98]	F = 13.55, p = 0.0003	F = 1.55, p = 0.2150
	Follow-up	51.11 (2.06)	[46.95, 55.26]	52.12 (1.74)	[48.65, 55.58]	53.93 (1.85)	[50.27, 57.59]
DECA Attachment/Relationships	Baseline	46.09 (1.82)	[42.43, 49.75]	46.76 (1.58)	[43.62, 49.91]	45.82 (1.56)	[42.73, 48.90]	F = 25.92, p < 0.0001	F = 3.03, p = 0.0513
	Follow-up	48.37 (1.83)	[44.68, 52.07]	49.76 (1.60)	[46.59, 52.93]	52.59 (1.64)	[49.33, 55.85]
DECA Self-Regulation	Baseline	51.61 (2.20)	[47.15, 56.08]	51.85 (1.82)	[48.22, 55.48]	50.84 (1.78)	[47.30, 54.37]	F = 6.03, p = 0.0153	F = 0.80, p = 0.4526
	Follow-up	53.39 (2.22)	[48.90, 57.88]	52.97 (1.83)	[49.33, 56.60]	54.75 (1.90)	[50.98, 58.51]
DECA Initiative	Baseline	50.22 (2.02)	[46.15, 54.29]	51.06 (1.87)	[47.36, 54.77]	49.85 (1.82)	[46.24, 53.46]	F = 11.38, p = 0.0010	F = 2.27, p = 0.1066
	Follow-up	52.01 (2.04)	[47.91, 56.11]	52.33 (1.89)	[48.60, 56.07]	55.28 (1.99)	[51.35, 59.21]
SDQ Total Problems	Baseline	8.45 (1.04)	[6.36, 10.55]	8.04 (1.03)	[5.99, 10.08]	7.65 (0.99)	[5.69, 9.62]	F = 1.07, p = 0.3033	F = 0.05, p = 0.9487
	Follow-up	7.90 (1.05)	[5.79, 10.01]	7.77 (1.03)	[5.72, 9.81]	7.01 (1.05)	[4.94, 9.09]
SDQ Externalizing Problems	Baseline	5.49 (0.72)	[4.04, 6.95]	5.12 (0.73)	[3.68, 6.56]	4.65 (0.70)	[3.26, 6.04]	F = 0.40, p = 0.5305	F = 0.05, p = 0.9543
	Follow-up	5.42 (0.73)	[3.95, 6.88]	4.85 (0.73)	[3.41, 6.29]	4.33 (0.75)	[2.86, 5.81]
SDQ Internalizing Problems	Baseline	4.69 (0.49)	[3.70, 5.67]	4.09 (0.49)	[3.12, 5.05]	4.02 (0.47)	[3.10, 4.95]	F = 28.16, p < 0.0001	F = 9.65, p = 0.0001
	Follow-up	1.78 (0.50)	[0.79, 2.78]	3.25 (0.49)	[2.28, 4.22]	3.64 (0.50)	[2.65, 4.63]

Note. Bold values indicate statistically significant results (p < 0.05).

, and post-intervention pairwise comparisons are shown in

Table 4. Pairwise Intervention-Control Comparisons of Child Outcomes at T2 (Adjusted for Baseline).

Outcome	JS Go vs. HC2 Diff (SE)	JS Go vs. HC2 95% CI	Adj p	JS+ vs. HC2 Diff (SE)	JS+ vs. HC2 95% CI	Adj p
DECA Total Protective Factors	−2.82 (2.28)	[−7.45, 1.80]	0.816	−1.81 (1.95)	[−5.72, 2.10]	0.938
DECA Attachment/Relationships	−4.21 (1.93)	[−8.14, −0.28]	0.25	−2.83 (1.66)	[−−6.17, 0.51]	0.529
DECA Self-Regulation	−1.36 (2.44)	[−6.35, 3.64]	0.994	−1.78 (2.03)	[−5.88, 2.32]	0.951
DECA Initiative	−3.27 (2.13)	[−7.61, 1.06]	0.642	−2.95 (1.92)	[−6.80, 0.90]	0.644
SDQ Total Problems	0.89 (1.03)	[−1.23, 3.01]	0.954	0.75 (0.98)	[−1.21, 2.72]	0.972
SDQ Externalizing Problems	1.08 (0.73)	[−0.43, 2.60]	0.676	0.51 (0.71)	[−0.91, 1.93]	0.978
SDQ Internalizing Problems	−1.86 (0.52)	[−2.92, −0.81]	0.006	−0.40 (0.49)	[−1.38, 0.58]	0.966

Values are differences in Least Squares-means (SE) with Tukey–Kramer adjusted p-values. Note. Bold values indicate statistically significant results (p < 0.05).

