Prior Knowledge and Reading Comprehension Test Bias

Johnston, Peter H.

Permalink

https://hdl.handle.net/2142/68747 Copy

Description

Title

Prior Knowledge and Reading Comprehension Test Bias

Author(s)

Johnston, Peter H.

Issue Date

1981

Department of Study

Education

Discipline

Education

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

Education, Curriculum and Instruction

Abstract

• This dissertation examined bias, especially that due to prior knowledge, in tests of reading comprehension. A pilot study demonstrated that prior knowledge has a strong and measurable effect on comprehension from expository text, even when measured with multiple choice questions. This was demonstrated in two ways, by (a) measuring prior knowledge with passage independent questions, and (b) creating an unfamiliar text by substituting pseudowords for certain familiar words in the text.
• The main study had two main thrusts. One thrust was a within-subject examination of the mechanisms through which prior knowledge might influence reading comprehension test performance, and the second was an examination of ways to remove test bias. The mechanisms investigated were: decoding speed, short-term memory capacity, and long-term memory.
• Prior knowledge was assessed with content-specific vocabulary questions. Short-term memory capacity was measured by means of a listening task similar to that used by Jarvella (1971). Speed and accuracy of decoding content-related word lists was recorded. Reading comprehension was assessed with a text, followed by 18 multiple-choice questions. These 18 questions were composed of six of each of Pearson and Johnson's (1978) question types; textually explicit, textually implicit, and scriptally implicit, half of each addressing central information, half addressing peripheral information.
• Each of the above tasks was replicated three times for each subject, once in each of three content areas. Long-term memory demands upon the reader were manipulated by (a) allowing the reader to refer back to the text while answering the questions (low demand), (b) preventing the reader such text access (moderate demand), and (c) preventing text access and delaying question answering (high demand).
• It was found that while prior knowledge influenced decoding speed and reading comprehension, and mean short-term memory capacity influenced mean reading comprehension, within subjects no direct effects of short-term memory or decoding speed on reading comprehension were observed. The extent to which readers were dependent on long-term memory influenced reading comprehension test performance, and there were indications that this might be moderated by the extent of the reader's prior knowledge.
• Not only were quantitative effects of prior knowledge on reading comprehension demonstrated, but qualitative effects were also demonstrable through an examination of performance on different question types. The availability of the text during question answering was found to exert a powerful influence on performance on certain question types. Peripheral textual items were most sensitive to such influence, central items and scriptal items were least sensitive to such influence. It was noted that performance on central questions actually improved when readers could not refer back to the text.
• The second thrust of the study involved the problems of removing bias from tests of reading comprehension. This was investigated by administering the tests to two subpopulations of eighth graders, one urban, and one rural.
• Bias is considered to be at the individual level, and thus should only be removed at that level, not at the population level. Using the specific vocabulary test, this was readily accomplished for bias due to prior knowledge, and incidentally resulted in a decrease in the bias due to intelligence. A conventional approach to bias removal (collapsing across several text content areas) also removed the bias due to prior knowledge, but at the same time it increased the bias due to intelligence. This latter bias was also found to be increased when readers were able to refer back to the text while answering the questions.
• Results are interpreted to suggest quite drastic modifications of our current reading comprehensions tests and our methods of dealing with bias.

Graduation Semester

1981

Type of Resource

text

Permalink

http://hdl.handle.net/2142/68747

Owning Collections