| Description |
The purpose of the present research was to investigate text; comprehension of narrative text;s at varying levels of comprehension and examine how metacomprehension varies as a function of the level of comprehension when making retrospective (posttest) confidence judgments of performance. Using Kintsch's construction-integration theory of text; comprehension, three types of question were developed to probe text;base and situation model levels of text; representation at three levels of difficulty: (a) text;base, literal (easiest), (b) situation model, temporal ordering (low difficulty inferences), and (c) situation model, propositional logic (high difficulty inferences). Differences in percent correct, response time in milliseconds per character, and max amplitude of pupil size confirmed the predicted difficulty of the three question types, except that there was no significant difference in pupil size between the literal and temporal ordering questions. The three types of questions were then used to examine the effect of question difficulty on metacomprehension judgments of confidence, absolute accuracy (calibration accuracy and bias), and relative accuracy (Goodman-Kruskal gamma coefficient or G). Results showed that readers were sensitive to different levels of comprehension and showed different levels of metacomprehension confidence and accuracy depending on the type of question. As predicted, absolute accuracy showed the effects of anchoringand-adjustment when making these judgments across question type. That is, subjects appeared to be anchoring on a moderate estimate of success that corresponded most closely in this study to performance on literal questions and adjusted their confidence for temporal ordering and propositional logic questions. The results related to bias provided support for the hard-easy effect, with propositional logic questions (i.e., hard questions) showing overconfidence and literal questions (i.e., easy questions) showing no significant bias, although bias scores did not discriminate between temporal ordering and propositional logic questions. As predicted, relative accuracy (G) appeared to be stable across question types with no significant differences by question type. As with previous studies, the differences in the results concerning absolute versus relative accuracy suggest that the two types of accuracy are measuring different components of metacomprehension. |