. Children demonstrated significant improvements in several social–emotional domains related to prosocial behaviors. Across all groups, TPF that support resilience improved over time (F = 13.55, p < 0.001), although there was no intervention-by-time interaction (F = 1.55, p = 0.22). Mean adjusted scores increased from 49.94 (SE = 1.72) to 52.12 (SE = 1.74) in JS, 49.23 (SE = 2.05) to 51.11 (SE = 2.06) in JS Go, and 48.66 (SE = 1.68) to 53.93 (SE = 1.85) in HC2.

Within the DECA Attachment and Relationships domain, a significant time effect emerged (F = 25.92, p < 0.0001) and a near-significant intervention interaction (F = 3.03, p = 0.051). Scores increased from 46.76 (SE = 1.58) to 49.76 (SE = 1.60) in JS, 46.09 (SE = 1.82) to 48.37 (SE = 1.83) in JS Go, and 45.82 (SE = 1.56) to 52.59 (SE = 1.64) in HC2.

Children also demonstrated significantly improved scores on the DECA Self-Regulation domain over time (F = 6.03, p = 0.02), without interaction (F = 0.80, p = 0.45). Mean scores increased from 51.85 (SE = 1.82) to 52.97 (SE = 1.83) in JS, 51.61 (SE = 2.20) to 53.39 (SE = 2.22) in JS Go, and 50.84 (SE = 1.78) to 54.75 (SE = 1.90) in HC2.

Time effect was also significant within the DECA Initiative domain (F = 11.38, p = 0.001), with no significant interaction (F = 2.27, p = 0.11).

In contrast, a strong time effect (F = 28.16, p < 0.001) and a robust intervention-by-time interaction (F = 9.65, p < 0.001) emerged within the SDQ Internalizing Problems subscale (Figure 2; Table 3). Adjusted means decreased from 4.09 (SE = 0.49) to 3.25 (SE = 0.49) in JS, from 4.69 (SE = 0.49) to 1.78 (SE = 0.50) in JS Go, and from 4.02 (SE = 0.47) to 3.64 (SE = 0.50) in HC2. Pairwise comparisons at post-intervention confirmed that children within JS Go had significantly lower internalizing scores than children within HC2 (difference = −1.86 [SE = 0.52], 95% CI: −2.92, −0.81, p = 0.01, Tukey–Kramer adjusted), while JS did not differ significantly from HC2 (p = 0.97) (Table 4).

At post-intervention, no significant group differences were found within the DECA Initiative, Self-Regulation, or TPF scores or within the SDQ Total or Externalizing Problems subscales (all p > 0.10).

Sensitivity analyses using center fixed effects produced results that were virtually identical to the primary models. The treatment-by-time interaction remained significant only for SDQ Internalizing (F(2,124) = 9.65, p = 0.0001), with all other outcomes showing no treatment-specific change over time. These findings confirm that the JS Go effects are robust to residual imbalance and center-level variation (Supplemental Table S2).

The combined program- and teacher-level intraclass correlation for SDQ Internalizing was modest (Intraclass correlation ≈ 0.04), corresponding to a design effect of ≈1.09 and an effective sample size of ≈140 children (vs. 153 nominal), indicating that precision was not meaningfully limited by the smaller number of teachers in JS Go.

3.3. Teacher and Classroom Outcomes

Detailed estimated means and model effects of teachers and classroom outcome are provided in

Table 5. Estimated Means (SE) for Teacher and Classroom Outcomes by Intervention Group and Time, With Main and Interaction Effects.

