How Early is Too Early? Identification of Elevated, Persistent Problem Behavior in Childhood

Abstract

We inquire how early in childhood children most at risk for problematic patterns of internalizing and externalizing behaviors can be accurately classified. Yearly measures of anxiety/depressive symptoms and aggressive behaviors (ages 6–13; n=334), respectively, are used to identify behavioral trajectories. We then assess the degree to which limited spans of yearly information allow for the correct classification into the elevated, persistent pattern of the problem behavior, identified theoretically and empirically as high-risk and most in need of intervention. The true-positive rate (sensitivity) is below 70% for anxiety/depressive symptoms and aggressive behaviors using behavioral information through ages 6 and 7. Conversely, by age 9, over 90% of the high-risk individuals are correctly classified (i.e., sensitivity) for anxiety/depressive symptoms, but this threshold is not met until age 12 for aggressive behaviors. Notably, the false-positive rate of classification for both high-risk problem behaviors is consistently low using each limited age-span of data (<5%). These results suggest that correct classification into highest risk groups of childhood problem behavior is limited using data observed at early ages. Prevention programming targeting those who will display persistent, elevated levels of problem behavior should be cognizant of the degree of misclassification and how this varies with the accumulation of behavioral information. Continuous assessment of problem behaviors is needed throughout childhood in order to continually identify high-risk individuals most in need of intervention as behavior patterns are sufficiently realized.

Achenbach and Edelbrock’s (1978) dimensional model of childhood psychopathology classifies dysfunctional, problem behaviors as either externalizing or internalizing in nature. Externalizing behaviors include aggressive behaviors as well as disruptiveness, hyperactivity, and delinquency. Internalizing behaviors include anxiety and depressive symptoms. Notably, aggressive behaviors and anxiety/depressive symptoms in childhood forecast problematic development. In the near-term, they can affect school and social functioning, disrupting an individual’s educational and relational trajectories (e.g., Campbell, Spieker, Burchinal & Poe, 2006). Moreover, children who exhibit high levels of aggression and/or anxiety/depressive symptoms are more likely to be victims of peer aggression and bullying (Fekkes et al., 2006; Glew, Rivara & Feudtner, 2000; Roland, 2002; Seals & Young, 2003). Importantly, the consequences resulting from persistent patterns of aggressive behavior and anxiety/depressive symptoms in childhood set into motion a sequelae of negative socialization processes and transactional relations among the child, parents, and peers that increase the risk for future problem behavior, including criminal involvement, alcohol and drug abuse, and depressive disorders (e.g., Broidy et al., 2003; Costello et al., 2002; Dodge et al., 2009; Hussong et al., 2016; Kokko, Tremblay, Lacourse, Nagin & Vitaro, 2006; Lussier, Farrington & Moffitt, 2009). As such, persistent displays of these behaviors or symptoms in childhood are public health concerns. Fortunately, there is evidence to suggest that these behavioral manifestations can be effectively treated (Blueprints for Healthy Youth Development, 2019). Advocates recommend intervention as early as possible to ward off persistence and future maladaptation in order to limit future costs to the individual and society (Kellan et al., 2008).

Targeted intervention strategies focus on high-risk youth (i.e., young people who are at high risk for maladaptive development), and accurate identification of high-risk individuals is needed. Without accurate identification, the impact of the intervention is weakened. For instance, iatrogenic effects are possible among false positives (those who were incorrectly classified as high-risk and subsequently targeted for intervention) as a result of contagion or stigma (e.g., Poulin, Dishion & Burraston, 2001). False positives are also a waste of resources. On the other hand, false negatives are those individuals denied intervention but who ultimately need services. In short, these are missed opportunities for prevention and treatment.

The developmental perspective is often used to inform accurate identification of high-risk individuals. Various theoretical frameworks suggest that biological dispositions and environmental contexts place children at risk for elevated, persistent patterns of problem behavior (e.g., Dodge & Pettit, 2003; Moffitt, 1993; Loeber et al., 1989). These perspectives often invoke the terminology of “early-starters” or “life-course persistent,” and “adolescent-limited” to describe patterns of problem behavior. Genetic underpinnings are one source of dysregulated behavior. Traits resulting from genetic factors or neurophysiological vulnerabilities render children ill-equipped to manage ordinary tasks of social life, particularly when faced with environmental adversity (e.g., poverty or poor family functioning; Dodge & Pettit, 2003; Moffitt, 1993). The result is a heterotypic display of persistent problem behaviors, including internalizing behaviors such as anxiety/depressive symptoms or externalizing behaviors including aggression, which begin at an early age and persist through adulthood.

