Social Information-Processing Skills and Aggression

The Reformulated Social Information-Processing Model

The design of the Making Choices program was primarily guided by social information-processing (SIP) theory (Crick & Dodge, 1994; Dodge, 1980, 1986, 2006; Huesmann, 1988). Stemming from early work on social problem solving (e.g., D’Zurilla & Goldfried, 1971; McFall, 1982), the SIP model was introduced by Dodge (1986). Later, Crick and Dodge (1994) reformulated the initial SIP model, such that responses to social situations are formulated through a series of five overlapping cognitive steps. The five cognitive steps include Step 1, encoding of external and internal cues; Step 2, interpretation and cognitive representation of those cues; Step 3, clarification and selection of a goal; Step 4, response search or construction; and Step 5, response decision (for reviews, see Crick & Dodge, 1994, 1996; Dodge, 2006). Summarized briefly, encoding refers to the perception of external and internal cues. External cues are comprised of social and environmental stimuli, whereas internal cues are comprised of one's cognitive indicators of emotional state (e.g., negative thoughts) and physical indicators of emotional arousal (e.g., heart rate, muscle tension, perspiration). Interpretation involves deriving meaning from internal and external cues and inferring the intentions of others. Goals are formulated during a goal formulation phase, and a range of potential responses are generated during a response search phase. In the response decision phase, children evaluate the appropriateness of potential responses and speculate about what might happen if they enact those response options. During this phase, children also assess their ability to successfully execute each of the various responses they have formulated. Enactment is the step during which children apply social skills to carry out selected responses.

The Role of SIP Skills in Aggressogenic Processes

The reformulated SIP model was intended to serve as a global framework for understanding the cognitive operations underlying child behavior, but its primary application has been toward the study of aggressive behavior in children. Many empirical studies have demonstrated the relationship between processing patterns at each of the five SIP steps and aggressive behaviors. Specifically, research suggests that, as compared with their nonaggressive peers, aggressive children encode fewer and more negative social cues (Step 1; Dodge & Newman, 1981; Gouze, 1987; Strassberg & Dodge, 1987); attribute more hostile intentions (Step 2; Feldman & Dodge, 1987); select goals that damage relationships (Step 3); generate fewer and less prosocial responses (Step 4; Pettit, Dodge, & Brown, 1988); and evaluate aggressive responses more favorably and expect more positive outcomes from aggressive behavior (Step 5).

The Role of Emotion Regulation in Aggressogenic Processes

Emotion is an integral component of the SIP model. Emotion is thought to affect behavior by motivating the processing of social information and by attenuating emotional and behavioral responses to physiological arousal (Arsenio & Lemerise, 2001, 2004; Dodge, 1991; Lemerise & Arsenio, 2000; Lemerise & Maulden, 2010; Campos, Mumme, Kermoian, & Campos, 1994). Poor emotion regulation has been linked to physical aggression (for a review, see Eisenberg & Fabes, 1999). When children who have difficulty managing their emotions encounter an emotion-arousing social situation, they more often rely on automatic schema and scripts rather than on unique cues. They also perceive fewer cues, generate fewer solutions, and are more likely to select aggressive responses (Bierman, Smoot, & Aumiller, 1993; Eisenberg et al., 2001; Izard, 2002). Moreover, children with deficits in emotion regulation skills are more likely to display strong affect, which elevates risk for peer rejection and victimization (Eisenberg & Fabes, 1992; Hubbard, 2001; Schwartz et al., 1999) and experience poor overall psychosocial adjustment (Lengua, 2003, Wyman et al., 2009).

