Explanation of Variables
The tasks that are used to identify the children are at risk for reading difficulties include the Deletion Task (Word Deletion: the number correct when deleting a word from a compound word; Syllable Deletion: the number correct when deleting syllables from a word; Onset-rime deletion: the number correct when deleting an onset, the first consonant cluster, from a rime, the vowel and remaining consonant clusters; Phoneme Deletion: the number correct when deleting phonemes from a word), the Blending Task (Onset-rime Blending: the number correct when blending onsets and rimes into words, Phoneme Blending: the number correct when blending phonemes to form words), thePhonemic Segmentation Task (Phoseg: the number correct when asked to identify the phonemes, in their correct sequence, in a word), the Dynamic Segmentation Task (Dyncor: the number correct when after several trials in which the student has been probed to determined the developmental status of phonemic segmentation, Prompts: the number of prompts necessary before the student is able to respond with the word properly segmented into its phonemes), Letter Identification Task (the Letter Identification subtest of the Woodcock Reading Mastery Test-Revised: the number of letters that the child was able to correctly name), Supple (theSupplementary Letter Identification task; the number of letter sounds that the child knew when presented with letters), Word Identification Task (the Word Identification subtest of the Woodcock Reading Mastery Test-Revised: the number of words that the child was able to correctly name),Word Attack Task (the Word Attack subtest of the Woodcock Reading Mastery Test-Revised: the number of pseudowords that the child was able to correctly name), Colors (the time it took for a child to name the colors of squares arranges in five columns of eight squares).
Each child's performance is examined by looking at the composition of phonological processing and reading measures. We have found that some of the measures are better predictors of reading performance at the end of the year. To adjust for this, some of the variables are given stronger weight in determining at-risk status.
Discriminant analysis was performed on the assessment variables to determine the structure coefficients that were most important in discriminating between the good and poor readers. For the first discriminant function, the largest structure coefficient, the correlation between the variable and the discriminant function, was that for the Letter Identification subtest (.77), followed by Word Identification (.59), the scores from the Peabody Picture Vocabulary Test- Revised (.42), the Supplementary Letter Identification subtest (.42) the Articulation Task (-.39), whether the child was read to (READTO, .37), and deleting the rime during the Onset-Rime Deletion Task (.33). The other structure coefficients were considerably smaller. It is interesting that the letter naming task (LETIDENT) and the task that required children to produce the sound that corresponded with letter names (LETSUPPLE) were so important for the first discriminant function. For the second discriminant function, the PPVT-R was once again important (.52), followed by how often the children were read to (OFTEN, .38), and the Onset-Rime Blending and Phoneme Blending tasks (.33 and .34, respectively). The first canonical variate explained 28.3% of the variance, with a significant canonical correlation of .53, F(44, 438) = 3.08, p < .0001.
The second canonical variate explained 18.7% of the variance with a significant canonical correlation of .43, F(21, 220) = 2.41, p < .0008.
The identification procedure determines at-risk group membership in an involved manner. First, discriminant analysis is performed on the previous year's data to determine the canonical structure coefficients for the variables of Letter Identification, Supplementary Letter Identification, Word Identification, Word Attack, Phonemic Segmentation, Dynamic Segmentation-Correct, Dynamic Segmentation-Prompts, Deletion Tasks, Blending Tasks, and time to complete the Color Naming Task. These are the variables that resulted in the best accuracy in discriminating group membership at the end of first grade based on their assessment scores. The structure coefficients from this analysis are used as weightings for determining at risk status in the following way: the loadings were multiplied by 10 and rounded to remove decimals. A score that was one standard deviation or more below the mean on a variable was given the largest at-risk weight, a score that was between one-half and one standard deviation was given an at-risk weight 20% that of the previous weight, a score that was between the mean and one-half a standard deviation below the mean was given an at-risk weighting 40% less than the original weight. Any score equal to or above the mean was not given an at-risk weighting. The at-risk weightings differed by variable based on its canonical structure coefficient. For example, a variable that had a canonical structure coefficient of .92 would have weightings of 9 (.92 x 10 rounded to 9), 7 (9 x .80 rounded to 7), and 5 (9 x .60 rounded to 5) for a score 1 sd or greater below the mean, between 1 sd and .5 sd, and between the mean and .5 sd below the mean, respectively. A score that had a canonical structure coefficient of .69 would have weightings of 7, 6, and 4, etc. (see Table 6). The at-risk status of a student was then determined by summing the various weightings for each of the variables. The individual was identified as at risk if his/her at-risk status score was equal to or greater than 1 sd (13.96) above the mean (18.77) for at risk status.
The identification procedure resulted in highly accurate group membership placement (89.7% accuracy). The identification procedure is not difficult to use. If the same weighting scheme is used, one would only need to determine at risk status scores for each student, determine the at risk criterion of 1 sd above the mean on a the criterion variable and assign at risk group membership accordingly. The number of at-risk students identified could be modified by adjusting the criterion such that the desired percentage of students were identified. This would be useful when a school system has a larger at risk pool than could be accommodated in special services.
The identification procedure yielded an accuracy rate of 89.7% overall with a false negative rate of 6.2%. This procedure used the loadings from discriminant analysis to determine a weighting system to create an at-risk variable, which was then used to identify children at risk for reading difficulties. When the calibration data from the previous year's discriminant analysis were used to identify at-risk status, the accuracy fell to 80.2% with a 10.2% false negative rate. The major concern was the false negative rate since these individuals will eventually become poor readers at the end of first grade, are not being identified, and thus are not provided the opportunity to engage in training. Training procedures have been effective in increasing the phonological processing and reading abilities of young children and have also been effective in increasing these same abilities in children at risk for reading difficulties. The crucial component for the latter group of children is that they are accurately identified so that they can be exposed to the training procedure.
Although an accuracy of identification rate of 89.7% with a 6.2% false negative rate is quite good, there is still a false negative rate. It would be ideal if there were no false negatives. If this were the case then all of the children who are at risk and therefore likely to be poor readers in the future would be identified. Unfortunately, it may not be possible to have a 0 false negative rate.