The notion of stability is confirmed by research indicating that behavior at age 5 predicts internalizing and externalizing behaviors in middle childhood (Bates, Pettit, Dodge & Ridge, 1998), adolescence, and adulthood (e.g., Moffitt et al., 2003). For this reason, early measures of a given problem behavior are noted to be the single best predictor of future problem behavior (e.g., Moffitt, 1993; Patterson, 1993). It is not surprising then that targeted intervention strategies adopt a developmental approach where predictors/behaviors assessed at a single point in time (and as early as possible) are used to predict future behavior. Unfortunately, this identification strategy, which can be decomposed into classification ability (e.g., sensitivity, specificity, and positive predictive value; Meehl & Rosen, 1955), does not necessarily equate to predictive accuracy (see Clemans et al., 2014). For instance, Moffitt et al. (1996) found that nearly half of early-onset youth did not ultimately persist in problem behavior. Bennett and colleagues (1999) found that the use of externalizing behaviors at age 5 and 6 resulted in substantial misclassification for elevated externalizing behaviors 30 months later. As such, the authors suggested that the public health impact of targeted interventions at early ages will not be optimal.

Substantial misclassification at early ages is not inconsistent with developmental frameworks that highlight a group of individuals for whom problem behavior will begin early and persist through adulthood. Notably, various developmental perspectives allow for individuals who do not have the same biological underpinnings and/or adverse early childhood environments to develop persistent displays of problem behavior as they progress through childhood and adolescence (Moffitt, 1993; Pettit & Dodge, 2003; Patterson, DeBaryshe & Ramsey, 2017). Family (e.g., inter-parental conflict or divorce; Amato, 2001; Davies & Windle, 2001), school (e.g., classroom environment or academic difficulty or failure; Roeser & Eccles, 2000), and peer factors (i.e., modeling or peer rejection; Dodge et al., 2004) may accumulate and/or interact over time to promote temporary or lasting manifestations of problem behavior. For example, a child may experience numerous protective factors (e.g., supportive family environment) resulting in limited stimuli for reactive, aggressive behavior (Vitaro et al., 2006) or anxiety/depressive symptoms. Conversely, a child may display a high level of anxiety/depressive symptoms at age six associated with school entry, but this may not reflect a continued pattern of dysfunctional behavior. Therefore, manifestations (or lack thereof) of problem behavior in childhood may be more reflective of one’s circumstances or maturational development at that time and may not represent a steadfast pattern of problem behavior. In fact, prior work documents this variation in aggressive behaviors in childhood (i.e., stable high fighters, variable high fighters, late initiating fighters and non-fighters; Haapasalo & Tremblay, 1994).

Group-based trajectory models (GBTM; Nagin, 2005) summarize trends in behavior that transcend transitory, within-individual variations and describe longitudinal patterns of behavior based on information (i.e., data) available over time. Importantly, this strategy was developed to represent general patterns of behavior across a population (e.g., Broidy et al. 2003; Janosz et al., 2008). Extant applications of GBTM confirm general patterns of aggressive behavior and depressive/anxiety symptoms in childhood and adolescence (e.g., Broidy et al., 2003; NICHD, 2004). For physical aggression, there tends to be an initial rise in early childhood (up to approximately 42 months) followed by a general decline in middle and late childhood (e.g., Tremblay et al., 2004). For anxiety or depressive symptoms, there is an increase through middle and late childhood (e.g., Leve, Kim & Pears, 2005). Notably, trajectories of these behaviors tend to vary in degree and not form because they are less influenced by temporary deviations.

