Evaluation of Reciprocal Teaching

Meta-analysis

In "reciprocal teaching", the teacher models by showing how she or he would try to understand the text, using two or more combinations of four strategies: question generation, summarization, clarification, and prediction of what might occur. Rosenshine& Meister (1994) conducted a meta-analysis Systematic comparing of differerent scientific studies on a topic, often reporting an effect size of the combined results. on 16 reciprocal training studies.  Rosenshine& Meister used the criteria of selection that was adopted by us: Studies had to be an experimental study with controls and use random assignment or matching of conditions. The grade levels studied were 1 through 8, distributed as level 1, n=1; level 2, n=1; level 3, n=4; level 4, n=6; level 5, n=3; level 6, n=4; level 7, n=4; and level 8, n= 1.  The modal grade for reciprocal teaching was grade 4 but high numbers occur for grades 3 through 7 in these studies (4 on average).  Reciprocal teaching using multiple strategies presumes basic reading (decoding) skills, even on those 2 or more grades below level.

The kinds of strategies included varied from 1 to 4 components of Summarization, Question Generation, Clarifying, and Predicting.  Question Generation was most frequent (9 studies), followed by Summarizing (6 studies) and Comprehension Monitoring (3 studies). 

The effect sizes (Rosenshine & Meister, 1994, Table 5, page 194) for Experimenter Tests (10) studies averaged .88 ; for Standardized Tests (9 studies), the average effect size was .32.  These values were about the same for High and Low Quality studies (.88 and .86, respectively for Experimenter Tests; .31 and .36, respectively, for Standardized Tests).  The low quality studies showed the same effect (.87) for Experimenter Tests but a small negative effect (-.12) for Standardized Tests.  Excluding the Low Quality studies, the effect size for Standardized tests was raised to .36 (7 studies).

Effect size varied as a function of reader ability.  The table below summarizes these data.

Effect Size as Function of Reader Ability

(data from Rosenshine & Meister, 1994, Table 7, page 197).

From the table, it can be seen that the magnitude of the effect size for Experimenter Tests was larger for below average or poor readers.  Despite greater efficacy of specific training, Standardized Tests declined as did the ability of the reader.  These data suggest that good readers benefit and generalize what they learn as strategies more than poor or below average readers.

Rosenshine & Meister (1994) tested for the significance of effect sizes and examined their results as a function of grade level, excluding below average readers.  These data are summarized in Table 12.  Their results show that reciprocal teaching of strategies is not significant for grade 3, is mixed for grades 4, 5, and  6, and significant for grades 7 and 8.  Thus, as measured by significant effect sizes, the older readers benefit most from reciprocal teaching.

Effect Size Significance and Grade Level

Reciprocal Teaching Studies not reviewed by Rosenshine& Meister, 1994

The NRP located 11 studies on reciprocal teaching that were not covered in the meta-analysis of Rosenshine & Meister (1994).  These studies covered grade levels from 1 to 6 (level 1 =1, level 2, N= 1; level 3, N=3, level 4, N=3, level 5, N= 3, and level 6, N= 1). These studies tended to use more strategies (7 had combinations of Summarization, Question Generation, Clarification, and Prediction) and added, in one case each, either monitoring or collaborative learning. Four of the studies report improvement on Experimenter Tests and three report significant improvement on Standardize tests.  These data are consistent with those of Rosenshine & Meister (1994).

Summary of Reciprocal Teaching of Multiple Strategies.

There is strong empirical evidence that the direct instruction of more than one strategies in a natural context leads to the acquisition and use of these reading strategies and transfers to standard comprehension tests.