Measures that Hold Promise for Screening
In this section, I review measures that have been used for screening and/or that hold promise as candidates for screening, based on evidence from classification or criterion validity studies. I confine my review to child performance assessments, leaving out assessments that are based on teachers' ratings of children.
Tables 2-5 identify several candidate-screening measures and the evidence that supports them. Measures are shown according to the grade level where they pertain: Early and mid-kindergarten; late kindergarten; early grade 1; late grade 1 and early grade 2.
Early- and Mid-Kindergarten Screening Measures
Table 2 shows a selection of measures used in early and mid-kindergarten to predict poor reading outcomes at end-of-grade 1. These measures may be useful to schools that provide early intervention during the kindergarten year.
Table 2. Early and Mid-Kindergarten Screens
| Measure/Study | Sample | Type of Evidence | Result | |||
| Letter Identification (Scanlon & Vellutino, 1996) |
1407 Early-Mid K | Classification | At Risk 10% |
Sensitivity 32% |
Specificity 95% |
Criterion Severe reading difficulty grade 1 (teacher-identified) |
| Letter Identification (Scanlon & Vellutino, 1996) |
1407 Early-Mid K | Classification | At Risk 35% |
Sensitivity 75% |
Specificity 75% |
Criterion Severe reading difficulty grade 1 (teacher-identified) |
| Combination of: Phoneme Segmentation Letter Naming Fluency Syllable Elision O'Connor and Jenkins (1999) |
129 Nov. K | Classification | At Risk 18% |
Sensitivity 100% |
Specificity 88% |
Criterion Below 8th percentile WRMT Basic Rd. Scale Grade 1 |
| Same combination with revised cut-scores | 101 Nov. K | Classification | At Risk 20% |
Sensitivity 100% |
Specificity 86% |
Criterion Same |
| Same combination with revised cut-scores | 215 Nov. K | At Risk 18% |
Sensitivity 91% |
Specificity 86% |
Criterion Same |
|
| TRPI Combination of: Letter Sound Identification Blend Onset-Rimes and Phonemes Foorman et al., 1998 |
421 December K | Classification | At Risk 56% |
Sensitivity 95% |
Specificity 56% |
Criterion Below 23rd percentile WJ Broad Reading Spring Grade 1 |
| DIBELS-OnRF Good et al (2001) |
353 Winter K | Concurent Validity | .36 WJ-R Reading Readiness Cluster | |||
| 378 Winter K | Predictive Validity (Mid K - Spring Grade 1) |
.36 WJ-R Reading Cluster .45 CBM-ORF |
||||
The table illustrates several of the study differences that I have already mentioned. First, most researchers used some version of the Woodcock end-of-grade tests for the grade 1 criterion, but they focused on different subtests and scales (e.g., WJ-R Readiness Cluster, Letter-Word ID, Broad Reading; WRMT-R Basic Reading Cluster). Second, the three classification studies (Foorman et al, 1998; O'Connor & Jenkins, 1999) employed different criterion performance levels (i.e., unsatisfactory reading--performance below the 23 percentile; very unsatisfactory reading--performance below the 8th percentile; teacher-judgment of severe reading difficulties). This in turn affected the percent of kindergartners identified as at risk by the researchers (56% -18%).
Third, all the kindergarten measures assessed some aspect of phonological awareness (e.g., blending onset-rimes) and/or letter knowledge (e.g., naming letters and/or their sounds). Fourth, some kindergarten measures assessed knowledge (e.g., ability to name letters), others assessed fluency (e.g., speed with which letters are named).
Fifth, two classification studies ascertained that a multiple assessments in combination worked better than single assessments (e.g. letter naming) as screens. Specifically, Foorman, et al., (1998) and O'Connor & Jenkins (1999) both reported that screening with a combination of phonological awareness and letter knowledge measures produced high sensitivity (i.e., the screen correctly identified 95% to 100% of the students who later performed at an unsatisfactory level on the criterion measure). Results for specificity were also positive, but more variable (the screen correctly identified 56% - 87% of those who later performed at a satisfactory level on the criterion measure).
Sixth, two classification studies (Foorman et al., 1998; O'Conor & Jenkins, 1999) identified screening cut-points in a post-hoc fashion, selecting cut score on the screen to minimize false negatives, and one (Scanlon & Vellutino, 1996) identified several screening cut-points and their classification accuracy. O'Connor and Jenkins demonstrated the importance of cross-validation to ensure that cut-points appropriate for one sample apply as well to other samples. Foorman et al., selected cut-points based on statistically estimated generalizability.
What is not shown in Table 2 are the measures that the researchers eliminated as screening candidates. For example, before arriving at their set of screening measures O'Connor and Jenkins (1999) eliminated PPVT-R, rhyming, blending syllables, segmenting syllables, blending phonemes, and isolating the first sound in spoken words because these measures did not discriminate risk status as well as those that were selected for the screen. Likewise, in arriving at their set of screening measures Foorman et al. (1998) eliminated phoneme blending, comparing first sounds in spoken words, phoneme elision, sound categorization, phoneme segmentation, phonological memory, PPVT-R, rapid naming, visual-motor integration, and visual perceptual matching. It is interesting that some measures eliminated by one research group (e.g. phonological memory eliminated by Forman et al.) strongly resembled those chosen by another research group (sound repetition chosen by O'Connor and Jenkins, 1999).
Previous Page | Next Page
(Combined Measures) | (End of Kindergarten)