From a developmental and empirical perspective, there is substantial stability in internalizing behaviors, such as anxiety/depressive symptoms, and externalizing behaviors, including aggression, but theory and prior research suggest that there is likely to be significant misclassification among those who will display persistent, elevated levels of problem behavior, which has a low base rate in the population. This is notable given that predicted patterns of problem behaviors are justification for early intervention. In this study, we address the following research question: at what age can we confidently determine whether a youth will follow a high-risk behavioral trajectory of problem behavior? Importantly, we are interested in correctly identifying individuals who will display an elevated, persistent pattern of problem behavior from childhood through early adolescence. We view this pattern of problem behavior not only as an outcome of interest, but we also note that it is a risk factor for compromised contemporaneous and future individual functioning (e.g., Campbell et al., 2010; Snyder et al., 2009).

Building upon prior work that assesses the predictive accuracy of behavioral evaluations/screenings at one point in time on subsequent problem behavior (e.g., Bennett et al., 1999; Clemans et al., 2014), we assess whether or not the accumulation of behavioral information with each passing age can improve predictive accuracy into a high-risk group who displays a persistent, elevated level of a problem behavior. To do so, we first estimate trajectories for two problem behaviors, anxiety/depressive symptoms and aggressive behavior, respectively, from middle childhood through early adolescence (measured at ages 6–13) to confirm patterns of each problem behavior and ascertain the base rate for our outcomes of interest. We then employ an extension of group-based trajectory modeling called ‘prob updates’ (Nagin et al., in press), which allows for the calculation of the probability of classification into one’s ultimate trajectory group using limited waves/information. We then evaluate the predictive accuracy of classification into the high-risk group based on limited age-spans of behavioral information with the following metrics (Meehl & Rosen, 1955): 1) the base rate of the high-risk problem behavior, 2) the true positive rate or sensitivity, 3) the true negative rate or specificity, 4) the positive predictive value (the probability that individuals will be classified as high-risk and actually display an elevated, persistent pattern of problem behavior), and 5) the negative predictive value (the probability that an individual who is classified as high-risk does not ultimately display an elevated, persistent pattern of problem behavior). We hypothesize that substantial misclassification will exist when behavioral information is limited to the earliest age (i.e., age 6) and predictive accuracy will increase with behavioral information accumulated from each subsequent age.

Method

Data

Data are drawn from the Rochester Intergenerational Study (RIGS), the intergenerational extension of the Rochester Youth Development Study (RYDS). A brief summary of these studies is provided. Detailed information regarding the two longitudinal, companion studies is presented elsewhere (Thornberry, Henry, Krohn, Lizotte, & Nadel, 2018). The RYDS data is comprised of a birth cohort of 1,000 adolescents, representative of the 7^th and 8^th grade public school population in Rochester, New York in 1988. Adolescents (Generation 2/G2) who were at a high risk for antisocial behavior were oversampled (males and adolescents residing in high crime rate areas of the city were oversampled) in line with the goals of the study. RYDS participants were followed through adulthood to 2006. RIGS began in 1999 with the identification of the oldest biological child of RYDS participants (n=370 in Year 1). New firstborns were added in each subsequent year when the child turned two years of age. Annual interviews of the RYDS participant and the other primary caregiver (if G2 was a male) have been completed each year since the start of participation in RIGS and continue until the child turns/turned 18. Children (Generation 3/G3s) complete annual interviews once they turn eight. Data collection is ongoing. Currently, there are prospective, longitudinal data for 539 parent-child dyads. Data collection procedures were approved by the University at Albany’s Institutional Review Board.

Participants

The current analysis focuses on 334 first-born children of RYDS participants for which yearly interviews were completed by the mother (either G2 or other primary caregiver) between child ages six and 13. Yearly information was needed for at least seven of the eight years (age six to 13) in order to ensure predictive ability based on earlier measures of behavioral information (see Analysis Plan).¹ The sample is relatively evenly split by sex (49% male and 51% female), and 62% of children are black, 18% are Hispanic, 12% are non-Hispanic white, and 8% are mixed race. Almost all mothers (97%) reported living with their child between child ages six and 13. Missing data comparisons revealed that across various demographic characteristics (i.e., child biological sex, race/ethnicity, living with biological mom, G2 poverty level, and community arrest rate), no significant differences between the retained analytic sample and all G3s emerged with the exception of parent age at birth.

Measures

Childhood Problem Behaviors.