Outcome	Time	JS Go Mean (SE)	JS Go 95% CI	JS+ Mean (SE)	JS+ 95% CI	HC2 Mean (SE)	HC2 95% CI	Time Main Effect (F, p)	Intervention × Time (F, p)
HERS Classroom Safety	Baseline	4.69 (0.14)	[4.41, 4.97]	4.75 (0.08)	[4.59, 4.90]	4.68 (0.08)	[4.52, 4.84]	F = 12.88, p = 0.0010	F = 0.11, p = 0.8986
	Follow-up	5.02 (0.16)	[4.71, 5.34]	5.11 (0.08)	[4.95, 5.26]	4.96 (0.08)	[4.80, 5.13]
HERS Classroom Behavior	Baseline	4.56 (0.12)	[4.33, 4.80]	4.51 (0.07)	[4.38, 4.64]	4.60 (0.07)	[4.47, 4.73]	F = 9.66, p = 0.0037	F = 5.05, p = 0.0118
	Follow-up	5.00 (0.13)	[4.74, 5.26]	4.90 (0.07)	[4.77, 5.03]	4.54 (0.07)	[4.40, 4.69]
HERS Classroom Communication	Baseline	4.28 (0.22)	[3.85, 4.72]	4.51 (0.12)	[4.27, 4.75]	4.54 (0.12)	[4.30, 4.77]	F = 0.51, p = 0.4778	F = 0.60, p = 0.5541
	Follow-up	4.66 (0.24)	[4.18, 5.14]	4.45 (0.12)	[4.21, 4.68]	4.54 (0.13)	[4.29, 4.80]
HERS Classroom Resiliency	Baseline	3.82 (0.21)	[3.41, 4.24]	3.85 (0.12)	[3.61, 4.09]	4.20 (0.12)	[3.96, 4.44]	F = 14.84, p = 0.0005	F = 8.95, p = 0.0007
	Follow-up	5.11 (0.23)	[4.65, 5.57]	4.42 (0.12)	[4.18, 4.66]	4.00 (0.13)	[3.75, 4.25]
Brief Resiliency Scale	Baseline	18.16 (0.75)	[16.69, 19.63]	17.81 (0.40)	[17.03, 18.59]	17.84 (0.35)	[17.16, 18.52]	F = 0.70, p = 0.4067	F = 1.99, p = 0.1492
	Follow-up	16.16 (0.75)	[14.69, 17.63]	18.28 (0.40)	[17.50, 19.06]	18.19 (0.38)	[17.45, 18.93]
Teacher Opinion Survey	Baseline	47.68 (1.80)	[44.15, 51.21]	46.88 (1.06)	[44.80, 48.96]	47.57 (0.96)	[45.69, 49.45]	F = 0.18, p = 0.6714	F = 0.90, p = 0.4147
	Follow-up	44.68 (1.80)	[41.15, 48.21]	47.63 (1.06)	[45.55, 49.71]	48.37 (1.05)	[46.31, 50.43]
Childcare Worker Job Demands	Baseline	2.74 (0.27)	[2.22, 3.26]	2.70 (0.15)	[2.41, 2.99]	2.70 (0.13)	[2.45, 2.95]	F = 1.34, p = 0.2539	F = 0.72, p = 0.4914
	Follow-up	2.64 (0.27)	[2.12, 3.16]	3.12 (0.15)	[2.83, 3.41]	2.95 (0.14)	[2.68, 3.22]
Childcare Worker Job Resources	Baseline	4.61 (0.19)	[4.24, 4.98]	4.52 (0.11)	[4.30, 4.74]	4.59 (0.09)	[4.41, 4.77]	F = 0.09, p = 0.7649	F = 0.38, p = 0.6866
	Follow-up	4.44 (0.19)	[4.07, 4.81]	4.62 (0.11)	[4.40, 4.84]	4.52 (0.10)	[4.32, 4.72]
Childcare Worker Job Control	Baseline	3.44 (0.23)	[2.99, 3.89]	3.26 (0.13)	[3.01, 3.51]	3.22 (0.11)	[3.00, 3.44]	F = 4.75, p = 0.0350	F = 0.78, p = 0.4671
	Follow-up	2.82 (0.23)	[2.37, 3.27]	3.01 (0.13)	[2.76, 3.26]	3.11 (0.12)	[2.87, 3.35]

Note. Bold values indicate statistically significant results (p < 0.05).

, and intervention-specific contrasts at follow-up appear in

Table 6. Pairwise Intervention-Control Comparisons of Teacher Outcomes at T2 (Adjusted for Baseline).

