Introduction

Problem behaviors, such as oppositionality, hyperactivity/impulsivity, rule-breaking, aggression, and violence, are among the most burdensome and prevalent forms of psychopathology faced by children and adolescents [1–3]. These outwardly directed behaviors load onto an “externalizing” factor of psychopathology [4], which includes clinical disorders, such as attention deficit/hyperactivity disorder (ADHD), oppositional defiant disorder, conduct disorder, and substance use dependence [4–6]. Externalizing problems are a major public health concern that place youth at risk for poor outcomes into adulthood, including academic underachievement, interpersonal problems, employment difficulties, incarceration, long-term substance dependence, and persistent antisocial behavior [7–9].

An underexplored question in the field is whether low rates of externalizing problems are driven by the absence of something negative (e.g., lack of neighborhood deprivation), the presence of something positive (e.g., neighborhood affluence), or both [10–12]. This has important implications for the field’s conceptual understanding of factors associated with reductions in antisocial behavior, and for informing intervention. If, for example, the absence of neighborhood poverty predicted low externalizing problems, but the presence of neighborhood affluence did not, it would be clear that interventions should focus on shifting neighborhoods from low to middle income, but not necessarily on making neighborhoods wealthy.

A trichotomization approach is one way to test this question. Trichotomization compares the middle half of a variable’s distribution to the upper and lower quartiles to test for linear or non-linear effects [11, 12]. A linear effect is indicated when children falling in the mid-range have a fewer externalizing problems than those in the lower quartile, and those in the upper quartile have a fewer externalizing problems than those in the mid-range. Alternatively, there may be non-linear effects if, for example, only the upper quartile, but not the lower quartile, differs significantly from the mid-range on externalizing problems. Thus, it may be that the absence of risk (i.e., the mid-range compared with the risky quartile), the presence of something positive (i.e., the positive quartile compared with the mid-range), or both are associated with decreases in externalizing problems.

A separate question is whether variables have promotive or protective effects on externalizing psychopathology [13–15]. Promotive factors (i.e., direct protective factors [16]) are main effects associated with decreases in problematic outcomes. Protective factors (i.e., buffering factors [16]) buffer youth from externalizing problems in the face of risk. To distinguish whether a variable has a promotive or protective effect, studies must test an interaction term between a positive factor and a risk factor. A promotive effect would be indicated if there was a significant main effect, but not a significant interaction, such that rates of externalizing problems decrease as the positive factor increases (see Fig. 1 for an example plot). If the interaction is significant and suggests a protective effect, then the high risk group will experience reductions in externalizing problems when exposed to high, compared with low, levels of the positive factor, but externalizing problems will be low and unassociated with the positive factor in the low-risk group (see Fig. 2 for an example plot). Although there have been many reviews of risk factors for externalizing psychopathology among children and adolescents (e.g., [17–22]), only a few have included [23, 24] or focused specifically [16, 25–27] on promotive and protective factors. In addition, extant reviews have focused on identifying variables associated with decreases in externalizing problems, but there has been a lack of synthesis regarding which variables tend to have promotive versus protective effects.

Fig. 1
figure 1

Example of a non-significant interaction demonstrating a promotive (main) effect

Full size image
Fig. 2
figure 2

Example of an interaction demonstrating a protective (buffering) effect

Full size image

The current systematic review has two aims. The first is to review findings from longitudinal studies that employed trichotomization to test for whether the absence of risk and/or the presence of something positive relates to decreased likelihood of externalizing behaviors. The second aim is to review findings from studies that tested interactions between a positive factor and a risk factor to identify whether the positive variable has a promotive effect, or a protective effect that mitigates the association between a risk factor and externalizing problems.

Method

Searches were conducted in PubMed and PsycInfo to identify articles written in English and published in a peer-reviewed journal between January 2005 and December 2015. Search terms were: Risk and (Protective or Promotive) and (Externaliz* or Antisocial or Violen* or Aggress* or Problem Behavior or Delinquen* or Attention Deficit Hyperactivity Disorder or Conduct Disorder or Oppositional Defiant Disorder or Substance Use Dependen*) and (Child or Adolescen*). The initial search resulted in 899 articles from PubMed and 923 articles from PsycInfo. Once duplicates were removed, 1467 articles were screened.