Problem behaviors are assessed using a revised version of the Child Behavior Checklist (CBCL; Achenbach, 1991). The CBCL is a widely used instrument that documents child functioning, and its subscales have previously been used as screening instruments (e.g., Lochman et al., 1995; Hill et al., 2004). The original CBCL contains 118 items. Eleven subscales are included in the CBCL, including an externalizing behavior subscale, an internalizing behavior subscale, an aggressive behavior subscale, and an anxiety/depressive symptom subscale. The Rochester studies (RYDS and RIGS) include 64 of the original 118 items. Analyses revealed that the trimmed versions of the subscales maintained reliability and predictability in line with the original work of Achenbach and Edelbrock (1979; Lizotte et al., 1992). Additionally, other research has similarly used trimmed versions of the CBCL assessed through a mean score to assess risk (and predict future maladaptation; Hill et al., 2004).

In each yearly interview, G2 and the child’s other primary caregiver responded to 64 questions from the CBCL regarding the frequency of the focal child’s behavior. Recall, other primary caregivers were only interviewed if G2 was a male; therefore, CBCL information from two caregivers is not available for all children in RIGS. As such, we only use behavioral information reported by biological mothers in our analyses (over 97% of children resided with the biological mother). Given that only 64 of the original CBCL items were asked, the internalizing behavior subscale and the externalizing behavior subscale were significantly reduced in number (although both maintained reliability and predictive validity; Lizotte et al., 1992). Therefore, this analysis focused on the aggression subscale and anxiety/depressive symptom subscale because RIGS retained all items in the original subscales. The average score of the responses to 14 items (e.g., cries a lot, worries; responses included never [0], sometimes [1], and often [2]) included in CBCL anxiety/depressive symptom subscale was used to create the measure anxiety/depressive symptoms at each age. The reliability for the scale at each age from six to 13 was adequate (α =.83-.89). The average score of responses to 20 items from the aggressive behavior subscale (e.g., argues a lot, destroys his/her own things; responses included never [0], sometimes [1], and often [2]) was used to create the measure of aggressive behavior at each age. The reliability for the scale at each age was acceptable (α =.89-.91). Individual items for each scale are available in the Appendix.

Analysis Plan

Our analysis proceeds in several stages. First, we separately fit trajectories of anxiety/depressive symptoms and aggressive behaviors spanning ages six to 13 using group-based trajectory modeling (GBTM; Nagin, 2005).² We modeled each outcome using a censored normal distribution and based model selection on a combination of factors recommended by Nagin (2005), including the optimization of the Bayesian Information Criterion (BIC) and the significance of higher order parameters (Nagin, 2005). We also evaluated our trajectory solutions using additional parameters recommended by Nagin (2005).

We then employed an extension of GBTM called ‘prob updates’, which iteratively calculated the posterior probabilities of group membership using data though limited age spans in early childhood (Nagin, Jones and Elmer, working paper). This allowed us to evaluate the predictive capability of early manifestations of anxiety/depressive symptoms and aggressive behaviors, respectively, from a limited age range of behavioral information. The ‘prob updates’ extension makes a distinction between the traditional average posterior probability of group membership, which is calculated using all of the available outcome data through age T (i.e., age 13), and a redefined measure of the average posterior probability of classification that is calculated using outcome data only through age t. The former quantity is defined as

$$P\left(\text{Group = }j\text{| }{Y}_i\right)=P\left(Y_i\text{| }j\right){\widehat{\pi }}_j/{\sum }_{J}P\left(Y_i\text{| }j\right){\widehat{\pi }}_j$$

where Y_i denotes a vector of individual outcomes measured through age T and π_j denotes the (unconditional) probability of membership in group j. This quantity can be interpreted as the conditional probability of membership in trajectory group j given the full range of outcome data. Nagin et al. (in press) redefined this quantity to estimate each individual’s posterior probability of group membership using outcome data only through t. This quantity, calculated using only data observed through age t, is written as

$$P\left(\text{Group = }j\text{| }{Y}_i^t\right)=P\left(Y_i^t\text{| }j\right){\widehat{\pi }}_j/{\sum }_{J}P\left(Y_i^t\text{| }j\right){\widehat{\pi }}_j$$

Where $Y_i^t$ denotes vector of individual outcomes measured through age t. This redefined quantity addresses how well we can classify individuals in their ultimate trajectory group given measurements through age t.³