Social Cognitive and Emotion Regulation Skills Training for Children

Given the research findings showing linkages between SIP skills, emotion regulation skills, and children’s social adjustment, many programs have been developed and implemented in regular classrooms to teach children social and emotional regulation skills (e.g.,Durlak, Weissberg, Dymnicki, Taylor, & Schellinger, 2011; Wilson & Lipsey,2006, 2007). Some of these programs are multi-element programs that include teachers and/or families in addition to students (e.g., Fast Track, Conduct Problems Prevention Research Group [CPPRG], 1992; Positive Action: Beets et al., 2009); others are child-focused programs in which the child is the primary target of the intervention (e.g., Making Choices: Fraser et al., 2005; PATHS: CPPRG, 2010; and Second Step: Holsen, Smith, & Frey, 2008). Both types of programs have used a variety of theories in designing their curricula, with only a few explicitly using the SIP model or perspective. Although all of these programs were designed to enhance one or more SIP skills, only a few programs addresses process patterns at each SIP step (e.g., Fraser et al., 2005; Greenberg & Kusché, 1998; Kam, Greenberg, & Kusché, 2004; Nash, Fraser, Galinsky, & Kupper, 2003; Orpinas, Parcel, McAlister, & Frankowski, 1995; Van Schoiack-Edstrom, Frey, & Beland, 2002). The majority of evaluation studies of the child-focused programs have reported positive effects on reducing aggressive and disruptive behavior (e.g., Domitrovich, Cortes, & Greenberg, 2007; Fraser et al., 2005, 2011;Slough, McMahon, & Conduct Problems Prevention Research Group, 2008), including recent studies published after Wilson and Lipsey’s systematic review of SIP-based programs (2006, 2007). However, these studies have offered limited evidence to support a link among SIP skills, emotion regulation, and aggressive behavior.

This article reports the findings of a controlled trial of MC and MC+. This study would have two major contributions to the field of promoting social emotional skills and preventing aggressive behavior of children. One contribution of this study is that it uses a theory-based approach to evaluation, by estimating effects on theoretical mediators (SIP skills) and aggressive behavior and testing whether program effects on aggression are mediated through SIP skills. Second, multi-element programs are often achieved at high implementation cost and may pose significant implementation burden on teachers, principals, and other school staff (e.g., Foster, 2010). Thus, another contribution of this study is that it estimates the incremental benefit of adding teacher-and family-oriented strategies to a child-focused program. Evaluation studies which test different versions of a program allow program developers to improve their program and assess whether modifications to the core components are needed.

Prior studies on MC and MC+

Prior evaluation studies of the MC and MC+ programs found effects on SIP skills, peer acceptance, social competence, emotion regulation, and aggression (Fraser et al., 2005; Nash et al., 2003). A previous study using the same intervention cohorts but a different comparison cohort (N = 548) found positive effects on aggression and social competence, with a slightly broader pattern of effects for MC+ (Fraser et al., 2005). The study also detected an interaction effect for gender, with boys demonstrating greater reductions in aggression than girls. Posttest-only measures of SIP skills suggested that intervention students in both MC and MC+ were more skillful than comparison students in encoding social cues and setting prosocial goals. The current study includes data from a lagged-comparison cohort from whom both pretest and posttest SIP data were collected. It sought to extend the prior study, by estimating the effects of MC and MC+ on SIP skills at posttest and by comparing the effects of MC with those of MC+.

Participants

Two rural schools in Chatham County North Carolina participated in the study. These schools were selected from eight public elementary schools invited to participate in the program, based on their level of interest and the absence of potentially competing social–emotional skills training programs. Because the Making Choices curricum was adopted by the school district as a routine element of their health education curriculum, and because data were obtained for educational purposes, parental notification but not active parental consent was required. The participating schools coordinated data collection and provided de-identified data for analysis. Consent and procedures were approved by the institutional board of the University of North Carolina at Chapel Hill and the school district.

Study Design

Three cohorts of third-grade students from each school (N = 541) participated in the study. Cohort 1 (2001–2002; n = 185) participated in MC; Cohort 2 (2002–2003; n = 202) participated in MC+, and Cohort 3 (2004–2005; n = 154) had no exposure to the intervention (for a participant flowchart, see Figure 1). The addition of a lagged-comparison cohort (Cohort 3), CC, permitted the examination of pretest to posttest changes in SIP skills. This study design improves on the prior evaluation, which had simply estimated posttest-only effects on these skills (Fraser et al., 2005; Nash et al., 2003). Data from Cohort 3 (CC) were collected 1 year following the delivery of MC+ to Cohort 2. Observations of teachers’ classrooms and consultations with principals, teachers, and school social workers confirmed that this comparison cohort had not been exposed to the MC+ intervention. Therefore, the risk of experimental contamination was considered to be minimal (see Figure 2 for study design).

OPEN IN VIEWER

Figure 1.

Participant flowchart.

OPEN IN VIEWER

Figure 2.

