Rethinking Aggression: A Typological Examination of the Functions of Aggression

We compared five subgroups of aggressive youth (n = 1,723, Grades 5 through 10) on a number of adjustment correlates. The subgroups were determined by the self-reported functions (i.e., "why") of their aggressive behavior: (a) an "instru mental" group who were high on instrumental reasons only; (b) a "reactive" group who were high on reactive reasons only; (c) a "both" group who were high on both dimensions; (d) a "typical" group who were moderate on both dimen sions; and (e) a "neither" group who were low on both dimensions. The reactive and both groups showed consistent maladaptive patterns across the adjustment correlates. The instrumental and typical groups both showed generally adaptive and well-adjusted patterns. Surprisingly, the neither group revealed high levels of aggressive acts and showed consistent maladaptive patterns on the correlates. These distinct profiles highlight the utility of a typological approach to classifying aggressive youth and have implications for both assessment and intervention.

one-third to one-half of all such referrals (Kazdin, 1995). These high referral rates are alarming given the negative outcomes associated with aggression and the fact that aggressive behavior can often continue into late adolescence and adulthood (Olweus, 1979;Pulkkinen, 1996).
On the other hand, some aggressive behavior is quite normative during adolescence and may even have adaptive benefits (Hawley, 1999;Trem blay, 2000). In our view, the plethora of negative consequences associ ated with aggressive behavior, coupled with the seemingly paradoxical advantages of aggression, highlight the need for improved approaches to identify and classify subtypes of aggressive youth.
As Tremblay (2000) has recently noted, efforts to identify subtypes of aggressive individuals have had only limited success. In this study, we address this challenge by adopting a person-centered typological approach (Magnusson, 1998) to classify subgroups of aggressive youth. A typological approach has the potential not only to identify specific subgroups of youth for whom aggression is clearly maladap tive but also to reveal subgroups of aggressive youth for whom modest amounts of aggressive behavior are generally normative and possibly beneficial. In this study, we examine five distinct subgroups of aggres sive youth (see later) based on the functions (i.e., the whys, or purposes) of their aggressive behavior. Because a functional perspective directly addresses why aggressive behavior is occurring, it has great potential to broaden our understanding and treatment of aggressive behavior.