With the latter calculation, we assessed how well we can classify each individual into their ultimate behavioral trajectory. To do so, we distinguished between the actual (A) trajectory, which corresponds to group assignment using all data through age T, and the classified (C) trajectory, which assigns group membership using data through age t. Actual group classification is based on the highest posterior probability of group membership,$P\left(j|Y_i\right)$, calculated through age T, whereas the classified group is based on the highest posterior probability of group membership calculated through age t,$P\left(\mathit{j}\text{| }{Y}_i^t\right)$. We then calculated the true positive rate or sensitivity at age t and the positive predictive value (PPV) at age t. Additionally, we calculated the true negative rate or specificity, the false positive rate. and false negative rate (Meehl & Rosen, 1955). Notably, we computed the probability of classification using data through ages 6, 7, 8, 9, 10, 11 and 12, respectively, which allowed us to compare the ability to correctly classify individuals through each age t.

Results

Childhood Trajectories

Figure 1 displays the trajectory solutions based on the mean CBCL score for anxiety/depressive symptoms (panel A) and aggressive behaviors (panel B). For anxiety/depressive symptoms, the optimal solution was a three-group model (the BIC was lower than the best-fitting two-group, four-group and five-group solution), including a low group who rarely, if ever, exhibits anxiety/depressive symptoms (55.1% of the sample), a moderate group (32.6% of the sample), and a high group (12.3% of the sample). For aggressive behaviors, the optimal solution also yielded three groups (through optimization of the BIC in comparison to the two-group, four-group and five-group solutions). Again, we found a low group (43.4% of the sample), a moderate group (40.4% of the sample), and a high group (16.2% of the sample). Each solution passes all four model adequacy checks recommended by Nagin (2005; see Appendix). In addition, the posterior probability of classification using data through age 13 for the anxiety/depressive symptom trajectory groups and the aggressive behavior trajectory groups exceeds .90, suggesting a very low degree of classification error (Roeder et al. 1999). Thus, we are highly confident regarding ultimate trajectory classification.

Figure 1.

Childhood Problem Behavior Trajectories

Panel A: Anxiety/Depressive Behavior Trajectories Panel B: Aggressive Behavior Trajectories

Prediction and Classification

We now consider how early we can confidently assign a child into his or her ultimate trajectory of anxiety/depressive symptoms and aggressive behavior. Individuals are hard-classified into a behavioral trajectory group based upon their highest posterior probability of group membership through age 13, referred to as one’s actual (A) trajectory. We also have information about one’s classified (C) trajectory at each age t, which is one’s assigned group using posterior probabilities calculated with data through age t. We focus specifically on identifying youth who are ultimately classified into the high (H) trajectory (using data through age 13), as opposed to distinguishing between the low and moderate groups for anxiety/depressive symptoms and aggressive behaviors, as the high group is most likely to experience problematic development and subsequent maladaptation.

Figure 2 presents a cross-tabulation of the classified group versus the actual group and defines several classification quantities of interest. First, the true positive rate or sensitivity, P(CH | AH), where CH is classified high status, and AH is actual high status, describes the proportion of individuals are correctly identified in the high trajectory. The complement of this quantity is the false negative rate, P(CH’ | AH), which represents the proportion of individuals who ultimately follow the high trajectory but are not classified in the high trajectory through age t. This quantity describes the proportion of individuals who would not be identified as high risk through screening even though they ultimately would display an elevated, persistent level of problem behavior. The false positive rate, P(CH | AH’), represents the proportion of those who would be classified in the high trajectory but ultimately would not be high-risk based on their ultimate pattern of behavior (through age 13). Finally, the positive predictive value is proportion of those classified as high who are in fact high.

Figure 2.

