The comparison children had access to the entire library collection, and did 20 minutes per day of SSR. They had the "regular third grade curriculum, book talks, and read-aloud sessions in the media center" (p. 36). Facemore does not mention whether the AR group also had the regular curriculum.

Comparison subjects gained about three months on the STAR reading comprehension test. AR students gained more, about five months over the three month interval. The difference was statistically significant (t = 2.145, df = 14, comparing differences between gain scores). This is by far the most impressive result in favor of AR in the research literature.

Note, however, that the number of students in the study was small. With small samples, behavior of a few subjects can influence scores a great deal. The AR group, for example, contained one child who gained 2.3 years in nine weeks! If we remove this "outlier" the AR mean gain drops to about four months The comparison group had one child who got much worse, dropping more than one year (1.1) in nine weeks. If we remove this outlier, the comparison mean increases to three and a half months. Some researchers would discard these subjects because they fall so far away from the group mean. (Facemore, in fact, discarded two students in the AR class who could not pass any of the AR tests and "had chosen not to participate in any way in the AR program" (p. 32).) A more prudent policy, in my opinion, is to include analyses with and without outliers. Excluding the outliers results in practically no difference between the groups.

Finally, it is likely that the AR students read more than the comparison subjects. They had "at least" 20 minutes per day of SSR, while comparisons had exactly 20 minutes. While comparions had "access" to the library, AR students had a regularly scheduled 80 minutes per week.

This study is an important step in the right direction, in my opinion. There is a comparison group, and an attempt is made to note the amount of reading done. More precise data on the amount of actual access to books and reading time would be helpful, as would a larger sample size.

Table 11: Comparisons of AR with Control Group that did Recreational Reading

What can we conclude? In agreement with McQuillan (1997), I conclude that the research provides inconsistent evidence in support of incentive-based reading management programs. In a review of 14 reports in which standardized tests were used to determine the effectiveness of incentive-based systems (table 7), in all cases except one (Idaho study), gains were not present or small, limited to a subgroup of students, inconsistent with other data on the same students, or only a subgroup of scores were reported. Of three studies in which AR is compared to traditionally taught comparison groups (table 8), a clear advantage for experimental students was found in only one study (Peak and Dewalt, 1994).

Even if the results of all of these studies had favored students in incentive-programs, we would not be able to determine why the program worked. To do this requires comparison with a group that also engaged in recreational reading and had plenty of access to books, but did not have tests and rewards. Three studies were found that claimed to compare AR with such comparison groups. In two cases, the AR students actually had more exposure to comprehensible text. Also, in these studies, the comparisons were not in a purely recreational reading program; in one case they wrote book reports, in another they also had incentives. AR did very well in a third study, but the sample size was small, the duration short, there were two serious outliers in the sample, and it is likely that AR students did more reading.

12