Dimensions of Aggressive Behavior
Recent studies generally support the distinctiveness and differential predictive characteristics of various dimensions of aggression (Atkins & Stoff, 1993;Crick, Casas, & Mosher, 1997;Poulin & Boivin, 1999, 2000Price & Dodge, 1989;Pulkkinen, 1996). For example, assessing relational aggression (instead of relying only on overt-physical measures of aggres sion) has narrowed the gap in the reported rates of aggression for males and females (Crick & Grotpeter, 1995;Delveaux & Daniels, 2000). Simi larly, Boivin (1999, 2000) differentiated between proactive and reactive facets of aggression and found that boys' proactive aggres sion predicted increased conflict in stable friendships over the course of a school year, while reactive aggression predicted a decrease in conflict.
Although such studies have identified distinct dimensions of aggression, nearly all of them have confounded the behaviors' form and function and, not surprisingly, have consistently found quite high correlations between the different facets of aggressive behavior (Crick & Grotpeter, 1995;Dodge & Coie, 1987;Little, Jones, Henrich, & Hawley, 2003;Poulin & Boivin, 2000;Price & Dodge, 1989). For exam ple, both instrumental and reactive aggression are often measured with items that also tap overt behaviors as the form of aggression and typi cally correlate in the .6 to .7 range (Poulin & Boivin, 2000;Price & Dodge, 1989). Because of this confound, proper identification, assess ment, and treatment of aggressive youth is undermined. In our view, the ability to purely identify the functions of aggression (i.e., knowing the reasons why an adolescent is aggressive separately from how he or she is aggressive) is critical because different intentions would necessi tate different approaches to intervention and treatment.
To address this common confound, Little et al. (2003) developed a measurement and analysis system that disentangles various forms (overt vs. relational) and functions (instrumental vs. reactive) of aggres sion. In their framework, overt aggression encompasses physical and verbal behaviors directed at another individual while relational aggres sion includes purposeful manipulation of and damage to another's social relationships. Instrumental aggression incorporates behaviors that portend self-serving outcomes (i.e., more offensive in function), while reactive aggression is more defensive in nature, emerging in response to perceived provocations or similar antecedent situations. Using a confirmatory structural equations modeling approach, Lit tle et al. (2003) found strong support for both the internal (i.e., the con vergent and discriminant features of the measured aggression con structs) and the external (i.e., the predictive relations with various outcomes) validity of their instrument. Consistent with the literature, they found that the overt and relational forms of aggression are highly overlapping but uniquely predictive aspects of aggressive behavior. How ever, in stark contrast to past research, Little et al. also found that, after controlling for the over-riding forms of aggressive behavior, instrumental and reactive aggression were, in fact, orthogonal facets of aggressive acts. This novel outcome emerged because their distinctive measurement and analysis system eliminates the inherent confound between the forms (e.g., overt vs. relational) of aggression and its functions (i.e., the whys, or pur poses). Disentangling the "what" from the "why" of aggressive behavior has the advantages of examining (a) how each dimension relates to oth ers and (b) how they relate uniquely to different outcome measures. of aggressive behavior (Ledingham, Younger, Schwartzman, & Berg eron, 1982). For example, Pakaslahti and Keltikangas-Jaervinen (2000) examined peer, teacher, and self-ratings of overt and rela tional aggression and found that the correlation was highest between the peer and teacher reports, followed by teacher and self-reports, with the lowest correlation between peer and self-reports. They also found higher consistency among reporters for overt forms of aggres sion than for relational forms. However, the majority of these cross rater correlations were in the .10 to .65 range, suggesting that reporters disagree more than they agree when it comes to identifying aggressive behavior.
A likely explanation for these discrepancies is that different raters view a target individual's aggressive behaviors in different contexts and situations (Achenbach, McConaughy, & Howell, 1987). Given each reporter's unique perspective and potential biases, a more thoughtful rationale is warranted when choosing which rater to use. In our view, taking into account the functions and forms of aggressive behavior that are of interest can provide such a basis. To examine the instru mental or reactive functions of aggression, self-ratings would be pre ferred because of the inherent difficulty in judging intent and because, beginning in late childhood, one's sense of self is sufficiently developed such that the individual would have some insight as to why he or she is acting aggressively. On the other hand, to assess the forms of aggres sion, multiple sources would be preferred because of the context speci ficity of the expressed forms of aggression. Instead of relying on any one reporter or an aggregate of all reporters, taking a unique look at several reporters (self, friend, peer, parent, and teacher) to identify cross-situation similarities and differences in the expression of aggres sive behavior is warranted.
Typologies of Aggression Based on Self-Reported Function For this study, we were particularly interested in the relations between the instrumental and reactive functions of aggression for iden tifying and classifying subgroups of aggressive youth. Our specific interest in focusing on the functions of aggressive behavior stems from the fact that although attention to the distinction between different forms of aggressive behavior (e.g., overt vs. relational) is instructive, it provides little information about why individuals engage in aggressive behavior. In our view, a better understanding of the functions, or pur poses, of aggressive behaviors is necessary to develop more compre hensive conceptualizations and more powerful approaches to assessing and treating aggressive behaviors.
As mentioned, a notable outcome of the Little et al. (2003) study was that, when the confound of the form of the aggression is con trolled for, the instrumental and reactive reasons for aggressive behav ior emerge as uncorrelated dimensions. This finding was a primary impetus for the current study. Using the same sample of adolescents as the Little et al. study, we significantly extend their earlier validation study by applying a typological analysis to the instrumental and reac tive reasons for aggressive actions.
The independence of the two functional dimensions is particularly well suited to a typological examination because individuals can be logically divided into meaningful subgroups based on where they lie along the two dimensions. A typological classification based on the relative standing along these independent dimensions offers a unique perspective on the resulting subgroups that is not captured by typical continuous-variable approaches (e.g., interaction terms; Hawley, Little, & Pasupathi, 2003). In the absence of absolute cutoffs, the categories are created by dividing the distribution of self-report scores on reactive reasons into thirds (i.e., at the 33rd and 66th percentiles) and crossing them with a similar tripartite breakdown on instrumental reasons (see Hawley et al., 2003;Hawley, this issue). This procedure yields a bal anced 3 by 3 matrix of possible groups (i.e., because the two distribu tions do not overlap systematically, a given score on one dimension is equally likely to be coupled with a high, medium, or low score on the other dimension). On the basis of their logical similarities, some of the cells (e.g., the three middle cells) can be further collapsed to yield at least five distinct typological profiles of aggressive youth (see Figure 1). Note that because these subgroups are defined based on the form-free reasons for their aggressive behavior (i.e., the form of the aggression is controlled), the absolute amounts of the aggressive behavior can vary freely. For example, a child who aggresses primarily for instrumental reasons might behave aggressively quite rarely or quite often.