Predicted vs. Actual Classification

Note. “Other” includes low and moderate trajectory groups; Accuracy = (True Positives + True Negatives) /n; Positive predicted value = (True Positive + False Positive)/n; Negative predictive value = (True Negatives + False Negatives)/n

We now consider classification ability for the high anxiety/depressive symptom trajectory (see Table 1). The base rate for this condition is the group mixture probability, 12.3%. The true positive rate or sensitivity calculated using information through age 6 only, which would amount to a single summary of observation upon school entry, is 44%. In other words, less than half of those who ultimately follow the high anxiety/depressive symptom trajectory would be classified as high at this age. Sensitivity increases with age (65% at age 7), and it crosses the generally accepted threshold of predictive ability (.70) using the sum of information through age 8 (.77). It then increases dramatically to over 90% using information through age 9 or older. We also consider the false negative rate. It remains at a nontrivial level using information through age 8 (56% at age 6, 35% at age 7, and 24% at age 8), and it declines dramatically using information through age 9 (3%), remaining low thereafter. The positive predictive value, which describes the proportion of those classified as high who are in fact high, increases from 71% at age 6 to 85% at age 9 to 100% by age 12. Finally, the false positive rate, which indicates how likely it is that children who ultimately will not display persistently high levels of anxiety/depressive symptoms are classified as such, is quite low. Even at early ages, it never exceeds 4.2%. Taken together, these results suggest nontrivial misclassification at early ages, particularly in the form of false negatives, but by age 9, it is possible to classify children into the high anxiety/depressive symptoms pattern of behavior with more confidence.

Table 1.

Error Rates for Classification into High Trajectory Groups

Panel A. Anxiety/Depressive Symptom Classification
	Age 6	Age 7	Age 8	Age 9	Age 10	Age 11	Age 12
True Positives (Sensitivity)	0.441	0.647	0.765	0.971	0.941	0.971	1.000
False Positives	0.024	0.032	0.029	0.025	0.025	0.004	0.000
True Negatives (Specificity)	0.976	0.968	0.971	0.975	0.975	0.996	1.000
False Negatives	0.559	0.353	0.235	0.029	0.059	0.029	0.000
Positive Predictive Value	0.714	0.733	0.788	0.846	0.842	0.971	1.000
Negative Predictive Value	0.928	0.953	0.967	0.996	0.992	0.996	1.000
Panel B. Aggressive Behavior Classification
	Age 6	Age 7	Age 8	Age 9	Age 10	Age 11	Age 12
True Positives (Sensitivity)	0.558	0.674	0.810	0.786	0.833	0.881	0.905
False Positive	0.041	0.042	0.038	0.042	0.034	0.026	0.004
True Negatives (Specificity)	0.959	0.958	0.962	0.958	0.966	0.974	0.996
False Negative	0.442	0.326	0.190	0.214	0.167	0.119	0.095
Positive Predictive Value	0.706	0.744	0.791	0.767	0.814	0.860	0.974
Negative Predictive Value	0.924	0.943	0.966	0.962	0.970	0.978	0.983

With respect to aggressive behavior, the base rate is the mixture probability of 16.2%. It is a bit higher than the expected base rate in a normative population, but it is not unexpected given our sample is comprised of children from an at-risk population (Conduct Problems Prevention Research Group, 2004). The true positive rate or sensitivity is little better than chance at age 6 (56%), and it is still less than the accepted threshold (.70) using data through age 7 (67%). The sensitivity increases to 81% at age 8 and it only exceeds 90% using information through age 12. The false negative rate is also very high at age 6 (44%) and remains over 20% using information through age 9. Using information through age 10, the false negative rate falls below 20%, and it is 10% using information through age 12. The positive predictive value increases from 71% at age 6 to 81% using information through age 10. It is 97% using information through age 12. Lastly, the false positive rate for aggressive behaviors is very low, ranging from 3.2% to nearly 0% using information through age 11. In sum, the results suggest that significant misclassification into the highest pattern of aggressive behavior using information from earlier ages (through age 8 or 9), particularly in the form of false negatives.

Discussion

Designing, implementing, and improving the effectiveness of prevention programs, which seek to reduce the prevalence and incidence of childhood and adolescent problem behaviors, is a sound investment, particularly if these programs promote socio-emotional competence and health at a young age (American Psychological Association’s Task Force on Prevention: Promoting Strength, Resilience & Health in Young People, 1998 as referenced in Payton et al., 2008). Effective prevention programming should be informed by extant theory and target empirically verified age-appropriate risk factors. More so, effective prevention programming should be tailored to the individual and applied among appropriate groups identified by age and risk. This research examines the presentation of a risk factor across age to inform treatment selection models related to prevention programming. Our goal was to analyze how different longitudinal spans of behavioral information for two childhood problem behaviors, anxiety/depressive symptoms and aggressive behavior, affect the ability to accurately classify individuals into groups at the highest-risk for near- and long-term maladaptation. The results of our efforts lead to multiple considerations for extant theory and policy.

