Abstract
Classification of reading difficulties: Cheap screening can be accurate
Purpose: Three factors are important for identification of students in need of remedial instruction: accuracy, timeliness, and cost. The identification has to be accurate to be of any use, the identification has to be timely to allow for optimal remediation, and the procedure should preferably be inexpensive to allow wide-spread adoption. These criteria are, however, often in conflict: Early screening may be less accurate than late screening, and comprehensive and expensive testing is possibly more accurate than simple, inexpensive testing. The present study investigated the classification accuracy of three screening models varying in timeliness and cost.
Method: We compared the ROC statistics of three logistic models for predicting end of Grade 2 reading difficulties in a sample of 164 students: 1) an early, comprehensive model using a battery of Grade 0 tests, including phoneme awareness, rapid naming, and paired associate learning, 2) a late, comprehensive model adding reading measures from January of Grade 1, and 3) a late, inexpensive model using only group-administered reading measures from January of Grade 1.
Results: The late models provided acceptable classification with little difference in classification accuracy between the comprehensive and inexpensive models (AUC = .93 vs .91). The late models provided better classification than the early model (AUC = .82).
Conclusion: Cheap group-administered testing in mid Grade 1 provided good classification. Choice of early intervention based on Grade 0 results should take into account the potential high cost of many false positives.
Purpose: Three factors are important for identification of students in need of remedial instruction: accuracy, timeliness, and cost. The identification has to be accurate to be of any use, the identification has to be timely to allow for optimal remediation, and the procedure should preferably be inexpensive to allow wide-spread adoption. These criteria are, however, often in conflict: Early screening may be less accurate than late screening, and comprehensive and expensive testing is possibly more accurate than simple, inexpensive testing. The present study investigated the classification accuracy of three screening models varying in timeliness and cost.
Method: We compared the ROC statistics of three logistic models for predicting end of Grade 2 reading difficulties in a sample of 164 students: 1) an early, comprehensive model using a battery of Grade 0 tests, including phoneme awareness, rapid naming, and paired associate learning, 2) a late, comprehensive model adding reading measures from January of Grade 1, and 3) a late, inexpensive model using only group-administered reading measures from January of Grade 1.
Results: The late models provided acceptable classification with little difference in classification accuracy between the comprehensive and inexpensive models (AUC = .93 vs .91). The late models provided better classification than the early model (AUC = .82).
Conclusion: Cheap group-administered testing in mid Grade 1 provided good classification. Choice of early intervention based on Grade 0 results should take into account the potential high cost of many false positives.
Originalsprog | Engelsk |
---|---|
Publikationsdato | 11 jul. 2012 |
Status | Udgivet - 11 jul. 2012 |
Begivenhed | Nineteenth meeting of the Society for the Scientific Study of Reading - Montreal, Canada Varighed: 11 jul. 2012 → 14 jul. 2012 |
Konference
Konference | Nineteenth meeting of the Society for the Scientific Study of Reading |
---|---|
Land/Område | Canada |
By | Montreal |
Periode | 11/07/2012 → 14/07/2012 |