Specific Hypotheses
Our first goal was to examine the amount or frequency of overt and relational aggression exhibited by the different subgroups. As men tioned, the subgroups are defined based on the reasons for their aggres sive behavior, which are independent of the amount of the aggressive behavior. We were particularly interested in the patterns of aggression exhibited by each subgroup as rated by each reporter (i.e., self, friend, peer, parent, and teacher). As a side note, we expected the correlati among raters to be generally low, with the highest cross-rater corr tions emerging between peer and teacher reports, because both observers of the school context. We also expected to find higher c sistency among reporters for overt aggression than for relatio because overt aggression is more visible than relational aggression.
Our second goal was to examine subgroup differences on a numb of adjustment variables in an effort to identify potential deficits, as we as benefits, of aggressive behavior depending on the defining typol cal characteristics of each subgroup. To accomplish this goal, we ch a representative set of constructs that could adequately characteri each subtype, covering (a) personality, (b) social motivation and co petence, and (c) academic adjustment. Because of the cross-secti nature of this study, we refer to these variables as correlates of ag sion. For the first set of adjustment correlates, we chose three pers ality characteristics that have been associated with aggressive behav hostility, frustration intolerance, and shyness (Asendorpf, 1989;Co Dodge, 1998;Keltner, Young, & Buswell, 1997;Seitz & Rausche, 199 We obtained self-report, peer-nomination, and best-friend assessme on these three characteristics. As an adjunct to these dimensions, we selected two academic indices (problem-solving ability and actual school performance) to evaluate the academic status of these sub groups. Finally, we assessed three self-reported indices of social moti vation and competence (intrinsic vs. extrinsic social orientation and social self-concept) in an effort to identify potential antecedents and consequents of the different subgroups of aggressive youth.
In terms of hypotheses, we generally expected that the different subgroups would exhibit differences in the amounts of aggressive behavior (i.e., on the overt and relational forms of aggression) as well as unique profiles on the selected correlates. In terms of specific sub group hypotheses, we expected the typical group to exhibit generally normative patterns of aggressive behavior (i.e., low to average levels of relational and overt aggression) and exhibit a generally positive profile on the correlates (i.e., low to average levels on shyness, hostility, and frustration intolerance; and average to high levels on social motivation, social competence, and school performance), because this group lies in the middle ground on both instrumental and reactive reasons for their aggressive behavior (see Figure 1). Because the instrumental use of aggression is planful and reward or outcome oriented (Coie & Dodge, 1998), we expected the instrumental group to exhibit average levels of overt and relational aggression but to not show signs of maladaptation on the adjustment correlates (i.e., low levels on shyness, hostility, and frustration intolerance; and average to high levels of social competence and school performance). That is, we expected the instrumental group to be very much like the typical group. Because the reactive use of aggression is associated with a number of self-regulatory and attribu tional deficits (Coie & Dodge, 1998), we expected the reactive group to reveal average to high levels of relational and overt aggression; average to low levels on school performance, social motivation, and social competence; and high levels on hostility, shyness, and frustration intol erance. Because the both group employs aggression purposefully and in response to provocation, we expected them to exhibit high levels of overt and relational aggression; low levels of social competence and school performance; and high levels of shyness, hostility, and frustra tion intolerance. That is, we expected the both group to be similar to the reactive group and perhaps exhibit even more negative patterns because of the expected elevated frequency of aggressive acts. Lastly, given their low standing on both instrumental and reactive reasons for their aggressive behavior, we expected the neither group to be generally reserved and meek (i.e., shy, evince low levels of overt and relational aggression, and average to low levels of social competence, frustration intolerance, and hostility).