First, we add to the abundance of research to date documenting between- and within-individual heterogeneity in anxiety/depressive symptoms and aggressive behaviors in childhood and early adolescence (spanning ages 6–13). To be sure, we are not the first study to demonstrate these patterns of behavior (e.g., Broidy et al., 2003; Hussong et al., 2011). We only confirm that varying patterns of elevated anxiety/depressive symptoms and aggressive behaviors, relative to one’s age mates, emerge among school-aged youth. Even when taking into consideration individual noise or variation in behavior, these patterns of anxiety/depressive symptoms and aggressive behaviors, only differentiable by degree, tend to be relatively stable from childhood through early adolescence. As such, patterns of anxiety/depressive symptoms and aggressive behaviors in childhood that manifest at an earlier age and endure over time should continue to be viewed as an indicator of maladaptation on its own and an informative risk factor for near-term and long-term maladjustment.

A key argument that we seek to advance is that limited age spans of observed/reported behaviors fail to suitably implicate all individuals who will display an elevated level of problem behavior over time. Our assessment of classification ability into the high group (i.e., elevated, persistent display of problem behavior) for both anxiety/depressive symptoms and aggressive behaviors based on limited spans of behavioral information (i.e., observations from only age 6, ages 6 and 7, or even ages 6 to 8) suggests a considerable amount of misclassification. The noise or variation in behavior at early ages precludes the ability to correctly identify a long-term, elevated level of problem behavior for a non-trivial percentage of children who ultimately display elevated, persistent involvement in each problem behavior. Moreover, the results demonstrate that misclassification is driven primarily by false negatives, as a non-trivial proportion of youth who ultimately display persistent, elevated levels of problem behaviors are not correctly identified at early ages. As a result, if intervention efforts solely focus on the identification of “early-onset” individuals using limited information on anxiety/depressive symptoms or aggressive behaviors at younger ages (through age 8), they will likely fail to identify a key subset of individuals who would also benefit from targeted prevention efforts. This is noteworthy since the presumption is that screening is most valuable at earlier ages because early and enduring prevention efforts are most successful.

In line with extant theory, some youth display a level of problem behavior that is maladaptive, likely invoking a clinical diagnosis or sub-clinical significance of symptoms at an early age. For these youth, correct classification into the high group at an early age is obvious, as is the need for screening and targeted intervention as soon as possible. Our findings support this notion as the false-positive rate for correct classification in the high group for each problem behavior is extremely low, even at age 6. Thus, the application of screening and targeted intervention at early ages for those with the most extreme levels of problem behavior is logical. Still, typological, developmental theories (e.g., Moffitt, 1993; Dodge & Pettit, 2003) allow for protective factors such as supportive family environments or schools to prevent the manifestation of problem behaviors at earlier ages. In addition, some youth encounter snares whereby individual and life circumstances promote a later onset (although still in childhood) of enduring, elevated levels of problem behavior. Both sets of youth are not as easily identifiable at early ages, and the results of this analysis suggest that they outnumber the early onset individuals. Therefore, screenings should endure through childhood in order to continue to identify children in need of targeted intervention programs (at least through age 9). While detailed screenings across age may be burdensome or costly, an important topic for further research is to determine if the ultimate benefits of improved classification and targeted provision of services outweigh these costs. In addition, other more feasible evaluation tools such as assessments from teachers/administrators at the end of the year (which are often used to inform class placement in the subsequent year) or parent/teacher behavioral referrals for discipline could be used to supplement initial screenings (which have a high false-negative rate) with additional behavioral information (i.e., onset of problem behavior and display of behavior over subsequent years) to identify youth in need for more formal screening and potential targeted intervention at later ages. As such, we stress that while early identification for targeted prevention strategies is needed to prevent a spiral of problem behavior at the earliest ages (i.e., at the beginning of school) for some, the prediction ability for all high-risk individuals at early ages is limited. Thus, ongoing evaluation should be used to inform subsequent formal screening and targeted interventions in order to be most effective if continuous formal screenings are too costly/burdensome.

