Diagnostic accuracy of the Dynamic Indicators of Basic Early Literacy Skills in the prediction of first-grade oral reading fluency

Research in the area of beginning reading has given educators both, the knowledge of the critical foundational skills that comprise reading, and the tools to assess such skills early to prevent the development of reading problems. The Dynamic Indicators of Basic Early Literacy Skills (DIBELS) are a series of brief measures that can be used to identify children who are at risk of developing reading problems as soon as they enter school. In this era of high stakes testing and accountability, educators must ensure that students are on their way to become proficient readers, well in advance of third grade when standardized tests are typically administered. In the interest of prevention and early intervention, authors of the DIBELS provide a timeline and recommended benchmarks to guide instruction and intervention. This study examines the diagnostic accuracy of DIBELS to predict oral reading fluency using author recommended cut-scores and alternative cut-scores identified as a result of Receiver Operating Characteristic (ROC) analysis. The accuracy of the DIBELS was assessed across the range of all possible cut-scores in an effort to maximize desirable test characteristics such as sensitivity, specificity, predictive power, or more broadly, decision validity. A sample of 122 students were administered the DIBELS measures in kindergarten and the middle of first grade, followed by oral reading fluency at the end of first grade. Analysis of decision accuracy indicated that the DIBELS measures are highly sensitive in identifying students who are at risk of developing reading problems; however, this occurred at the expense of an inordinate number of false positives. This has important implications for the utility of the DIBELS as a decision-making tool. In an effort to maximize the accuracy of the DIBELS, ROC curves were generated and alternative cut-scores were identified which improved specificity, predictive power, and the percentage of correct classifications