Study design.

Implementation Characteristics

Sessions were led by program specialists who had received a Bachelor’s degree in social work or education. Students in both intervention conditions (MC and MC+) were offered a 23-lesson classroom curriculum. Lessons were delivered weekly for an average duration of 45 minutes. In addition, students in MC+ were offered supplemental classroom activities led by their teachers and their parents were invited to attend multifamily group meetings. On average, MC+ teachers implemented approximately one supplemental activity from the teacher activity manual each week, (according to teacher records and observational data from the first few months of the intervention). In addition, a classroom management component was added. Teachers at one school elected to implement a classroom management strategy called the Good Behavior Game and received training on how to implement this strategy (Barrish et al., 1969), while teachers at the other school opted to rely on existing classroom management protocols. Parents of third-grade children in the MC+ cohort were invited to five orientation programs that were delivered in English and Spanish from 5:00 to 7:00 p.m. in the school cafeteria.

Dosage

Although the intervention was a required part of the school day, student attendance varied due to external factors, such as teachers pulling students out of the classroom for alternative classes, such as English as a Second Language. Attendance was taken at every session. Student’s exposure to the program was estimated by calculating the percentage of sessions attended. About four of five students received at least 90% of the sessions and about two of five attended all of the sessions. In contrast, attendance for family nights (offered to parents of students in MC+) was low. Although project staff attempted to remove barriers to participation and provide incentive for participant engagement (e.g., offering free transportation, providing child care, food, and mailing bilingual monthly newsletters), only 27% (55 of 202) of the children in MC+ had a parent who participated in at least one session.

Analytic Sample

Despite preserving assignment to original study condition (dropping retained students from the enrolled sample to prevent cross-cohort contamination), and including study participants who did not attend all sessions (including 9 who did not attend a single session) were included in the analytic sample, this analysis was not fully intent to treat. Of 541 students, 62 were excluded from the analysis because of missing pretest or posttest data (i.e., CC: 24; MC: 29; MC+: 9), leaving a sample of 479 students. As shown in Table 1, the analytic sample (N = 479) was gender balanced and ethnically diverse (i.e., 45% Latino, 34% White, non-Latino, and 17% African American across cohorts). Study participants ranged from 7.2 to 11.6 years of age (M = 8.7; standard deviation [SD] = .66). Across both schools, 53% of the students were eligible for the free and reduced lunch program (an indicator of poverty status), which exceeded the 2005 national average for fourth graders (41%; US Department of Education, 2006).

OPEN IN VIEWER

Table 1.

Characteristics of the Making Choices, Making Choices Plus, and Comparison Condition Cohorts.

	MC	MC+	CC	Total
N (count)	156	193	130	479
Race/ethnicity
Latino	40% (62)	41% (79)	56% (73)	45% (214)
White	34% (53)	40% (77)	25% (33)	34% (163)
Black	21% (33)	16% (30)	15% (20)	17% (83)
Other	5% (8)	4% (7)	3% (4)	4% (19)
Gender
Male	51% (80)	51% (99)	45% (58)	50% (237)
Female	49% (76)	49% (94)	55% (72)	50% (242)
Pretest scores
Encoding	.44 (.17)	.48 (.19)	.59 (.17)	.50 (.19)
Hostile attribution	.66 (.26)	.58 (.22)	.64 (.21)	.62 (.23)
Goal clarification	.76 (.29)	.77 (.28)	.75 (.26)	.76 (.28)
Response decision	.60 (.34)	.71 (.31)	.68 (.31)	.67 (.32)
Emotion regulation	3.22 (.85)	3.26 (.94)	3.36 (.78)	3.27 (.87)
Aggression	.40 (.25)	.38 (.25)	.39 (.17)	.39 (.23)

Note. CC = comparison condition. Percentage refers to row percentage. Parentheses denote cell count (n).

Measures

Using teacher and child reports, six outcomes (i.e., encoding, hostile attribution, goal formulation, response decision, emotion regulation, and aggression) were assessed twice a year. Pretest data were obtained in October, so that teachers had sufficient knowledge of their students in order to accurately assess their behavior, and posttest data were obtained in April, so that preparation for end-of-grade tests did not pressure teachers to hurriedly complete the surveys. Data on children’s SIP skills were also collected in classroom settings in October, for pretest, and April, for posttest.