Participants
Participants consisted of 5th to 10th graders (mean age, 14.0 years) who were selected from each of the basic school types (elementary, college-bound, vocational, and comprehensive), serving lower to upper middle class suburbs of Berlin, Germany. In total, 1,723 stu dents from five schools participated, reflecting over 75% school-wide participation. Written informed consent was obtained from both par ents and youths. The ethnic makeup of the sample was roughly 82% German, 12% Turkish, and 6% other. For this study, gender, grade, and ethnicity effects, which were generally minimal, were included in all analyses, but, because our focus was on the subgroups of aggressive youth and we did not have specific predictions regarding interactions among them, we report only pronounced (p < .005) interactive effects.

Measures
As part of a larger study, participating students completed ques tionnaires in groups of approximately 30 per class during three 45 minute sessions spanning approximately 2 weeks. For the 5th-and 6th grade participants, a proctor read the questions aloud while a second proctor assisted. Participants in the upper grades completed the ques tionnaires independently while a proctor circulated to answer any questions. Order of questionnaire administration was counterbal anced. Except for the aggression instrument, which was codeveloped in English and German, the various measures were either translated into German (using back-translation and bilingual committee evaluation procedures) or adapted from established measures in the literature.
To obtain peer nominations of the constructs, we used standard sociometric procedures (i.e., students were asked to nominate up to three classmates for each item). For the friendship assessments, students listed up to three very best friends and rated each on a number of dimensions (see later). Only mutual (i.e., both friends reciprocally nominated each other) best friends' evaluations of the focal participant were utilized (η = 1,164). Approximately 70% of the teachers and parents also filled out a questionnaire about the participating children, rendering valid responses for 733 of the participants from teachers and 877 from parents.

