The Role of Teacher-Child Interaction Patterns in Social Relationships: Drawing Causal Inferences Using Propensity Scores

• In Event: Teacher-Student Interaction Patterns: Integrating Models of Instructional and Social Dynamics in the Classroom

Abstract

Objectives
This study explores the association between teacher-child interaction quality in elementary school classrooms and children’s social relationships in the classroom. We employ the Classroom Assessment Scoring System (CLASS; Pianta et al., 2008), an observational measure of teacher-child interaction quality frequently used in both research and practice settings, and use advanced methodologies to address two common limitations of research using the CLASS. First, high intercorrelations among the three CLASS domains limit the ability of researchers to identify components of teacher-child interaction quality that are most important for specific outcomes (Hamre et al., 2013). Second, making causal conclusions based on CLASS scores is difficult because children are not randomly assigned to their contexts; that is, spurious correlations may be due to confounding variables.

Theoretical framework
Our research is guided by the bioecological theory of development (Bronfenbrenner & Morris, 2006), which views development as the product of an individual’s characteristics, multiple contexts (e.g., family, school), and the interactions between the individual and his or her contexts.

Methods
Pairs of observers rated teacher-child interaction quality at the beginning of the year using the CLASS. Approximately six weeks before the end of the school year, we assessed children’s perceptions of closeness to the teacher (7 items; α = .82), peer community (5 items; α = .82), and loneliness (3 items, α = .87). Numerous teacher and child demographics and classroom characteristics were collected as potential confounds.

Analytic strategies involved structural equation modeling for the bifactor model, logistic regression to obtain propensity scores, and multilevel models for the outcome analyses.

Data sources
Data were collected from a diverse sample of 2274 children in 109 1st, 3rd, and 5th-grade classrooms in Pennsylvania, Illinois, and Indiana. 1956 children (86%) participated at least once; of these children, 1714 were in the classroom the entire year and were used in the analyses.

Results
The bifactor modeling strategy resulted in three uncorrelated factors: a “general” factor that accounted for shared variance among all ten domains and two “specific” factors that accounted for the organizational and instructional elements of interactions. The general factor is the focus of the current analysis. Following Hamre and colleagues (2013), we labeled this factor “Responsive Teaching” (RT).

Classrooms were classified as high or low RT using a median split (median = 4.6). Propensity score models estimated a classroom’s likelihood of having high RT based on covariates. Classrooms with similar propensity scores were paired, and a dataset of matched classrooms was created (N=70 classrooms). Multilevel models showed significant effects of RT on perceived teacher closeness and loneliness, and marginally significant effects of RT on peer community (see Table 1).

Scientific or scholarly significance of the study or work
Responsive Teaching was a strong predictor of children’s relationships with both teachers and peers, suggesting that teachers’ direct interactions with students may also affect how peers interact with each other. Using propensity scores is a promising approach for examining the effects of classroom contextual variables while avoiding spurious associations due to confounding variables.

Authors

• Rebecca Madill, Pennsylvania State University

• Scott D Gest, The Pennsylvania State University

• Philip C. Rodkin, University of Illinois