Data Analysis

Hiearchical and multiple regression methods were used. Prior to conducting explanatory analyses, attrition and selection bias analyses were conducted to evaluate potential threats to internal validity. All analyses controlled for baseline outcome measures and the demographic characteristics of the participants. Two-tailed tests were used to test study hypotheses, using a critical p value of .05.

Multilevel models

Multilevel models (MLMs) are generally recommended when assessing the effects of programs that are administered at the group level (Krull & MacKinnon, 1999). One rule-of-thumb for determining whether a multilevel approach should be used is to consider the level of intraclass correlation (ICC) needed to produce a design effect (DEFF). The ICC (ρ) is a measure of the proportion of between-group to total variability, for two-level models, ρ = τ²/(σ² + τ²). The DEFF is the ratio of the actual variance to the variance computed under the assumption of simple random sampling, DEFF = 1 + δ (n – 1), where n is the cluster size and δ is the ICC (Shackman, 2001). An ICC of .06 was selected as the cutoff because it produced a design effect of 2.0, which is considered high enough to warrant the use of a multilevel approach (Muthén & Sattorra, 1995). Two of six dependent variables (i.e., hostile attribution and response decision) did not have sufficiently large ICCs. For these variables, single-level regression models were estimated using maximum likelihood estimation (MLE). Effects on the other variables were estimated using SAS Proc Mixed Version 8.0 (SAS, 2000); the unconditional ICCs for these variables were aggression (ρ = .09), emotion regulation (ρ = .23), goal formulation (ρ = .06), and encoding (ρ = .09). All models were adjusted for the effects of the outcome-specific pretest measure and demographic covariates (i.e., male, African American, and Latino). For the MLM models estimating the effects of intervention versus comparison, the statistical covariates were included at Level 1 (student level) and fixed effects for the two program indicators (MC and MC+) were included at Level 2 (classroom level). Several preliminary models were run to determine whether the models fit better when including random slopes for the effects of pretest on posttest; however, the random slopes were not found to be significant.

Figure 3 shows the baseline multilevel equation used to estimate the effects of MC vs. CC and MC+ vs. CC on overt aggression. The subscript “ij” refers to an individual student’s score on a particular scale (e.g., student “i” in classroom “j”); the subscript “j” refers to a classroom’s average score on a particular scale. The subscripts for MC and MC+ are “j” because the interventions were administered at the classroom level, meaning all students belonging to a particular classroom were either assigned to the intervention or not assigned to the intervention.

OPEN IN VIEWER

Figure 3.

Example of multilevel equation to estimate intervention versus comparison group effects.

Not shown in Figure 3, the difference between MC and MC+ was estimated by replacing the variable for MC with CC, permitting the effect of MC+ to be estimated as the difference between MC and MC+. Moderating effects were tested by adding two cross-level interaction terms, B_5j (MC × male) and B_6j (MC+ × male), to the model. Consistent with the recommended strategy for testing fixed effects in HLM (Raudenbush & Bryk, 2002), interaction effects were estimated using full MLE procedures and then tested using a deviance test approach. A normal theory chi-square difference test was also used to test interaction effects in the single-level models.

Effect sizes (i.e., Cohen’s ds) represent the magnitude of program effects in the standardized units. For MLMs, the effect sizes were estimated by dividing the parameter estimates for MC and MC+ by the square root of the total variance (δ = β/[(τ² + σ²)^1/2]); Raudenbush & Bryk, 2002). For single-level models, effect sizes were estimated using the Cohen’s d formula, Cohen’s d = (Mean_p − Mean_c)/ s (Cohen, 1988), which calculates the difference in adjusted means divided by the pooled standard deviation of the data.

Effect sizes were estimated for MC and MC+ relative to the comparison condition. When program effects were moderated by gender, effect sizes were estimated separately for males and females, using conditional models (see Table 4 for effect size estimates).