These findings further inform prevention science as they can be applied to cost-benefit analyses, which consider the cost of false negatives and false positives in the context of specific types of interventions weighed against the benefits of the program. We argue that cost-benefit analyses are surely needed, particularly for screening and prevention programming implemented at early ages as misclassification is certain. For some targeted intervention programs, the costs of a false positive might be considerably high both in terms of monetary costs, as well as unanticipated costs such as labeling and stigmatization. Moreover, missed opportunities for targeted intervention may be costly to the individual and society. As such, some programs may see greater benefits (and limited costs) when applying a continuum of intervention, beginning with universal strategies at younger ages and more targeted strategies at older ages (i.e., Incredible Years, Webster-Stratton et al., 2004). Our findings should be used to inform a more thorough cost-benefit analyses of screening and intervention, as it provides potentially useful information on the base rate of misclassification of targeted youth as well as the positive predictive value, which illuminates balanced cost efficiency.

It is important to reconcile these findings with other research suggesting that individual assessments in kindergarten can be used for prevention efforts (see Masse & Tremblay, 1997). We highlight that these assessments were personality-based, and while personality is relevant to patterns of behavior, personality traits are also more difficult for practitioners to identify without standardized measurement tools. Moreover, personality information is not commonly included in screening, including school-based prevention programs. Conversely, normative behaviors are commonly used in screening instrumentation (i.e., Bennet & Offord, 2001; Clemans et al., 2014; Hill et al., 2004), as behavioral measures are easier for family members and practitioners, such as doctors/nurses and/or school counselors, to monitor over time/grade level. Moreover, there may be a misalignment of personality and behavioral manifestations of problem behavior as protective factors in one’s life (e.g., supportive family environments) may impede the manifestation of problem behaviors at earlier ages (Moffitt, 1993). Nevertheless, if practitioners have the ability to assess personality traits, such as psychopathy or negative emotionality, it may be possible to more accurately identify youth who are most likely to display elevated, persistent levels of aggression or anxiety/depressive symptoms at earlier ages. However, given the need for clinicians to administer such tests, the wide-scale applicability of these methods may be limited.

While the results of this study are informative, some limitations are worthy of note. We focused on two childhood problem behaviors that are indicative of maladaptation. The focus on a limited number of problem behaviors was beneficial to the purpose of this research endeavor, but it is important to recognize that maladaptation can result from elevated, persistent involvement in various problem behaviors (i.e., rule-breaking or defiance, anxiety symptoms alone). However, these behaviors often cluster together (e.g., aggressive behavior and rule-breaking behavior); therefore, examining the predictive accuracy of one long-term pattern of elevated, persistent problem behavior likely speaks to the early classification ability for other patterns of problem behavior as well (results were similar for anxiety/depressive symptoms and aggressive behaviors). Second, our sample size is small. We recommend replication of this analysis with larger samples to confirm our results. Relatedly, we were unable to include information on the targeted problem behaviors from earlier ages due to further reduced sample sizes at younger ages (prior to age 6). However, those who identify youth for targeted prevention programming are most often located in schools or examine school age youth (i.e., school counselors and psychologists) and information prior to kindergarten (age 5 or 6) is largely unavailable unless parents/guardians are willing to share this information. We also note that the sample for this analysis is predominantly African-American and consists of the offspring from a high-risk sample. Replication among a more representative sample is worthwhile, particularly if base rates for elevated, problem behaviors are lower (which may further speak to limited classification ability at younger ages). Finally, prior work documents that base rates for problem behavior vary across sex. In line with extant theory, this work viewed biological sex as a risk factor for level and overall pattern of behavior. If screening tools and targeted interventions set the criteria for intervention to be unique across sex (in terms of degree or duration), then it would be informative for subsequent research to replicate these analyses separately across sex.

Our investigation emphasizes that relying on single observations of maladaptive development in early childhood may result in substantial misclassification for long-term, elevated levels of problem behavior. For the majority of children most in need of targeted interventions, accurate classification into the high group or high-risk for future problem behavior increases with the accumulation of behavioral information through the later years of elementary school. As such, those that seek to provide prevention or intervention programming should stay steadfast in the goal of continued monitoring and assessment through elementary school when problem behavior patterns are sufficiently realized.