We screened articles in a three-stage process. First, we screened titles and abstracts to exclude studies of outcomes other than externalizing problems as defined by the search terms (e.g., gambling, problematic video game use, sexual risk taking, teen parenthood, internalizing problems; n = 517); child welfare services delivery and utilization (n = 45); and populations with limited generalizability (e.g., autism spectrum disorders, child soldiers; n = 77). Qualitative, measurement development, and intervention development studies (n = 106) and other miscellaneous studies (n = 13) were also excluded.

Second, we read papers and excluded studies that were not empirical (e.g., introductions to special issues; n = 89), sampled adults (n = 103), did not include protective/promotive factors as predictors (n = 74), and did not include temporal ordering in the measurement of protective/promotive factors and externalizing outcomes (n = 222). This process yielded 203 empirical articles. Reference lists were reviewed for additional papers that met inclusion/exclusion criteria (n = 31). Third, we screened the final pool of 234 studies and identified those that used trichotomization (n = 7) to review for the first aim, and those that tested interaction effects (n = 53) to address the second aim.

Results

The absence of risk versus the presence of positive factors

In the following section, we review evidence from studies that trichotomized predictor variables to examine whether it is the absence of risk (e.g., not having low IQ) or the presence of something positive (e.g., having above-average IQ) that is associated with decreased rates of externalizing problems. We review variables that are continuously distributed (e.g., low to high IQ) and, therefore, could be trichotomized to test for linear or non-linear effects. We do not review dichotomous variables, because the interpretation of the results is the same regardless of which value is the reference group. It is equally true, for example, that being a girl and not being a boy are associated with a fewer externalizing problems. We present evidence for linear versus non-linear effects of individual-, family, peer-, school-, and neighborhood-level variables. See Table 1 for a summary of findings.

Table 1 Summary of findings
Full size table

Individual-level variables

Some individual-level variables have demonstrated linear effects, such that youth with scores in the positive end of the distribution have lower delinquency compared with youth with scores in the mid-range. Likewise, youth with scores in the mid-range have lower delinquency compared with those with scores in the risky end. One study reported this pattern for IQ, such that children with above-average IQ had significantly lower risk of externalizing problems than children with IQ in the average range, and children in the average range had significantly lower risk of externalizing problems than children whose IQ fell in the below-average range [10]. Similarly, low, compared with average, attention problems/ADHD is associated with lower rates of violence [28, 29], and average, compared with high, attention problems/ADHD is associated with less violence [28]. Of note, only one of the studies [28] on attention problems/ADHD reported both comparisons, and the other [29] did not report differences between mid-range and high attention problems/ADHD.

Regarding children’s temperament, van der Laan and colleagues [30] found support for linear effects of surgency (e.g., activity level, behavioral inhibition, and impulsivity) and effortful control (e.g., ability to shift attention when needed and suppress inappropriate responses). Low surgency and high effortful control were associated with lower delinquency compared with average scores on these measures, and average scores were associated with lower delinquency compared with high surgency and low effortful control. Taken together, these linear effects suggest that the absence of risk (i.e., not having low IQ, high ADHD symptoms/attention problems, high surgency, or low effortful control) and the presence of the positive side of the variables (i.e., having high IQ, low ADHD symptoms/attention problems, low surgency, and high effortful control) are associated with a fewer externalizing problems.

Several variables demonstrated non-linear effects in the direction that suggests that the absence of risk, rather than the presence of something positive, drives reductions in externalizing problems. In other words, youth with scores in the mid-range have lower delinquency compared with those with scores in the risky end, but youth with scores in the positive end do not have lower delinquency compared with those with scores in the mid-range. This pattern was supported for delayed visual memory, such that the mid-range, compared with low, scores were associated with lower delinquency, but mid-range and high scores did not differ significantly on delinquency [10]. Similarly, compared with youth with above-average interpersonal callousness, youth with mid-range scores have lower rates of violence, but youth with below-average interpersonal callousness are not less violent than those in the mid-range [31]. In sum, not having poor delayed verbal memory and high interpersonal callousness decreased rates of externalizing problems.