Aggression
The self-reported aggression constructs were assessed with six 6 item subscales designed to differentiate the underlying forms and th functional expressions of aggression-overt aggression ("I'm the kin of person who often fights with others," a = .79), overt-reactive aggr sion ("When I'm hurt by someone, I often fight back," a = .82), and overt-instrumental aggression ("I often start fights to get what I want," a = .84) as well as relational aggression ("I'm the kind of person who tells my friends to stop liking someone," a = .62), relational-reactive aggression ("If others have threatened me, I often say mean things about them," a = .63), and relational-instrumental aggression ("I often tell my friends to stop liking someone to get what I want," a = .78). A 4-point scale (never, seldom, often, always) was used (see Little et al., 2003, for all items and detailed validity information).
Following the analytic procedures outlined by Little et al. (2003), the form information (i.e., overt or relational) was separated from the instrumental and reactive aggression constructs, leaving two variables that reflect the unconfounded functions of aggression. Specifically, each construct that contains functional information (e.g., overt reactive aggression) is regressed on to the construct that contains only the form information (e.g., overt aggression). The scores for the two instrumental residuals were averaged to index instrumental aggres sion (a = .82) and the scores for the two reactive residuals were aver aged to index reactive aggression (a = .67). Participants were assigned to the five subgroups based on their standing on the two indexes.
A first subgroup, termed both, includes those participants above the 66th percentile on both instrumental and reactive reasons (i.e., this group reports that their reasons for aggressing are both defensive, in reaction to provocation, and offensive, in planful pursuit of a desired objective). A reactive subgroup includes participants above the 66th percentile for reactive reasons but below the 66th percentile for instru mental reasons. Similarly, an instrumental subgroup includes those above the 66th percentile for instrumental reasons but below the 66th percentile for reactive reasons. A neither subgroup includes those par ticipants below the 33rd percentile for both instrumental and reactive reasons. Finally, the typical group includes those who score below the 66th and above the 33rd percentile range on both self-reported instru mental and reactive reasons. The frequencies for the subgroups are listed in Table 1.
For the other reports of overt and relational aggression (using the same "pure" overt and relational aggression items as the self-report battery), all raters showed acceptable levels of reliability. For overt aggression the reliabilities were: .88 for teachers, .81 for parents, .65 for best friends, and .89 for peer nominations. For relational aggression the reliabilities were: .76 for teachers, .67 for parents, .72 for best friends, and .87 for peer nominations. For the social motivation variables, 18 items were selected from Multi-CAM questionnaire (Little & Wanner, 1997) to examine intr sic social motives (e.g., "Why is it that you try to make new friend it because you like to do it?"; Cronbach's a = .88) and extrinsic soc motives ("Is it because you want to be popular?"; a = .87). The soci competence measure was adapted from Harter's (1982) self-con measure ("I find it easy to be with others," "I find it hard to mak friends"-reversed; a = .75). For the measures of shyness, we adap Asendorpf's (1989) subscales to include self-report (e.g., "I feel tim and shy around others"; a = .91), a single peer-nomination it ("Who is shy?"; a = n/a), and friend report (e.g., "Is this friend shy a = .76). Both the hostility and frustration intolerance measures w adapted from subscales of Seitz and Rausche's (1992) personal inventory for children. Hostility also included self-report (e.g., "I o pick on others"; a = .71), peer nomination (e.g., "Who is mean to o ers?"; a = .70), and friend report (e.g., "Is this friend nice to ers?"-reversed; a = .72). Frustration intolerance also included report (e.g., "I easily get angry when things don't go the way I wan = .83), a single peer-nomination item ("Who gets upset or angry e ily?"; a = n/a), and friend report (e.g., "Does this friend get angry ily with others?"; a = .70). Regarding school-related correlates collected the teacher-assigned marks in verbal and math perform as a combined achievement variable (a = .79) and we used Rav progressive matrices as an index of problem-solving ability (Raven, 1971 α = .91 In terms of the obtained latent correlations among the reporters, peer and teacher reports, as expected, had the highest cross-rater corre lations ranging from .32 to .52 (see Table 2). Parent-and self-report ratings were modestly correlated, ranging from .26 to .33, while those between parent and friend report were extremely low, ranging from .00 to .11. Overall, higher consistency was found for overt aggression between the different reporters, ranging from .24 to .52, with the excep tion of friend and self-report, friend and parent report, and peer and parent report. Bearing in mind that the correlations in Table 2 are dis attenuated (i.e., the effects of unreliability are controlled), the high cor relations between overt and relational aggression within each reporter are somewhat higher than reported in the literature but at a level one would expect given the typical levels of reliability in assessing aggres sion (Crick & Grotpeter, 1995;Delveaux & Daniels, 2000).
To examine the between-subgroup mean differences on the various measures, we conducted a series of three set-wise inclusive MANOVAs followed by univariate ANOVAs that included gender, grade, and eth nicity as well as their interactions. Specifically, for each outcome mea sure, we first included gender, grade, ethnicity, and all of their possible interactions. In the second step, we added the subgroup variable to evaluate the significance of its contribution above and beyond the demographic variables. This step reflects our focal examination of the subgroup differences. In the third and final step, we added all the pos sible interaction terms of the subgroup variable with the demographic variables to explore any possible interactions with the aggression sub groups. Because of the sample sizes and the lack of specific interaction hypotheses, we report only those effects that were significant at the .005 level. The MANOVA groupings (20 in all) were based on source of report (self, peer, friend, teacher, parent) and class of variable (i.e., overt and relational aggression were grouped; hostility, frustration, and shyness were grouped; Raven and achievement were grouped; and motivation and self-concept were grouped). We also conducted four planned comparisons among the five subgroups. First, we compared the instrumental and typical groups against the reactive and both groups because we expected generally opposite patterns across these groups. For the second and third comparisons, we contrasted (a) the instrumental against the typical group and (b) the reactive group against the both group. For the last comparison, we contrasted the nei ther group with the typical group. Figure 2 displays the mean-level differences in the amount of aggression among the five subgroups as rated by the different reporters. As seen in Table 3, each MANOVA and ANOVA was signif icant, with the exception of the teacher ratings. Although a supple mental MANOVA that included rater as a factor revealed that the sub group by rater interaction was significant, F(32, 12264 = 2.63), ρ < .01, visual inspection of Figure 1  and relational aggression. A few effects for the control variables emerged. Boys were higher than girls on overt aggression as rated by the self, F(l, 1722 = 6.42), ρ < .01, peers, .F(l, 1722 = 34.87),ρ < .01, and teachers, F(l, 729 = 24.96),p < .01. For best-friend ratings, a gen der by grade interaction emerged, F(2, 1163 = 5.27), ρ < .01, reflecting large gender differences (favoring boys) in the later grades but smaller in the younger grades. No gender differences emerged for the ratings of relational aggression, but two grade effects were found, one for best friend ratings, F(2, 1163 = 4.66), ρ < .01, and one for teacher reports F(2, 729 = 4.06), ρ < .01. Teachers reported more relational aggression with age while best friends reported less relational aggression with age.  .01, older students were higher than younger s 26.46), ρ < .01, and majority students were higher dents, F(l, 1722 = 13.40),/? < .01. The three-way i by grade by ethnicity, F(8,1722 = 3.32),/? < .005, uninterpretable, suggesting that it emerged as a Self-reported hostility was greater in older studen dents, F(2, 1722 = 5.36), ρ < .01, and it was m minority students than in the majority students both subgroups and less pronounced in the typica subgroups, F(4, 1722 = 4.53),/? < .01. Both peer no = 28.74),/? < .01, and best-friend rated, F(l, 11 hostility were higher in boys than girls. Finally, peer was greater in girls than boys, F(l, 1722 = 6.29),/ Regarding our focal analyses of differences amo sion subgroups, the adjusted means (i.e., controlli effects) for these subgroups are presented in Figu parison purposes, Appendix A contains the raw m deviations for all variables). Three patterns are of these figures and supported by the planned compa 3. First, the reactive group and the both group sh adaptive patterns across the outcomes and the generally more pronounced negative patterns. Sec tal group was very much like the typical group w ing generally adaptive and well-adjusted patterns Finally, contrary to our expectations, the neither ber of patterns that are particularly maladaptive ( H Raven Β Achievement 0 Self-Concept ED Intrinsic Motivation S Extrinsic Mo Aggression Subgroup Figure 3. Self-reported mean-level differences on the possible a consequent correlates.

