Predictors of Adolescents' Generation of Causal and Noncausal Inferences

• In Event: 68.018 - Expanding Our Understanding About Comprehension: Theory, Assessment, and Intervention

Abstract

Objectives or purposes. The purpose of this study was to evaluate the effects of an array of reader and text characteristics on adolescents’ generation of causal and non-causal inferences.

Perspective(s) or theoretical framework. According to the construction-integration model (Kintsch, 1994; 2004), a reader can derive one or more levels of comprehension (i.e., surface, textbase, situation model). The situation model is a mental model of text information that is complete, integrated, and meaningful; this level is necessary for learning from text. The situation model extends beyond the literal meaning of the text, derived through the comprehension and integration of implied as well as explicit relationships in the text. Text-based inferences related to the causes or consequences of events serve to integrate and reinforce text information within a coherent representation in memory (Magliano, Trabasso, & Graesser, 1999).

Methods, techniques, or modes of inquiry. Inference generation was measured through the analysis of think-aloud verbal protocols generated as students read narrative or informational text. Codes were analyzed quantitatively along with student assessment data and text variables. Multilevel models of person and text characteristics were fit to predict inference generation.

Data sources, evidence, objects, materials. Participants were 325 students in grades 7, 9, and 11, who were assessed with a range of reading measures, including the collection of think-aloud protocols (Author et al., 2015). For the current analysis, text segments previously coded as inferences were recoded as either causal or non-causal. We also utilized data from assessments of word reading, vocabulary, and working memory. Participants were designated as adequate or poor comprehenders based on the state reading test. Students were stratified by grade and comprehender group and randomized to read either informational or narrative text. Each participant read two texts in the assigned genre. One was accessible (aligned with the student’s Lexile reading level), and the second was challenging (about 350 Lexiles higher). Texts were individually selected for each student relative to their Lexile levels, necessitating the availability of a large corpus of 36 narratives and 36 informational texts. This procedure generated 650 verbal protocols that were transcribed verbatim, coded, and analyzed.

Results. When text and reader characteristics were considered simultaneously, the generation of both causal and non-causal inferences was significantly predicted by text narrativity, text causal cohesion, relative text difficulty for individual readers, as well as grade level and vocabulary knowledge. Absolute text difficulty (i.e., Lexile), word reading, and working memory were not significant predictors after accounting for other variables. Among the reader characteristics, vocabulary knowledge was the strongest predictor of inference generation.

Scientific or scholarly significance. This study demonstrated that the likelihood that adolescents will generate inferences is dependent on both reader and text characteristics, that inference generation continues to develop through grade 11, and that relative text difficulty for a given reader is more critical than the absolute level of text difficulty (grade-level or Lexile). This has important implications for educators in terms of instruction to support the generation of critical inferences in text.

Authors

• Yusra Ahmed, University of Houston

• Carolyn A. Denton, The University of Texas - Health Science Center at Houston

• Mischa L. Enos, The University of Texas - Health Science Center at Houston

• David J. Francis, University of Houston