On the other hand, some variables demonstrated non-linear effects suggesting that the presence of the positive end of the variable is associated with lower delinquency compared with the average range, but youth with average scores do not have lower delinquency compared with those with scores reflecting risk. High, compared with average, scores on sustained attention and delayed verbal memory are associated with lower delinquency, but mid-range and low scores on these measures do not differ significantly [10]. Similarly, youth with high, compared with mid-range, aspirations for higher education are less likely to perpetrate violence, but those with mid-range aspirations are not less violent than those with low aspirations [32]. The same pattern has been reported for shyness, with high shyness representing the positive end of the variable, and low shyness representing the risky end [30]. In addition, youth with low depressive symptoms are less likely to perpetrate violence than those in the mid-range, but those in the mid-range are not less violent than those with high depressive symptoms [31, 33]. Youth with higher ability to refuse engaging in antisocial behavior [28] and who espouse more negative attitudes toward delinquency [29, 31] engage in less delinquency compared with youth with mid-range scores, but youth with mid-range scores do not differ on rates of delinquency compared with youth with poor refusal skills or pro-delinquency beliefs. Findings suggest that it is the presence of the positive ends of these variables rather than the absence of risk that drives reductions in antisocial behavior.

It is unclear whether academic achievement has a linear or non-linear effect on antisocial behavior. Some studies reported non-linear effects, such that youth with above-average academic achievement had significantly lower risk for violence than those with average achievement, but those with average achievement did not differ from those with below-average achievement [28, 31]. Other studies identified a linear pattern, such that above-average versus average, and average versus below-average, academic achievement was associated with lower violence [30, 32, 33].

Family factors

The majority of family level variables tested in prior studies demonstrated linear effects. Good family management strategies (e.g., clear contingencies) [28] and healthy family functioning [30] are associated with less violence compared with the mid-range of these variables, and the mid-range is associated with less violence compared with poor strategies (e.g., unclear rules) and family dysfunction. Similarly, lower, compared with average, and average compared with high, parental stress is associated with decreased risk for delinquency [30]. These findings suggest that children living in families that have good management strategies, healthy functioning, and low parental stress, and do not have poor management strategies, dysfunction, and high parental stress, are at decreased risk for externalizing problems. In contrast, one study reported a non-linear effect; low, compared with mid-range, levels of parental overprotection were associated with less delinquency, but mid-range and high levels of overprotection did not differ significantly on delinquency [30].

Peer factors

One study found that youth whose ability to get along with peers fell in the mid-range engaged in less violence than youth who had below-average abilities to get along with peers [31]. However, youth whose ability to get along with peers was above-average did not engage in less violence than those with average abilities [31]. These findings suggest that the ability to get along with peers has a non-linear effect, such that the presence (versus the absence) of this ability lowers externalizing outcomes.

In the case of affiliating with delinquent peers, studies have reported linear and non-linear effects. Two studies found a linear effect, such that children with a few, compared with mid-range, and mid-range compared with many, delinquent peer affiliations had lower rates of antisocial behavior [31, 33]. A third study reported non-linear effects, such that youth with mid-range delinquent affiliations had lower rates of violence in adolescence compared with those with many delinquent affiliations, but youth with a few delinquent affiliations did not differ significantly from those in the mid-range [32]. However, the opposite non-linear effect emerged when predicting violence in young adulthood: youth with a few delinquent peer affiliations had lower rates of violence compared with the mid-range, but youth with mid-range and many delinquent affiliations did not differ significantly on violence. Compared with youth with peers falling in the mid-range of prosocial behavior, youth who affiliated with peers with above-average prosocial behavior was less likely to perpetrate violence [28]. The study did not report comparisons between youth with peers in the mid-range and with above-average prosocial behavior, making it unclear whether this represents a linear or non-linear effect.

School factors

Only non-linear effects have been reported for school variables. Youth with more positive attitudes toward school do not differ on violence from those in the mid-range, but those with mid-range school attitudes engage in less violence than those with negative attitudes [31, 33]. On the other hand, high, compared with average, school attachment is associated with lower delinquency, but children with low and mid-range school attachment do not differ on delinquency [28]. Thus, having positive school attachment, and not having negative attitudes/dissatisfaction, is associated with lower levels of externalizing problems.