Discussion
The unique profiles associated with the different subgroups of aggression illustrate the advantages of utilizing subtypes to examine aggressive behavior. Three of the profiles appear to be maladaptive in nature, but differ in their specifics. Two of the profiles appear to be quite adaptive (i.e., the typical and instrumental groups), suggesting that, at least in this age range, some aggression may be normative and beneficial to social developmental adjustment. The most obvious implication of the subgroup differences is for assessment and intervention programs, which would need to adopt different foci depending on where an indi vidual falls within this taxonomy. Before discussing the profiles of the subtypes, we turn first to the issue of differences among reporters' views.

Differences Among Reporters' Views
The cross-rater correlations for overt and relational aggression showed little agreement (see Table 2). This inconsistency emphasizes the importance of interpreting results cautiously depending on the reporter used. These discrepancies are likely related to the different contexts in which the different reporters interact with the participants and the attendant bias of a given reporter (Achenbach et al., 1987). Peer and teacher reports were moderately correlated given that they  share a similar vantage point on the school setting. The least consis tency was between friend and parent reports, suggesting that adoles cents behave differently when with their parents than with their friends.
There was a slight tendency for the cross-reporter correlations of overt aggression to be higher than those of relational aggression (particu larly for the peer-teacher cell), which makes intuitive sense given that overt aggression tends to be highly visible while relational aggression is a more subtle behavior that might not be apparent to all observers.
The self-, friend-, and parent-reports of aggression (see Figure 2) were generally similar in their relative patterns across the five groups, with a few notable exceptions (outlined later). This similarity is likely related to the fact that friends and parents have many interactions with the adolescent across numerous contexts, and thus know him or her better than peers and teachers. Regarding the reports of hos tility, frustration intolerance, and shyness, two general observations can be made. First, the between-group differences were larger for the self-reports than for the peer and friend reports (as was true for the aggression ratings). This difference is likely related to a self-report bias that enhances the pattern of differences, given that the sub groups are also based on self-report. However, the second general observation is that self-, peer-, and friend-reports showed roughly parallel patterns of mean-level differences, with a few notable excep tions.
The general convergence across reporters indicates that the patterns are veridical but the magnitudes of the differences depend on a given reporter's vantage point. On the other hand, the differences among reporters highlight the importance of understanding why and when to use different reporters to report on different aggressive behaviors.

