Given a hypothesis of partially dissociable alphabetic and orthographic skills, two separate (albeit highly correlated) factors were predicted. The first (representing decoding skills) would include nonword reading and spelling; the second (orthographic)
^ We have no reliable spelling data from 1" semester of the T‘ grade, but spelling skills apparently develop rapidly during that period, since they reach mean 70% accuracy in semester 2.
would involve word reading speed and word spelling accuracy. The remaining subskills (nonword reading time, word reading accuracy) should load on both factors.
All aspects of reading turned out to be strongly intercorrelated (table 6-2). This was particularly the case with reading speed (r=.947 between word and nonword tests}* and reading accuracy (r=.824 between word and nonword tests). Word spelling showed moderately strong correlation with various reading indices. However, the test of nonword spelling appeared to be unrelated to other measures, except for word spelling and nonword reading.
Reading- word- accuracy Reading- word-time Reading- nonword- accuracy Reading- nonword time Spelling- word- accuracy Spelling- nonword- accuracy R eading-w ord- accuracy - - . / / / * * * R ead in g-w ord tim e -.778*** - -.461*** R ead in g-n on w ord - accuracy .824*** -.734*** - R ea d in g-n on w ord tim e -.772*** .947*** -.690*** - S p ellin g -w o rd - accuracy .393** -.566*** .483*** - 469*** - Sp eliin g -n o n w o rd - accuracy .150 -.159 .339** -.064 .508***
Table 6-2. Correlations between reading and spelling measures. Time scores were log-transform ed before the analyses. Figures below diagonal express zero-order Pearson correlations. Figures above diagonal are partial correlations, controlling for chronological age, WISC Vocabulary (raw scores), Columbia (raw scores).
* p<.05; ** p<.01; *** p<.000
A factor analysis (Principal Component extraction. Direct Oblimin rotation) showed that the observed pattern of correlations could be reduced to two factors (table 6-3). The first received high loadings from all reading measures, and a weak loading from word spelling. It accounted for 63.4% of overall variance. The second factor encompassed both spelling tests, and explained 20% of overall variance.
^ Descriptive statistics revealed that the distributions of all time-based measures showed strong negative skewness (see appendix). To improve their fit for the subsequent analyses, word and nonword time scores were log-transformed. All correlation, factorial and regression analyses reported in this and following chapter are based on log-transformed time scores.
Factor 1 Factor 2 Reading-nonword-time -.973 Reading-word-time -.949 Reading-word-accuracy .921 Reading-nonword-accuracy .801 Spelling-nonword .971 Spelling-word .351 .673
Table 6-3. Pattern matrix of the factor analysis of reading and spelling measures. Time scores were log- transformed before the analysis. Values represent factor loadings and are sorted by size. Only factor loadings higher than .30 are displayed.
The first (general reading) factor was not decomposable any further. The second factor analysis, carried out on reading measures only (both speed and accuracy), identified just one factor which accounted for 84% of overall variance.
Generally, the results suggest a very strong relationship between different aspects of reading, those which are apparently phonological as well as those which are orthographic. Spelling constitutes a partially separate dimension. The latter conclusion should be treated cautiously, however, as it probably reflects differences in testing procedure.
It is important to add, however, that word spelling accuracy did load on the same factor as reading indices; it is only nonword spelling that truly stood out. That last variable also showed an atypical pattern of development, with scores decreasing between T' and 2"*^ grade level (figure 6-1). Different patterns of correlations for word and nonword spelling cannot be accounted for by differences in testing procedure (since both tasks were carried out during the same group session), but they may reflect low reliability of the nonword spelling task in grades 1 and 3 (see section 5.5.), and its generally low validity as a measure of phonological recoding (see previous section).
Nonword spelling was expected to correlate particularly strongly with nonword reading, since both are considered to be relatively pure measures of recoding skills. The correlation turned out to be significant, but weak (table 6-2), despite the comparable level of difficulty of both tasks (table 6-1). As pointed out before (sections 5.5. and 6.1.1.), a group administered nonword spelling test was not very reliable in grades 1 and 3, and may have generally low validity as a measure of phonological recoding.
In the case of reading, a single factor account of its different aspects and sub components suggests a uniform processing strategy. It is likely to be decoding-based: observation made during testing showed some (more or less overt) sounding out and
blending in most children, especially in grades 1 and 2. That analytic strategy involved single letters or graphemes, but most often occurred in the form of reading syllable-by- syllable. Systematic sounding out and blending behaviour explains extremely slow reading speed, observed in the T* grade. On the other hand, words are processed quicker and much more accurately than nonwords which suggests strong ‘top-down’ lexical influences on decoding (for further discussion, see the following section).
Although correlational and factorial analyses suggest a unitary nature o f reading skills at this early stage, there are also some signs of developmental asynchrony between ‘the phonological’ and ‘the orthographic’. Children seem to improve their decoding accuracy before they go on to make large progress in the automaticity (speed) of decoding and recognition. Longitudinal data, however, are necessary to approach the problem of developmental trajectories more directly.
Relatively low accuracy of nonword processing also suggest that the consistency of the Polish orthography must be rather low after all, since several years are required to master the system fully, even in case of children that (like my sample) showed above- average intelligence. I shall return to this issue in the following section.