Neighborhood factors

Perceived availability and exposure to marijuana in the neighborhood has a linear effect on violence, such that low compared with mid-range, and mid-range compared with high, levels of perceived availability and exposure to marijuana are associated with less violence [28]. Loeber and colleagues [10] found a non-linear effect of housing quality, such that the presence of good housing quality compared with the mid-range, but not mid-range compared with poor housing quality, was associated with low delinquency. There is mixed evidence on the type of non-linear effect that community crime exerts on violence. One study found that youth living in neighborhoods with average levels of crime/poverty had lower rates of violence than those living in neighborhoods with high crime/poverty, but youth exposed to low versus average crime/poverty did not differ significantly on violence [31]. Another study found the opposite non-linear pattern: youth living in low crime neighborhoods were less likely to engage in delinquent behavior compared with those in neighborhoods with average levels of crime, but delinquent behavior did not differ between youth living in average versus high crime neighborhoods [10]. Thus, it is not clear whether reductions in delinquency are more strongly associated with the absence of high crime rates or the presence of very low crime rates.

Promotive versus protective effects

The following section reviews studies that tested an interaction term to determine if a variable involves a promotive main effect or a protective buffering effect. We examined studies that tested for (1) individual characteristics that moderate effects of risky environments, (2) environmental characteristics that moderate effects of individual-level risk, and (3) environmental characteristics that moderate effects of risky environments. Table 1 shows findings by variable, and we provide a summary below. This review focuses on behavioral and cognitive individual difference variables as opposed to genetic or physiological ones. A recent review discussed biological protective factors [27]. We also did not review moderators of intervention effects, because we focused on the question of promotive versus protective effects within risky contexts.

Individual-level promotive factors and moderators of risky environments

Several studies tested whether individual strengths buffer risk associated with family environment, such as harsh discipline and hostility. Positive individual-level characteristics of higher cognitive ability [34], lower levels of difficult temperament [35], attending or having completed high school (or GED) [36], and low endorsement of aggressive norms [37] had protective effects within risky family contexts. One study found that easy infant temperament exerted a promotive effect on preschoolers’ problem behaviors, but did not significantly buffer against family level risk [38].

Studies have also identified individual-level moderators of risk associated with extra-familial contexts. The strength of the association between affiliating with delinquent peers and engaging in delinquent behavior is attenuated by spending more time in prosocial activities (e.g., sports team and musical activity) [39], low endorsement of aggressive norms [37], and high self-esteem [36]. One study found that higher, compared with lower, levels of self-confidence decreased externalizing problems among youth living in poverty [40]. Finally, one study identified a promotive effect of self-perceived scholastic competence, which did not buffer risk associated with peer rejection [41].

Some interaction effects indicated that decreases in externalizing problems occurred in low, but not high, risk environments. One study found that girls with social anxiety were at decreased risk for substance use if they reported that their peers had low, compared with high, rates of substance use [42]. This pattern does not support a buffering hypothesis, but rather suggests that an individual characteristic (social anxiety) may allow girls to benefit more from a low risk environment [43]. In addition, the degree to which individual-level characteristics can protect youth from environmental risks may depend on the number of stressors present in the environment. One study found that children’s individual strengths buffered risk for externalizing problems at low, but not high, numbers of family and neighborhood stressors [44].

Overall, the interaction effects identified in this section support a buffering hypothesis, such that positive individual characteristics tend to attenuate the relation between risky environments and externalizing problems. The only exceptions were for easy infant temperament and high scholastic competence, which had promotive effects associated with decreases in externalizing problems regardless of risk.

Environmental promotive factors and moderators of individual-level risk

Positive aspects of the family environment can have buffering effects for children with individual-level risk factors for externalizing problems. Among children born in Mexico, the relation between high prenatal lead exposure and ADHD is attenuated by higher maternal self-esteem [45]. Higher levels of positive parenting (e.g., monitoring, involvement, and warmth) buffer children against risk associated with prenatal cocaine exposure [46], difficult temperament [47–49], and history of externalizing problems [50–52]. Among children who have been physically abused, positive parenting is protective for children with low self-regulation [53]. In contrast, some studies found support for promotive effects of family factors rather than moderation of individual-level risk. One study found that parental disapproval of antisocial behavior had promotive effects, but did not moderate risk associated with a history of abuse [54]. Marsiglia et al. [55] found that among children from Mexican immigrant families, high family cohesion and familism had promotive effects on externalizing outcomes, but did not buffer against risk posed by low acculturation. Thus, although many positive aspects of the family environment buffer individual-level risks, some have promotive effects instead.