Outcome	JS Go vs. HC2 Diff (SE)	JS Go vs. HC2 95% CI	Adj p	JS+ vs. HC2 Diff (SE)	JS+ vs. HC2 95% CI	Adj p
HERS Classroom Safety	0.06 (0.18)	[−0.29, 0.41]	0.999	0.14 (0.12)	[−0.09, 0.37]	0.819
HERS Classroom Behavior	0.46 (0.15)	[0.16, 0.75]	0.042	0.36 (0.10)	[0.16, 0.55]	0.011
HERS Classroom Communication	0.12 (0.27)	[−0.43, 0.66]	0.998	−0.10 (0.17)	[−0.44, 0.25]	0.993
HERS Classroom Resiliency	1.11 (0.27)	[0.57, 1.65]	0.003	0.42 (0.18)	[0.05, 0.79]	0.24
Brief Resiliency Scale	−2.02 (0.84)	[−3.70, −0.35]	0.18	0.09 (0.56)	[−1.01, 1.20]	1
Teacher Opinion Survey	−3.65 (1.99)	[−7.63, 0.32]	0.46	−0.65 (1.38)	[−3.41, 2.10]	1
Childcare Worker Job Demands	−0.32 (0.29)	[−0.90, 0.26]	0.88	0.19 (0.19)	[−0.20, 0.57]	0.92
Childcare Worker Job Resources	−0.08 (0.21)	[−0.50, 0.35]	0.999	0.10 (0.14)	[−0.19, 0.39]	0.98
Childcare Worker Job Control	−0.30 (0.26)	[−0.81, 0.22]	0.86	−0.11 (0.17)	[−0.45, 0.23]	0.99

Values are differences in Least Squares-means (SE) with Tukey–Kramer adjusted p-values. Note. Bold values indicate statistically significant results (p < 0.05).

. Teacher’s classroom practices demonstrated multiple positive changes across the study period (Figure 3). Significant main time effect (F = 9.66, p = 0.004) and intervention-by-time interaction (F = 5.05, p = 0.01) emerged within the HERS Classroom Behavior subscale. Mean post-intervention scores were 4.90 (SE = 0.07) in JS, 5.00 (SE = 0.13) in JS Go, and 4.54 (SE = 0.07) in HC2. Pairwise contrasts showed that teachers within both JS (difference = 0.36 [SE = 0.10], 95% CI: 0.16, 0.55, p = 0.01) and JS Go (difference = 0.46 [SE = 0.15], 95% CI: 0.16, 0.75, p = 0.04) demonstrated significantly greater improvements in classroom behavior practices compared to HC2.

Mean scores assessing classroom resiliency practices increased moderately in JS (3.85 [SE = 0.12] to 4.42 [SE = 0.12]) and significantly in JS Go (from 3.82 [SE = 0.21] to 5.11 [SE = 0.23]) while decreasing slightly in HC2 (4.20 [SE = 0.12] to 4.00 [SE = 0.13]). Both the main effect of time (F = 14.84, p < 0.001) and the interaction (F = 8.95, p < 0.001) were significant. Pairwise comparisons showed that JS Go improved significantly relative to HC2 (difference = 1.11 [SE = 0.27], 95% CI: 0.57, 1.65; p = 0.003), while JS did not differ significantly from HC2 (p = 0.24). Figure 3 depicts changes in HERS Classroom practices over time for the three intervention groups.

Evaluations of teacher classroom safety practices also improved significantly over time (F = 12.88, p = 0.001), but significant intervention differences did not emerge (F = 0.11, p = 0.90). In addition, no significant time or intervention effects were observed within the HERS Classroom Communication, Brief Resilient Coping Scale, or Teacher Opinion Survey (all p > 0.10).

Measures of teacher workplace stress (job demands, resources, and control) did not significantly change across time or intervention. Although Job Control declined modestly over time (F = 4.75, p = 0.04), intervention effects were nonsignificant (F = 0.78, p = 0.47). Pairwise comparisons also confirmed no significant differences between JS or JS Go relative to HC2 on the Teacher Opinion Survey or the Brief Resilient Coping Scale as seen in Table 5 and Table 6.

4. Discussion

This study evaluated the effectiveness of two early childhood consultation models (JS and JS Go) in promoting children’s social–emotional development and enhancing teachers’ classroom practices, relative to an attention control group (HC2). Although we hypothesized that both JS conditions would outperform the attention control (HC2) on all child and teacher outcomes, study findings revealed a more nuanced pattern. Across JS and JS Go intervention groups, children and teachers demonstrated significant improvements over the span of the study, yet between-group differences varied. Specifically, the most consistent advantages emerged for JS Go, particularly in reducing children’s internalizing behaviors and improving teachers’ classroom resiliency and behavioral support practices, relative to JS and HC2.