Sociodemographic Effects

Consistent with prior research (Hughes, Meehan, & Cavell, 2004), gender differences in SIP skills were found. Across cohorts, boys tended to be less skillful than girls in emotional regulation (Table 2; B = −.16, p < .005), encoding (B = −.06, p < .001), response decision (B = −.12, p < .001), and goal formulation (Table 3; B = −.10, p < .001). In addition, African American and Latino children reported less neutral/friendly social goals (African American: B = −.07, p = .034; Latino: B = −.06, p = .018) than White, non-Latino children. Ethnic differences in hostile attribution were also found, with Latino students being rated by teachers as having higher levels of hostile attribution than non-Latino students (Latino: B = .04, p = .043).

OPEN IN VIEWER

Table 2.

Fixed and Random Effects for Conditional and Unconditional Models of the Effects of MC and MC+ Models without Interactions.

		Emotion Regulation		Encoding		Hostile Attributiona		Response Decisions
Level	Effect	Est.	SE	Est.	SE	Est.	SE	Est.	SE
Student	Intercept	3.43***	.09	.65***	.01	.56***	.01	.79***	.02
	Pretest SIP skill	.61***	.04	.33***	.04	.38***	.04	.37***	.04
	Male	−.16**	.06	−.06***	.01	−.01	.02	−.12***	.03
	African American	−.14†	.08	−.03	.02	.05†	.03	−.03	.04
	Latino	.08	.07	.00	.02	.04*	.02	−.04	.03
Classroom	MC vs. CC	.40†	.23	.04	.03	−.06*	.02	.07*	.03
	MC+ vs. CC	.37	.22	.01	.03	−.12***	.02	.14***	.03
	MC+ vs. A	−.03	.12	−.02	.03	−.06*	.02	.07*	.03
Random effects	L1 variance(e)	.347***		.021***		.041***		.07***
	L2 variance(u11)	.204***		.003**		n/a		.00

Note.CC = comparison condition; Est = estimate; SE = standard error; L1 = Level 1; L2 = Level 2; SIP = social information processing. ^aSingle-level model.

^†Marginally significant effect (p < .10). *p < .05. **p < .01. ***p < .001.

OPEN IN VIEWER

Table 3.

Fixed and Random Effects for Conditional and Unconditional Models of the Effects of MC and MC+ Models with Interactions.

Level	Effect	Aggression		Goal Clarification
		Est.	SE	Est.	SE
Student	Intercept	.15***	.02	.05	.01
	Pretest	.71***	.05	.30***	.04
	Male	.04†	.02	−.10***	.02
	African American	.03	.03	−.07*	.03
	Latino	−.03	.03	−.06*	.02
Interacted Terms
MC vs. CC	For Females	−.03	.06	−.02	.04
	For Males	−.22**	.06	.11*	.04
	Diff. (Interaction)	−.20***	.05	.13*	.06
MC+ vs. CC	For Females	−.05	.06	.07†	.04
	For Males	−.22**	.06	.20***	.04
	Diff. (Interaction)	−.17**	.05	.13*	.05
MC+ vs. MC	For Females	−.02	.06	.09*	.03
	For Males	.01	.06	.09*	.04
	Diff. (Interaction)	.03	.05	.00	.05
Random effects	L1 Variance(e)	.05***		.054***
	L2 Variance(u11)	.01**		.001

Note. CC = comparison condition; Est = estimate; SE = standard error; L1 = Level 1; L2 = Level 2.

^†Marginally significant effect (p < .10). *p < .05. **p < .01. ***p < .001.

Intervention Effects

Effects on SIP skills were varied. Although MC and MC+ had no statistically significant effects on encoding and emotion regulation, both interventions were associated with improved response decision (B _MC = .07, p = 0.029 and B _MC+ = .14, p < .001) and lower hostile attribution (B _MC = −.06, p = .021 and B _MC+ = −.12, p < .001; Table 2). In addition, the interventions improved goal formulation among boys. Boys in MC and MC+ (but not girls) were significantly more likely to select neutral/friendly goals in response to hypothetical peer provocations (B _MC = .11, p = .016; B _MC+ = .20, p < .001) than their same-sex peers in the comparison group (Table 3). On each outcome for which statistically significant effects were found, MC+ was found to be more efficacious than MC. On average, children in MC+ reported significantly greater improvement in hostile attribution (B = −.06; p = .021), response decision (B = .07; p = .037), and goal formulation (B = .09, p < .001) than children in MC.

