1989). Nevertheless, oral passage reading most directly requires the earlier component skills proposed by developmental reading theorists: decoding and fluency. According to theory, greater growth on tests directly requiring decoding and fluency should be, and indeed was, manifest as earlier reading stages in the earlier grades.
By contrast, the CBM maze task may more directly require not only decoding and fluency, but also comprehension: To score well on the maze (or to index improvement over time), students must decode text, proceed fluently, and understand the content for successful blank restoration. Developmental theory would predict that this more comprehensive set of component skills required by the maze would result in more similar rates of growth over the grades. This predicted pattern actually was demonstrated in the current study. Our findings are related to and supported by those reported and discussed by (a) Jenkins and Jewell (1990) in their correlational comparison of CBM measures of oral passage reading and maze and (b) Shinn and colleagues (1992) in their construct validity analysis of reading measures.
Data for spelling correct letter sequences and words are shown for the common and the graded measures: Year 1 results appear in Table 3; Year 2 results, in Table 4. The following statistics are presented by grade level: (a) average slopes and standard deviations, (b) percentages of individual student regressions for which the quadratic term significantly contributed to the modeling of student progress, (c) information about the distribution of the slopes across students, and (d) percentages of students with negative slopes. In footnotes to these tables, information also is presented about results of ANOVA, testing for the effect of grade level on slope.
How well does a linear relationship model student progress within 1 academic year? Preliminary analyses, assessing the extent to which a linear relationship adequately modeled student progress within one academic year, revealed that, for the Year 2 graded measure the quadratic term significantly contributed to the modeling of student growth for 2-13% of the individuals when LS was the score and for 4-21% of the students when correct words was the score (see Table 4 for percentages by grade). For the common measure in Year 2, percentages ranged between 4 and 16 both for LS and words correct (see Table 4). As with the reading slopes, for almost all cases, these curvilinear relationships indicated slightly negatively accelerating patterns across one academic year. A linear relationship contributed significantly to the modeling of student progress for the following percentages at Grades 2-6, respectively 70, 50, 24, 28, and 48 for the graded LS score; 67, 25, 32, 19, and 19 for the grades words score; 89, 83, 66, 43, and 45 for the common LS score; and 59, 41, 20, 13, and 21 for the common words score.
How well does the normal distribution characterize the distribution of slopes! Distributions of the Year 1 slopes appeared normal for approximately 40% of the grades (i.e., kurtosis and skewness fell within 2 standard errors). Only 20-40% of the Year 2 distributions appeared to conform to normality (see Table 4). In all cases, when kurtosis exceeded 2 standard errors, the distribution was lepokurtic (i.e., more peaked than a normal distribution). With respect to skewness, distributions that exceeded 2 standard errors were, in all but two cases, positively skewed with a few extreme positive slopes. Despite that many spelling slope distributions failed to conform to normal distributions, distortion from normality, both in terms of kurtosis and