What do working memory span tasks like reading span really measure?

Figure 2.1 (a) The all-components model. (b) The text + integration model.

Figure 2.1 (c) The text-only model. (d) The integration-only model.

(p.38) the text-only model, one comparing the text + integration model to the integration-only model, and one comparing the text-only model to the integration-only model. The results of these tests revealed that none of the three models was significantly different from each other, χ² (1) = 1.78, χ² (1) = 1.35, and χ² (0) = 0.43, respectively. In circumstances such as this one, the general rule is to adopt the most parsimonious model(s). If we adopt such a rule, we can conclude that the text-only and integration-only models are better than the text + integration model for describing what is tapped by working memory. All in all, the SEMs supported the zero-order correlations by showing that the two working memory span tests tap the components that draw on the new text information (text memory, text inferencing, knowledge integration) more so than the pure knowledge access components and the speed component.

Working memory span, component processes task and reading comprehension

Our final step was to determine the extent to which the component processes task and the two working memory span tasks, reading span and operation span, made overlapping or independent contributions in their prediction of reading comprehension. To do this, we conducted pairs of stepwise regression analyses: in one analysis, working memory span (either reading span or operation span) was entered as the first predictor into the regression model; in the second analysis, working memory span was entered into the regression model after the variance accounted for by the components of our component processes task was partialed out.⁴ Table 2.4shows the results for the pair of regression analyses using reading span as the measure of working memory capacity. Table 2.5 shows the results for the equivalent pair of regression analyses using operation span as the measure of working memory capacity.

As displayed in Table 2.4(a), when entered first, reading span accounted for 20 per cent of the variance in reading comprehension performance, a finding that is consistent with the reliable correlation between working memory span measures and comprehension (see Daneman and Merikle 1996; Turner and Engle 1989). However, note also that our text inferencing, speed, and high-knowledge integration components accounted for a further 28 per cent of the variance in reading after the effects of reading span were removed. On the other hand, when reading span was entered into the regression equation after the 46 per cent of variance accounted for by the component processes was partialled out, reading span accounted for only an additional 2 per cent of unique variance (Table 2.4b). The picture was very similar when operation span was used as the measure of working memory capacity, although now operation span did not contribute any additional unique variance after the variance accounted for by the component processes was partialed out (see Table 2.5b). Together, these analyses show that the component processes task is accounting for most of the variance in reading comprehension that is tapped by a typical test of the combined processing and storage capacity of working memory. Note that the two working (p.39)

Table 2.4 Regression analyses on reading comprehension scores with reading span and component processes as predictors (n = 206)

Variable	R	R ²	ΔR ²	F
a) Reading span as first predictor of reading comprehension
1 Reading span	0.443	0.196	0.196	49.79
2 Text inferencing	0.573	0.328	0.132	39.98
3 Speed	0.647	0.418	0.09	31.16
4 High-knowledge integration	0.694	0.481	0.063	24.25
(b) Reading span as predictor of reading comprehension after variance accounted for by the component processes has been partialed out
1 Text inferencing	0.517	0.267	0.267	74.19
2 Speed	0.613	0.376	0.109	35.58
3 High-knowledge integration	0.678	0.459	0.083	30.93
4 Reading span	0.694	0.481	0.02	8.48