A gender-biased intervention effect was found for aggression, with statistically significant effects for boys but not for girls (Table 3). Intervention boys had significantly lower scores on aggression at posttest (B _MC = −.22, p = .002 and B _MC+ = −.22, p = .002), while intervention girls had scores that did not significantly differ from comparison girls (B _MC = −.03, p = not significant [ns] and B _MC+ = −.05, p = ns). For both boys and girls, the comparison between MC and MC+ was not significant.

As shown in Table 4, the effect sizes for MC and MC+ for SIP skills varied in magnitude from small (for MC+ on encoding: d = .065) to medium (for MC+ on goal formulation, males: d = .853); and effect sizes for aggression were large for males (d = .898 for MC and MC+) and small for females (d = −.122 for MC and d = −.204 for MC+).

OPEN IN VIEWER

Table 4.

Effect Sizes for MC and MC+.

Outcome Variables	MC	MC+
Emotion Regulation	0.539	0.498
Encoding	0.258	0.065
Hostile Attribution	−0.251	−0.557
Response Decision	0.213	0.452
Goal Formulation (Males)	0.469	0.853
Goal Formulation (Females)	−0.085	0.298
Aggression (Males)	−0.898	−0.898
Aggression (Females)	−0.122	−0.204

Note. Cohen’s d effect size, using adjusted regression coefficients multilevel models (encoding, emotion regulation, goal formulation, and aggression) and adjusted mean differences for single-level models (hostile attribution and response decision).

Limitations

The study design and analysis limited in several ways. First, the use of a purposive sample prohibits generalizability to a higher-order population of schools across the country. Second, participants were not randomly assigned to study conditions. Lack of randomization increases the potential for selection effects between conditions on observed and unobserved factors. Although the use of a cohort design within schools controls, in part, for site and other potential sources of selection bias, comparisons on background measures revealed several between-cohort differences on pretest measures. These potential selection effects were controlled statistically. Third, the use of a treatment-withdrawal cohort design increases potential for experimental contamination. Prior to obtaining consent for this wave of data collection, teachers reported having discontinued the use and display of MC materials in their classrooms, and they stated that they were no longer implementing MC-related lessons and activities. The threat of treatment contamination appears minimal. Fourth, history is a potential confound in cohort designs. During the 2-year lag between the intervention and the comparison cohorts, the population of Latino children increased from 41% to 57% (n = 479). Indeed, statistical tests conducted with the entire sample showed only one significant difference. At pretest, Latino children were rated lower on cognitive concentration, t(478) = −2.09, p = .037; n = 479, than non-Latino children. Finally, to our knowledge, the cohorts were not differentially exposed to major historical events, such as the implementation of the No Child Left Behind reforms, which could produce systematic differences across conditions.

Other study limitations relate to model specification. First, the use of an MLM limited statistical power to detect small main and moderated effects by decreasing sampling units from the number of students (N = 479) to the number of classroom clusters (J = 28). On balance, the resulting low power does not appear to have compromised significance testing, as main findings fell in the moderate to large effect size range. Second, given that quasi-experimental studies are prone to selection bias, a propensity score analysis might provide an alternative means for balancing cohorts and estimating program parameters. It is not clear, however, that a propensity score approach would provide advantages over the covariance control that we employed (Guo & Fraser, 2010; Shadish, Clark, & Steiner, 2008).

Implications for Prevention

Two major implications can be derived from this study. First, as suggested in prior studies of Making Choices, school-based programs with conceptual foundations in SIP appear to hold the potential to disrupt developmental pathways leading to conduct problems in childhood and early adolescence. Few SIP-focused, school-based prevention programs have measured SIP skills, despite a growing evidence base suggesting that deficits in SIP skills are related developmentally to aggressive and delinquent behavior. This study suggests that SIP skills are responsive to classroom-based intervention strategies.

Second, findings also suggest that it may be possible to strengthen the existing interventions by addressing gender-specific risk factors. Boys appear to have differentially benefited from the programs. Both the MC+ and the more economical (i.e., lower implementation burden) MC interventions had large effects on aggression in males. This study suggests that a universal, school-based primary prevention program targeting SIP skills can reduce aggression in males, who are at greatest risk for engaging in delinquency and crime. Given these findings, further research to explore whether gender-sensitive and gender-specific violence and delinquency prevention programs might yield stronger effects.