Increasingly, randomized trials have been analyzed with attention to moderating roles of person and place, where interventions may fit the needs of some children, in some contexts more than the intervention might for other children, in other contexts (Gorman-Smith & Tolan, 1998). The role of moderators was explored in some studies reviewed here, but not in others, and they represent a very promising direction for future research.
A small number of studies considered the role of child characteristics, such as child gender, race/ethnicity, English-language-learner status, and risks for self-regulatory or expressive language difficulty. For example, children at higher levels of behavioral and cognitive risk (e.g. those children who are more temperamentally or neurocognitively prone to high levels of shyness, impulsivity, or distractability) might be expected to benefit more greatly or less greatly from interventions (see Bierman et al., in press-a for review). Yet this review suggests that few of the socioemotionallyoriented. hybrid, or cognitively-oriented interventions (in ISRC and PCER) considered whether intervention impacts were greater or smaller for children with greater proneness to regulatory skill or difficulty. One exception was the nonexperimental finding that children with greater proneness to shyness had significantly more difficult time establishing positive relationships with teachers in nonexperimental analyses of one PCER- funded intervention (Justice et al., under review). Importantly, child temperament moderated relations between childrens language skills and student-teacher relationship, where children who were temperamentally prone to anger and had low expressive language abilities were at particularly high risk of conflictual relationship with their preschool teachers (Justice et al., under review). Additional findings of moderation of intervention impact by child risk were found for Ravers team for observational measures of child behavioral problems (Raver et al., revised and resubmitted). In future, it will be important to carefully consider whether program impacts are larger or smaller for children with differing profiles of strength versus risk.
Family level risk may also be important and parsimonious way to consider fit of different intervention models for families with substantially differing economic and psychosocial resources. Findings by Piantas team of clear, larger benefit of the MTP program for serving very high-poverty classrooms as compared to programs serving proportionally fewer poor children highlights the importance of including family-level income poverty and related risks in models. A third important set of moderators are those of program type and program resources. For example, an intervention targeting the emotional climate of classrooms may be difficult to implement in settings that are chaotic or disorganized, or under-resourced (see Raver et al., 2008 for review). In contrast, programs that have mental health consultants on staff, on-site personnel to address teacher training, quality improvements, etc. may already be sufficiently resourced that they are likely to show little, if any benefit of additional services implemented through our intervention efforts. In short, it is important to include some observable indicators of level of program resources as covariates and as moderators, to detect whether programs with higher organizational capacity are able to benefit from intervention more so than others (see Assel, Landry, Swank, & Gunnewig, 2007 for examples of heterogeneity of child level program impacts across program type).
Past reviews have highlighted the importance of including socioemotional measures as well as cognitively oriented measures when benchmarking intervention impact (e.g. Raver & Zigler, 1997). There are several key benefits (highlighted earlier) for including socioemotional measures at both child- and classroom levels, even when interventions are targeted toward childrens language and literacy. The inclusion of child social skills and behavior problem measures in the PCER evaluation and some individual PCER studies (e.g., Klein et al., in press) helps to rule out concern, for example, that there may be iatrogenic sequelae from the introduction of interventions targeting language and literacy. Similarly, the inclusion of childrens language and math skills in interventions that target only classroom socioemotional processes offers the opportunity to test whether there are costly tradeoffs (in terms of lower instructional time) or unanticipated benefits (in terms of childrens language gains) when focusing program improvement efforts on socioemotional processes. This cross-domain integration of measures at child- and classroom levels represents an important area of future collaboration and future research.
From a methodological standpoint, the impact of a number of these interventions on childrens socioemotional development was difficult to interpret for this review because of variability in the ways that data were analyzed and reported. A substantial number of studies provided careful, sophisticated analyses of program impact, using Intent-to-treat analyses, multi-level modeling (e.g., HLM), and clear description of model specification so that the role of cluster-randomized status to treatment versus control groups could be clearly identified. In contrast, a smaller number of studies limited their reports to analyses of program fidelity as a predictor of child-level or classroom-level outcomes, effectively reintroducing selection bias into designs that were initially randomized. Future research in this area would be substantially strengthened by a tiered reporting process, whereby intention-to-treat (ITT) analyses and treatment-on-treated/ dosage analyses could both be encouraged.
"index.pdf" (pdf, 307.32Kb)
"apa.pdf" (pdf, 78.45Kb)
"apb.pdf" (pdf, 401.85Kb)
"apc.pdf" (pdf, 288.78Kb)
"apd.pdf" (pdf, 90.04Kb)
"ape.pdf" (pdf, 190.7Kb)