Characterizing the Subgroups
The Typical Group Along with the instrumental group, the typical group was the lea aggressive on both the overt and relational forms of aggression. Th youth also exhibited generally adaptive and well-adjusted patterns the correlates: They do well in school, they are reasonably smart, th have a positive self-concept, they are not too shy, they are able to r late their feelings of frustration, they do not harbor hostile feeli toward others, and they are more intrinsically than extrinsically m vated to acquire and maintain friendships. This typical group refle the bulk of our sample of youth and for the most part they should considered our "control" group. The validity check regarding the r tions across the correlates for this group generally supports this view The Instrumental Group As mentioned, the instrumental and typical groups were genera similar in their levels of aggression as reported by self, peer, and ent. The patterns diverged for friend and teacher reports. The fri of the instrumentally aggressive youth saw them as more relationa and more overtly aggressive than did the friends of the typical gr Perhaps the more intimate perspective of the best friend may be p to the few episodes of aggression that result in beneficial outcom Friends may also be accomplices to acts of relational aggress against a particular target. The teachers of the instrumentally ag sive youth rated them as quite low on relational aggression and, s prisingly, the teachers rated typical children as above average (the tern was similar but nonsignificant for overt aggression). If instrumental youth are more planful and goal oriented about u aggression, they likely would exhibit their aggressive acts more c fully, out of view of the teacher. The general pattern suggests that lescents who utilize aggression in instrumental ways appear to be more aggressive than typical children; however, when they use agg sion, they appear to do so in measured ways and in contexts w friends see it, but teachers do not.
Like typical children, instrumental youth do well in school and have a positive self-concept. Their levels of intrinsic and extrinsic m vations were both about equal and at the average for the sample. A like the typical group, the instrumental youth do not harbor hos feelings toward others and they are able to regulate feelings of fru tion, even more so than the typical children. The story for shyness somewhat different. The peers (and to some extent friends; see Figu of the instrumental youth see them as quite shy, but the youth th selves report that they are no more shy than average. Children w aggress to get what they want would need to feel capable of exert themselves when desired (i.e., not feel shy); these episodes may be s ciently rare that peers do not perceive such a connection. A clear o all pattern for this subgroup that emerges is that they do not appea have any associated social or academic deficits, even though they fully and planfully employ aggression to meet their goals.

The Reactive Group
The levels of aggression for the reactive group were genera above average, but the notable finding was that the parents of re tively aggressive youth reported considerable levels of overt and tional aggression. If parents are witness to these levels of aggressio suggests that the home and family related contexts may be the sett in which these children are most aggressive. Perhaps siblings or the children of the parents' friends are the recipients of the aggressive behavior. It might also be that the parent-child relationship is conflict ual characterized by both overt (e.g., physical tantrums) and relational (e.g., playing one parent against the other) forms of aggression.
Except for the slightly lowered levels of intellective skill, the reactive group was pretty much at the average on achievement, self-concept, and motivation. As expected the reactive youth showed considerable frus tration intolerance and elevated levels of hostility. Because the reactive group mostly responds to perceived provocation with aggressive behav ior, their lack of ability to control their frustration is not surprising. These episodes of reactive aggression may contribute to the levels of hostility. Interestingly, a differential pattern emerged for shyness, which was pretty much opposite that of the instrumental group. Reactive youth self-report that they are shy, but their friends and their peers see them as generally outgoing (or at least near average in the case of the peer reports). Although reactive youth may see themselves as shy, their aggressive acts, which are generally visible because they are provoked, may be seen as a reflection of an outgoing person by peers and friends (i.e., an attributional bias, perhaps).

The Both Group
Those adolescents who use aggression in both ways (reactively an instrumental^) engage in aggressive behavior most often, alongsi the neither group. We expected that the both group would evince hi levels of aggression because of the greater number of contexts and c cumstances in which they potentially could aggress. In terms of th motivational correlates, their motivational imbalance is striking. The reported the lowest levels of intrinsic motivation to establish and ma tain friendships (i.e., for their inherent joy and satisfaction) and, at t same time, they reported the highest levels of extrinsic motivation (i to be popular, gain approval, etc.). The motivation literature has lon been clear on the undermining effects of extrinsic motivation (see e.g Deci & Ryan, 1985), and coupled with the low levels of intrinsic mo vation, puts this group at risk for generally dissatisfying relationshi that may lead to other long term adjustment problems and feelings alienation from the peer world. Although this group does not appea to suffer from a lowered self-concept, their friends' views might sugges otherwise. The reciprocal best friends of these children see them a highly relationally aggressive, moderately overtly aggressive, and ho tile. In terms of their academic profiles, this group was the lowest p forming and scored lowest on the measure of intellective skill. One might speculate that this group would be at risk for school drop out because of their low academic standing and the potential for later adjustment problems in their social worlds.

