|Institution:||The Ohio State University|
|Department:||ED Policy and Leadership|
|Full text PDF:||http://rave.ohiolink.edu/etdc/view?acc_num=osu1218626731|
The majority of incarcerated youth are unsuccessful in school and many have a significant reading deficit. This study aimed to determine if the Scholastic READ 180 program had a meaningful impact on the reading proficiency of low-achieving incarcerated youth in a large mid-western state, when salient subject covariates were controlled for their influences. The study was based on a longitudinal experimental design in which the eligible youth were randomly assigned to either the READ 180 program or a comparison group being instructed by a traditional reading program. The course of investigation lasted for one school year during which the subjects were measured for their reading proficiency by the Scholastic Reading Inventory (SRI) prior to the treatment and again at the end of each of the four terms. The mixed-effects models were applied to the sample data with a maximum of five repeated measures. Results indicated that subjects exposed to the READ 180 program demonstrated accelerated reading growth over time. In addition, it was found that the variability in the initial reading status of the low-performing incarcerated youth could be attributed to covariates including age, disability status, and another baseline reading test, while a baseline math assessment could account for the variability in a constant growth rate for reading proficiency over time. It also seemed that at higher grade levels, the reading growth of these youth was expected to decelerate over time. Another major focus of the study concerned identifying statistical properties of the probability distribution of the estimate for the READ 180 intervention effect on low-performing incarcerated youth as well as methodologically guiding future longitudinal research with power analysis. Based upon both the bootstrap and the Monte Carlo approaches, it was believed that the intervention effect estimate was modeled quite well using the original sample with sufficient power. Power simulations via the Monte Carlo method indicated that statistical power of detecting true treatment effect is substantially influenced by the magnitude of the effect and the sample size but not the number of points in time, given that highly unbalanced data pattern is expected for similar longitudinal studies.