How do behavior problems change over time in childhood? Evidence from the early childhood longitudinal study

Behavior problems (internalizing and externalizing) are prevalent from early to later childhood and are linked to adverse adult outcomes. Prior research on developmental patterns has been inconsistent, often assuming homogeneity of change across children. Drawing on dynamic systems theory (nonlinear, time-embedded development) and acknowledging individual differences, this study investigates heterogeneous trajectories of internalizing and externalizing problems in a large, ethnically diverse longitudinal cohort. Guided by ecological systems theory, it also examines child (e.g., gender, race), family (e.g., socioeconomic status), and psychological (e.g., self-control) predictors of these trajectories. The study addresses two questions: (1) How do behavior problems change during childhood? (2) What factors influence the developmental patterns or trajectories of behavior problems?

Evidence on developmental trends is mixed: some studies report stability or increases in externalizing and stability or decreases in internalizing during childhood, whereas others report decreases in both. Person-centered approaches have shown multiple subgroups (typically 3–5) for externalizing, but fewer studies have examined internalizing subgroups. Individual differences are theorized to shape heterogeneous developmental pathways. Multiple predictors have been implicated: gender (boys often show higher externalizing; girls differing trends), race, family socioeconomic status (SES), and psychological factors such as self-control. Ecological systems theory suggests these multilevel influences and their interactions shape trajectories, supporting examination of demographic and psychological predictors alongside trajectory identification.

Design and data source: The study used nine waves of data from the Early Childhood Longitudinal Study, Kindergarten Class of 2010–2011 (ECLS-K:2011), which followed ~18,170 kindergartners from 2010–2011 through 2015–2016 with information from children, parents, teachers, and schools. Data collection occurred twice per year in the first three school years and once (spring) in grades 3–5. Sample: Teacher-reported internalizing and externalizing problem behaviors across nine waves were analyzed; parent interviews provided demographics. Children with special needs and those with only one wave of behavior-problem data were excluded. Final analytic samples were n = 11,179 for internalizing and n = 11,185 for externalizing (as indicated in Tables 2 and 3). Participants were ages ~5–13 across waves; chi-squared tests indicated no significant demographic differences across waves (gender, age, race, SES; p > 0.05). Measures: Behavior problems and self-control were assessed via the teacher-reported Social Rating Scale (SRS; adapted from Gresham & Elliott, 1990). Subscales: self-control (4 items), interpersonal skills (5 items), externalizing problem behaviors (6 items), and internalizing problem behaviors (4 items). Items rated on a 4-point frequency scale; higher scores reflect more frequent behaviors. Internal consistency (Cronbach’s alpha) across waves: internalizing = 0.76–0.79; externalizing = 0.86–0.89. Self-control measured at Wave 1 (alpha = 0.81). Sociodemographic variables: gender and race at Wave 1; age each wave; SES at Waves 1, 4, and 9 per NCES procedures. Analytic strategy: After testing homogeneity across waves (chi-squared tests), growth mixture modeling (GMM) was used. Unconditional GMM identified latent trajectory classes for internalizing and externalizing problems. Conditional GMM examined predictors (gender, race, SES at Waves 1/4/9, and Wave 1 self-control) of class membership and within-class growth factors (intercepts, slopes). Model fit indices included BIC, SSABIC, Entropy; Lo-Mendell-Rubin adjusted LRT (aLRT) and bootstrapped LRT (BLRT) tested whether an n-class solution improved fit over n−1. Missing data were handled via full-information maximum likelihood (FIML). Analyses were conducted in Mplus 7.4. Ethics approvals and data protections followed NCES guidelines for public data use.