4.1. Child Social–Emotional Outcomes

Across child outcomes, several prosocial domains showed improvements for all groups. Consistent with prior ECMHC research (Natale et al., 2025b; Silver et al., 2023; Stein et al., 2024), children across all groups exhibited significant improvements in child prosocial skills, including attachment, self-regulation, and initiative. These findings highlight the capacity of all three consultation models in strengthening children’s prosocial skills through enhanced teacher support and the consistent use of implemented classroom strategies. That is, both ECMHC and obesity prevention interventions frequently operate through caregiver-mediated strategies, enhancing the capacity of teachers to provide responsive, supportive, and structured environments. These approaches possibly foster child secure attachment, emotional competence, and self-regulation by promoting sensitive, consistent adult-child interactions and co-regulation practices (Finlay-Jones et al., 2021; Housman, 2017; Silver et al., 2023).

While JS and JS Go were expected to outperform the attention control, the similarity in overall child prosocial skill improvements across groups may be partly explained by the indirect benefits of the HC2 program. Although HC2 was designed as an obesity prevention curriculum, its focus on health-promoting classroom practices such as structured physical activity, nutritious eating, and reduced screen time likely supported children’s emotional regulation, focus, and planning (Becker et al., 2014). This active control group may have therefore reduced the observed effect size, leading to an underestimation of the true impact of the JS and JS Go programs. In fact, obesity-prevention programs often strengthen routines, nutrition, physical activity, and screen-time practices, which collectively enhance children’s sleep, energy regulation, and physiological readiness to learn. These improvements can indirectly boost emotional regulation, attention, and overall behavior, leading to stronger-than-expected effects even in the absence of explicit behavioral content (Guzmán-Muñoz et al., 2025). Therefore, future studies should more explicitly examine how integrated approaches that combine health and mental health promotion may yield complementary benefits for young children’s behavior, prosocial development, and overall well-being. Alternatively, future research may want to include active controls that are less likely to impact child prosocial skills such as injury prevention or creative story telling curriculums.

With respect to child externalizing and internalizing behaviors, JS Go demonstrated the most pronounced improvement in children’s behavioral functioning. Children in JS Go classrooms showed significant reductions in SDQ Internalizing Problems relative to HC2, suggesting that the AI-enhanced format effectively supported teacher use of emotion-coaching and resiliency strategies. It is possible that the JS Go’s on-demand guidance and brief, video-based learning modules and AI chatbot may have helped teachers integrate personalized strategies more consistently between consultations, bridging the gap between learning and daily practice. However, no significant changes were observed in children’s externalizing behaviors across time or intervention condition. One possible explanation is that the majority of children began the intervention with externalizing behaviors already within the typical developmental range, leaving limited opportunity for measurable improvement. When baseline levels of hyperactivity, impulsivity, or conduct problems are low, ceiling effects can attenuate apparent intervention gains even when teachers apply new behavior-support strategies effectively.

4.2. Teacher and Classroom Outcomes

Whereas prosocial skills improved across all three groups, likely reflecting shared classroom supports and general developmental growth, selective advantages emerged for JS Go in domains more sensitive to consultation strategy uptake, such as internalizing symptoms and teacher resiliency practices. At the teacher and classroom levels, Jump Start interventions outperformed HC2, particularly in domains central to social–emotional and behavioral support classroom practices. Teachers in JS Go and JS conditions demonstrated greater improvements in classroom behavioral management practices and overall classroom resiliency practices relative to teachers assigned to HC2. Specifically, teachers in JS Go classrooms reported greater adaptability, problem-solving, and optimism, which are key dimensions of resilience known to buffer against occupational stress and improve child outcomes (Natale et al., 2025b). Notably, HC2 classrooms showed slight declines over time, reinforcing the added value of structured, relationship-based professional development offered through Jump Start.

These domains represent core components of emotionally supportive environments (Gu & Day, 2007; Stein et al., 2024). Further, the stronger gains in JS Go suggest that AI-enhanced coaching in combination with human consultation can be as effective and in some cases more efficient than traditional in-person consultation. This is especially notable given the shorter live consultation contact time (30 min vs. 60 min) in JS Go relative to traditional JS. The asynchronous AI-based support may have enabled more frequent practice, self-reflection, and reinforcement of key strategies.