Positive aspects of the peer, school, and neighborhood environments can also promote positive outcomes or protect children with individual-level risk factors. For example, for girls, being liked by peers attenuates the relation between difficult temperament and hyperactivity [56]. Another study found that high commitment to school was associated with a fewer externalizing problems, but did not buffer risk associated with prior maltreatment [54]. Neighborhoods with higher social cohesion provide protective effects for youth with histories of externalizing problems [57]. In sum, some extra-familial environmental characteristics have promotive effects, and others have protective effects for youth with risky individual characteristics.

Environmental promotive factors and moderators of environmental risk

Children’s environments comprise a mix of risk and positive factors that can interact to reduce externalizing problems. High maternal empathy [58] and sensitivity [59] have protective effects in the face of household stress and conflict. High parent–child relationship quality buffers children from risk associated with a parent’s history of delinquency [60]. Grandmother involvement is protective for children exposed to harsh maternal discipline [61]. The relation between harsh discipline and externalizing problems is also attenuated by positive parenting [62–64]. There is some evidence that this effect may be culture-specific. Lansford and colleagues collected data in China, Colombia, Jordan, Kenya, Philippines, Thailand, Italy, and the US, and found that maternal warmth buffered the effects of corporal punishment on aggression only among African American families in the US, and had promotive effects in Jordan and Kenya [64]. However, an additional study that used data from a large, US birth-cohort study that oversampled African American and Hispanic families found that maternal warmth did not have a promotive or protective effect on later aggression in the face of spanking [65]. This study did not test whether these effects differed by race or ethnicity, so it is unclear whether buffering effects held within the African American subgroup. Although there is some evidence for promotive effects, the majority these findings support the hypothesis that positive family factors can buffer risk associated with negative aspects of the family environment.

Several studies tested whether positive aspects of the family environment buffer youth against risks outside of the home. Positive parenting attenuates the relation between externalizing problems and attending a school with higher norms for aggression [66], living in a more disadvantaged neighborhood [67, 68], and exposure to community violence [69–72] or cumulative risk [73]. Perceived parental support [74], positive parent-adolescent interaction style [75], and maternal socialization of coping [76] buffer risk associated with life stress. Thus, it appears that positive family characteristics provide protective effects for youth exposed to stressful environments and negative life events. On the other hand, a handful of studies found support for promotive, rather than protective, effects of positive family factors in the face of peer substance use [77], neighborhood disadvantage [78], and community violence [79, 80].

Positive aspects of extra-familial environments can also promote positive outcomes or protect youth against risk factors within or outside of their families. Positive peer relationships buffer against externalizing problems for youth exposed to community violence [72]. Higher levels of school connectedness promote decreases in violence, but do not buffer against exposure to community violence [70]. Positive neighborhood factors, such as social cohesion and collective efficacy, buffer children against risk associated with neglect [81], caregiver depression (among neglected children) [82], and neighborhood deprivation [83].

Stoddard and colleagues [84] included individual, family, and peer factors in measures of cumulative positive and risk factors. They found that more cumulative positive factors buffered against violence for youth exposed to high, compared with low, cumulative risk. The cumulative positive score was not associated with violence among youth exposed to lower cumulative risks. Thus, it appears that overall, positive factors are protective in contexts of high, compared with low, risk.

In sum, our review suggests that positive aspects of the family environment can buffer risk associated with domestic violence, parental history of antisocial behavior, parental and life stress, and high norms for aggression at school. Findings were mixed regarding whether positive parenting practices have protective, promotive, or no effect against harsh discipline, community violence, and neighborhood disadvantage. A few studies examined whether positive peer, school, and neighborhood factors moderated associations between risk and externalizing outcomes. Those that did found support for buffering effects, expect for school connectedness, which had promotive effects.