The Neither Group
The high levels of aggression for the neither group was absolutely opposite of our expectations. They scored highest on intellective skill, but perform only slightly above average in school-a generally under achieving pattern. Although their motivations for social relationship were in the adaptive direction, their social self-concept was strikingly lower than any of the other subtypes. These youth report that they have hostile feelings toward others and their friends agree, as does the peer world. The friends of these youth also report that they are gener ally frustration intolerant. This group does not appear to be particu larly shy.
In our view, the profile of these youth depicts a hostile aggressive pattern, wherein aggression is neither provoked nor wielded for gain. Perhaps the aggressive episodes are an attempt to enhance a low self concept. Moreover, the underachieving nature of this group suggests that the poor adaptations in the social world may be undermining their ability to perform well in school. Clearly future work will need to identify the manner in which their academic world is hampered by their social world. Clues to this relationship could then be used to tai lor an intervention program specifically for this category of aggressive youth.

Limitations and Questions for Future Research
Some of the deficiencies of this study warrant discussion. First, the reliance on self-report of the functional purpose of aggressive acts is not without potential bias. The degree to which other factors such as self-presentation bias, acquiescent-response bias, and the like are involved does weaken the virility of our classifications. Clearly, further criterion-related validity work would help to establish the accuracy of the taxonomic classifications. In addition, other methodologies such as the use of vignettes, structured interviews, or controlled experimental procedures (e.g., Hawley & Little, 1999) could be used that reduce the potential impact of self-report biases.
A second limitation that future work will need to address is the lack of absolute cutoffs for classifying youth. We chose to use the rela tive differential in the scores from this representative sample of Ger man youth to establish the groups for this study. However, in order for practitioners to be able to use these categories meaningfully, diagnostic tools will need to be developed that include normed criteria for classi fication.
Third, our study only intimated potential antecedents and conse quents of the different subgroups of aggressive youth. This strictly cross-sectional approach does beg a number of questions that only lon gitudinal work could address. For example, are characteristics such as shyness, frustration intolerance, and hostility antecedents to aggressive behavior or are they consequents of it? Are the subgroups of aggres sive youth stable over time or are they dependent upon age and con text? In this longitudinal work, expanding the breadth of correlates would also serve to help elucidate the nature and meaning of these pro files.
Lastly, one question for future work that would provide a critical piece for this puzzle is a detailed examination of who the recipients of the different aggressive acts are. We would postulate, for example, that instrumentally aggressive youth would not choose "easy targets" like the neither group might, but instead would choose targets that hold a desired resource or who are at about the same level of social domi nance. A well-placed and successful aggressive act toward a near chal lenger would have a double effect of thwarting the challenger and send ing a message to all other challengers at or below the thwarted challenger's rank: "Don't even think about it." Similarly, the source of the elevated reports by parents would be better understood knowing who the recipients were.

Conclusions
Overall, our efforts to classify aggressive children into subgroups successfully revealed striking profiles associated with the subgroups with direct implications for intervention work. These unique configu rations suggest that an intervention designed for one type of aggressive behavior (e.g., reactive aggression or bullying) might not be effective for another. For example, an intervention trying to curb hostility will be more effective for children who display reactive aggression and less effective for children who exhibit instrumental aggression. Many inter vention programs for aggressive children target frustration intolerance, hostile attribution bias, and impulsive responding to provocative stim uli (Kazdin, 1995). Our findings suggest that although these dimen sions may be important points of intervention for some children (i.e., the reactive and both subtypes), they are likely to be less important in the treatment of other aggressive children (i.e., the instrumental and neither subtypes). For the neither group, other motivations need to be considered such as problems with self-esteem. In other words, more attention needs to be given to assessing the functions of aggression in order for intervention programs to be optimally effective.
A p p e n d i x A . R a w m e a n s a n d s t a n d a r d