No significant between-group differences were detected in classroom safety or communication practices. Given that both JS and HC2 models emphasized structured routines, health-promoting activities, and family communication within their consultation models, it is possible that these factors may have produced convergence across groups. Similarly, the absence of significant change in teacher self-reported stress (job demands and job resources), teacher coping and confidence may suggest that, while consultation enhanced observable classroom quality, it did not substantially alter teachers’ broader work environment or systemic pressures. This divergence between improvements in practice-proximal outcomes (e.g., behavioral management, resiliency) and the relative stability of distal organizational indicators (e.g., job stress, perceived job control) is consistent with the variable ‘reach’ of consultation: strategies that directly support classroom interactions are more immediately influenced by coaching, whereas broader workplace climate reflects systemic conditions beyond the scope of an individual-level intervention (See Supplemental Figure S1). These findings are consistent with prior ECMHC research showing that meaningful changes in organizational climate typically require center-level support and sustained multi-tiered interventions beyond classroom consultation (Natale et al., 2023b).

Interestingly, a modest decline in teachers’ stress measured by perception of job control was observed across all groups over time. It is possible that this pattern may reflect ongoing workforce instability and structural challenges within the early childhood sector rather than a direct effect of consultation participation. Teachers may have experienced reduced autonomy due to staffing shortages, heightened post-pandemic regulations, or increased emphasis on program fidelity and data collection. It is also important to interpret these trajectories in light of baseline caregiver demographic differences across groups. Although children were matched on age, gender, and prosocial skills, variations in factors such as caregiver education, language, and classroom composition may have influenced the rate or magnitude of change. These contextual differences could contribute to differential responsiveness to consultation supports and should be examined more systematically in future studies.

Overall, these findings suggest that while JS and JS Go successfully strengthened observable classroom practices most directly tied to children’s social–emotional learning, broader workforce indicators such as systemic workplace stress and perceived control may require organizational-level strategies and policy supports to achieve parallel gains (Natale et al., 2023b). While consultation can enhance teachers’ coping and classroom practices, it may not fully offset structural issues such as low pay, high ratios, and limited administrative support. Moreover, modest declines in perceived job control across groups may reflect ongoing post-pandemic workforce challenges and staffing shortages within participating centers.

4.3. Interpretation and Implications

Taken together, these findings suggest that Jump Start, particularly the JS Go format, produced selective but meaningful advantages in classroom practices and certain child outcomes relative to HC2. The clearest differences emerged in teacher and classroom domains (behavioral management and resiliency) and in children’s internalizing symptoms, where JS Go demonstrated superior improvements. These results indicate that a combination of targeted professional development, structured coaching, and classroom-based strategies can enhance teacher practice and promote children’s socioemotional wellbeing in early childhood settings.

The results also highlight the value of flexible implementation models. JS Go, while less intensive than JS, demonstrated comparable or stronger outcomes in several domains, suggesting that lower-dose or hybrid models may achieve high impact when grounded in robust coaching and fidelity frameworks. Future research should further explore mechanisms underlying these differences, particularly how teacher resilience, stress reduction, and classroom climate mediate improvements in children’s emotional adjustment. Continued attention to gender, cultural responsiveness, and sustainability will be key to optimizing scalability across diverse early childhood settings.

4.4. Limitations

While children across the three intervention groups were matched on gender, age, and children’s prosocial skills, the selection process can still lead to potential bias in the sample. Specifically, matching on a few key variables cannot account for all unmeasured differences (e.g., teacher skill, teacher attitudes about the intervention) that may influence child outcomes, leaving potential selection bias. Further, matching participants ultimately reduces the overall sample size and may limit generalization of the findings to a broader population. Future research should prospectively conduct a cluster-randomized three-condition trial that will reduce the potential risk of bias in samples across intervention groups.

An important consideration is the strength of the comparison condition. HC2 may have provided greater benefit than initially anticipated, particularly during the COVID-19 pandemic, which may have reduced the observable differences between groups.

Furthermore, longer-term follow-up is warranted to evaluate whether JS Go and JS produce delayed improvements that were not captured within the study’s timeframe.

5. Conclusions

This study provides initial evidence that early childhood consultation can meaningfully strengthen social–emotional development, but the pattern of effects makes clear that how consultation is delivered matters. The advantages of JS Go, particularly in reducing children’s internalizing symptoms and improving teachers’ behavioral support and resiliency, suggest that flexible, technology-enhanced coaching may help translate consultation into daily practice more effectively than traditional formats. At the same time, the limited changes in broader workforce stress highlight the need for complementary organizational supports if consultation is to achieve its full potential. Taken together, the findings point toward a next generation of ECMHC models that pair relationship-based consultation with scalable, adaptive tools to better meet the realities of today’s early childhood settings.