Discussion

The current review provides a novel synthesis of the literature on factors that decrease likelihood of externalizing problems in childhood and adolescence. We examined whether it is the absence of risk or the presence of something positive that drives reductions in externalizing behaviors. We also reviewed findings on whether certain variables tend to have promotive main effects or protective buffering effects in various risky contexts.

We reviewed studies that used a trichotomization approach to determine if reductions in externalizing problems involve either the absence of risk or the presence of something positive (i.e., non-linear effects), or both (i.e., linear effects). IQ, temperament, family management, family functioning, and parental stress had linear effects, such that lower delinquency was associated with the mid-range compared with the risky end of the variable, and the positive end compared with the mid-range. Fewer externalizing problems were associated with not having high attention problems/ADHD, poor delayed visual memory, high depression, high interpersonal callousness, poor skills to refuse engagement in antisocial behavior, high parental overprotection, affiliating with many antisocial peers, low ability to get along with peers, negative attitude/dissatisfaction with school, high perceived availability, and exposure to marijuana in the neighborhood. In contrast, the presence of the following (rather than the absence of the risky end of these variables) was associated with fewer externalizing problems: high academic aspirations, high sustained attention, high delayed verbal memory, high shyness, negative attitudes toward delinquency, affiliating with prosocial peers, high attachment to school, and good housing quality. There were mixed findings regarding whether academic achievement had linear or non-linear effects on antisocial behavior, and whether the absence of high crime or the presence of very low crime decreased externalizing problems. Future work should examine whether effects depend on contextual variables.

These findings provide guidance for whether interventions should aim to reduce risk, foster positive factors, or both to improve children’s outcomes. For variables demonstrating non-linear effects, resources should be allocated to move children out of the risky quartile and into the mid-range, or from the mid-range into the positive quartile, to reduce risk for externalizing problems. For instance, only youth with high, compared with average school attachment, but not average compared with low school attachment, were at decreased risk for externalizing problems. This suggests that interventions should not stop once children are no longer disengaged or antagonistic toward school, but rather aim to foster positive school attachment. For variables with linear effects, interventions that result in even incremental movements in the dependent variable will reduce risk for externalizing problems.

Our review of promotive versus protective effects revealed some consistent findings. Positive individual characteristics tended to have protective effects, which suggest that they are particularly important for youth in risky environments. In addition, positive parenting practices buffered youth from individual-level (e.g., difficult temperament), family level (e.g., harsh discipline), and neighborhood-level (e.g., neighborhood disadvantage) risks. There were mixed findings regarding whether positive parenting practices buffered risk associated with community violence. A few studies examined interactions between peer, school, and neighborhood characteristics on children’s externalizing behaviors, and findings were mixed regarding whether these characteristics tended to have promotive versus protective effects. Variables demonstrating protective effects should be considered particularly important targets for intervention for children exposed to risk, and variables with promotive effects should be targeted within universal prevention strategies.

The current review has several limitations. First, a relatively few studies used a trichotomization approach. Second, although trichotomization is a particularly appropriate technique to address our research question, it has drawbacks, such as utilizing relatively arbitrary cut-points. Third, the majority of trichotomization and interaction studies reviewed were conducted in the US or high-income European countries. Thus, findings may not generalize to low- and middle-income countries, and certain effects may be culture- or context-specific. For example, in Lansford and colleagues’ [64] study of families from developed and developing countries, maternal warmth mitigated adverse effects of harsh discipline only among African Americans in the US. Additional research should be conducted to gain an international perspective on promotive and protective factors for children. Finally, our review of promotive versus protective effects may be somewhat skewed toward identifying buffering effects, because studies may not publish nonsignificant interaction findings, even if results demonstrate a promotive effect.

In summary, determining whether reductions in externalizing and other common childhood problem behaviors are associated with the presence of positive factors and/or the absence of risk factors has implications for prevention and intervention goals. Knowledge of whether these positive factors have promotive or protective effects provides additional information about intervention targets. Future research should test whether findings generalize cross culturally and probe the sensitivity of the trichotomization approach using different cut-points (e.g., deciles [85]) or alternative statistical methods to determine whether a variable has a linear versus non-linear effect.