Descriptives and preliminary tests: Demographic characteristics were comparable across the nine waves (p > 0.05), supporting pooling across waves. Trajectory identification: - Internalizing problems (n = 11,179): Three latent classes were identified (preferred model per Table 2): • Low-increased pattern: 5.88% — low initial levels with increasing slopes. • Low-stable pattern: 87.23% — low initial levels with slight increase, largely stable. • Medium-decreased pattern: 6.89% — medium initial levels with decreasing slopes. - Externalizing problems (n = 11,185): Five latent classes were identified (preferred model per Table 3): • High-decreased pattern: 4.05% — high initial levels with decreasing slopes. • Medium-decreased pattern: 11.64% — medium initial levels with decreasing slopes. • Low-very high increased pattern: 0.20% — low initial levels with sharp increases. • Low-increased pattern: 10.45% — low initial levels with increases. • Low-stable pattern: 73.66% — low initial levels with slight increase, largely stable. Conditional models: Model fit improved when adding predictors (internalizing BIC = 39,596.861; SSABIC = 39,444.336; externalizing BIC = 37,181.377; SSABIC = 36,819.131). Predictors of class membership (selected comparisons, as reported): - Internalizing: Race, gender, and Wave 1 self-control predicted class membership. Comparing Class 1 vs Class 3, non-white race (β ≈ −0.105, OR ≈ 0.900, p < 0.05), girls (β = −0.436, OR = 0.647, p < 0.001), and lower self-control at Wave 1 (β = −0.581, OR = 0.560, p < 0.001) were more likely to be in Class 3 (medium-decreased). Comparing Class 2 vs Class 3, boys (β = 0.789, OR = 2.202, p < 0.001) and higher Wave 1 self-control (β = 0.401, OR = 1.493, p < 0.001) were more likely to be in Class 3. - Externalizing: Gender, SES, and Wave 1 self-control predicted class membership (baseline Class 5, low-stable). Compared to Class 5: Class 1 vs 5 — higher SES at Wave 9 (β = 1.247, OR = 3.479, p < 0.01) and higher Wave 1 self-control (β = 1.213, OR = 3.364, p < 0.05) associated with Class 5; Class 2 vs 5 — higher SES at Wave 9 (β = 0.882, OR = 2.416, p < 0.05) and higher Wave 1 self-control (β = 1.714, OR = 5.551, p < 0.01) associated with Class 5; Class 3 vs 5 — higher Wave 1 self-control (β = 1.860, OR = 6.427, p < 0.05) associated with Class 5; Class 4 vs 5 — girls (β = 1.417, OR = 4.124, p < 0.05), higher SES at Wave 9 (β = 0.982, OR = 2.670, p < 0.01), and higher Wave 1 self-control (β = 2.900, OR = 18.169, p < 0.001) associated with Class 5. Within-class predictors of growth factors: - Internalizing: Race (β = −0.008, p < 0.05) and Wave 1 self-control (β = −0.134, p < 0.001) negatively predicted intercepts; gender (β = 0.007, p < 0.05) and Wave 1 self-control (β = 0.013, p < 0.001) positively predicted slopes. - Externalizing: Wave 1 self-control negatively predicted intercepts across classes (e.g., C1 β = −0.341, p < 0.01; C2 β = −0.504, p < 0.001; C3 β = −0.556, p < 0.01; C4 β = −0.206, p < 0.001; C5 β = −0.614, p < 0.001). Self-control positively predicted slopes in several classes (e.g., C1 β = 0.068, p < 0.01; C2 β = 0.043, p < 0.001; C4 β = 0.021, p < 0.001; C5 β = 0.084, p < 0.05). In Class 4, gender negatively predicted slope (β ≈ −0.005, p < 0.05). In Class 5, SES at Wave 4 positively predicted slope (β = 0.087, p < 0.001) and SES at Wave 9 negatively predicted slope (β = −0.046, p < 0.05). Overall, trajectories were heterogeneous, with most children in low-stable patterns and meaningful minorities showing increasing or decreasing trends. Predictors at individual (gender, race, self-control) and family (SES) levels were associated with both class membership and within-class change.

Findings demonstrate heterogeneous developmental pathways for internalizing (three classes) and externalizing (five classes) problems from kindergarten through fifth grade. Most children followed low-stable trajectories, but notable subgroups exhibited increasing or decreasing trends, underscoring individual differences in development and aligning with dynamic systems theory. Multilevel predictors—gender, race, SES, and self-control—were linked to both class membership and within-class growth, supporting ecological systems theory. Protective patterns emerged: for internalizing, white race, female gender, and higher early self-control; for externalizing, higher SES, female gender, and higher early self-control. These results highlight the need for tailored prevention and intervention strategies that consider subgroup-specific developmental patterns and leverage modifiable factors such as self-control, alongside supports addressing socioeconomic disadvantage and attention to boys and minority children at higher risk.

Using nine waves of ECLS-K:2011 data, the study identified three internalizing trajectories (low-increased, low-stable, medium-decreased) and five externalizing trajectories (high-decreased, medium-decreased, low-very high increased, low-increased, low-stable). Class membership and within-class changes were predicted by child characteristics (gender, race), family SES, and early self-control. The results reinforce that behavior problem development is heterogeneous and emphasize incorporating individual differences into research and practice. Future work should integrate multiple informants, include time-variant predictors, and examine developmental outcomes (e.g., interpersonal and academic skills) associated with distinct trajectory classes.

- Behavior problems were assessed via teacher report only; parent reports were not included, potentially limiting perspective. - Predictors were time-invariant in this analysis; omission of time-varying predictors may miss dynamic influences on trajectories. - Outcomes of behavior problems (e.g., interpersonal skills, reading, writing) were not examined, limiting understanding of the consequences of different trajectories. - Children with special needs were excluded, so findings may not generalize to that population.