Sources of Variability for Students' Engagement in Science: Findings From a Cross-Classified, Multivariate Modeling Approach

• In Event: Affordances and Modeling of Intensive Data

Abstract

Objectives
Engagement is multi-dimensional (Fredericks, Blumenfeld & Paris, 2007) and the consequence of a number of sources (Skinner & Pitzer, 2012), yet how much of the variability in each of the dimensions of engagement can be meaningfully attributed to sources is based on very limited past research (e.g., Strati, Schmidt, & Shumow, 2017). In this work, we seek to understand how much variability in engagement is attributable to situational, individual student and program (or classroom) factors. The research questions that guide this study are as follows:
1. How much of the variability within each dimension of engagement is explained by individual, situational, and class sources?
2. How does the amount of variability explained by each source differ between each dimension of engagement?

Theoretical Framework
A student’s engagement at any given moment is a function of many complex personal, relational, and contextual factors (Skinner & Pitzer, 2012). One’s engagement in a learning task likely has something to do with their personal characteristics such as their gender, interests, or educational background. Engagement is also likely influenced by the immediate features of the situation at hand. Finally, engagement may be shaped by features of learning contexts like one’s teacher, classmates, or the broader learning climate. Although scholars have conceptualized the nature of engagement in multiple ways, it is commonly thought of as involving behavioral, cognitive, and affective dimensions (Fredericks et al., 2007; Sinatra, Heddy, & Lombardi, 2015).

Method and Data Sources
We explore these foundational issues using data from three studies (n = 1,174 students), each of which gathered Experience Sampling Method data on youth’s immediate reports of behavioral, cognitive, and affective engagement (n = 12,858 reports) within particular science learning environments (n = 51 out-of-school science programs or classrooms). Two studies were in formal, K-12 science education settings, while one was in an informal STEM education context. We use cross-classified, multi-level models to estimate the variance in each engagement dimension that is attributable to differences between situations, individuals, and programs. The design of these studies and the multi-level approach to analysis are ideal for identifying variation in engagement at the level of the situation, the individual, and the program.

Results
We present a description of the findings in Figure 1. Hypothesis tests for the variance components showed that across all three studies and all three dimensions of engagement, more variability in youth’s engagement responses was observed at the level of the individual than the immediate situation or program. Affective engagement depended more on situational factors than did behavioral or cognitive engagement. Generally, the program accounted for little variability across all three dimensions of engagement.

Significance
This work contributes to foundational understandings about the nature of engagement. Specifically, it provides an initial, but comprehensive picture of the relative influence of situational, individual, and program factors, and suggests that these influences may impact each dimension of engagement differently. In addition, this study also demonstrates the possibility of analyzing data associated with each dimension of engagement and source of variability in a single, multivariate model.

Authors

• Joshua Michael Rosenberg, The University of Tennessee - Knoxville

• Patrick Neil Beymer, University of Wisconsin - Madison

• Vicky Phun, Michigan State University

• Jennifer A. Schmidt, Michigan State University