Reading Span Measure
5.3.1 Reading Span Measure
Table 7 shows general descriptive statistics for the recall-criterion measure and overall
number of trials correctly recalled 1 . A paired T-test was performed on the recall-
criterion score (left column) to compare the most difficult sentence type (subject-object)
and the easiest (cleft subject) and returned T 48 = 5.03, p < 0.001. This supports the idea
that processing load influences the reading span score, and reinforces the modification of the WCH RST over the DC RST whereby sentence complexity is controlled. A
similar difference was found in the number of trials correct for these subtypes (T 48 =
7.90, p < 0.001) and the RTs for the on-line acceptability judgement: T 47 = 13.96, p <
0.001 (see Table 8 2 ). Figure 12 displays the means graphically.
Table 7: Span Score and Overall Number Correct for the Reading Span Test
Reading Span Score (marked as DC
No. of correctly-recalled sentence-final
RST)
words
Measure Range Mean SD Range Mean SD
Combined 1.00 - 5.50
2.79 1.07 29 – 68 (of possible
CS Span 1.00 - 6.00
3.72 1.66 7 – 20 (of 20)
CO Span 1.00 - 6.00
3.56 1.12 7 – 18 (of 20)
OS Span 1.00 - 6.00
3.29 1.45 5 – 18 (of 20)
SO Span 1.00 - 5.50
2.60 1.06 4 – 17 (of 20)
Note 1: CS (Cleft Subject), CO (Cleft Object), OS (Object-Subject), and SO (Participant-Object)
1 Span test data was prepared for analysis with specially-written Microsoft Visual BASIC programs. 2 One participant’s on-line data was removed because it was corrupted.
Table 8: Acceptability Judgement Reaction Times
Measure Range Mean RT SD
(msec)
Overall 2570.6 - 8043.3
CS Span 2050.1 - 6492.8
CO Span 2589.1 - 7923.0
OS Span 2962.7 - 9363.2
SO Span 2798.0 - 9374.9
CO - Medium
OS - Medium
SO - High
CO - Medium
OS - Medium SO - High
Sentence Type
Complexity
Complexity
Complexity Complexity
Sentence Type
(a)
(b)
7000 t RT 6000
ec rr 5000
n 4000 Co
M ea 3000 2000
CS - Low
CO - Medium
OS - Medium
SO - High
Sentence Type
(c)
Figure 12: (a) Sentence Type and Mean Type-Specific Span, (b) Sentence Type and Mean No. Correct for Sentence-Final World Recall and (c) Sentence Type and Mean Correct Reaction Time for the Acceptability Judgement.
For the thirty-three participants who completed both Phases I and II, mean error on the acceptability judgment was 15 (range: 1.25 – 35; SD: 7.9). This error did not correlate significantly with any measure of span. The recall-criterion measure of span did not correlate with average RT (including both incorrect and correct acceptability
judgements): r 32 = 0.18, p > 0.05. This is consistent with studies that show higher-span
readers do not advantage themselves by simply reading faster (e.g. Waters Caplan, 1996b; cf. Salthouse, 1993).
Correlations between the recall-criterion span measure and the processing-storage Z- score are presented in Table 9. Note that the correlations are negative because the traditional measure is based upon correct recall (the better the recall, the higher the score) whereas the processing-storage score is based upon recall error and processing speed (the better the performance, the lower score). Generally speaking, there is a good overlap between the recall-criterion and processing-storage Z-scores for CS and OS subtypes, but the relationship is poorer for measures SO and CO recall.
Table 9: Correlations between traditional, recalled-based measures of span and composite Z-score measures
SPAN CS OS SO CO ZSPAN ZCS ZOS ZSO ZCO
SPAN
r
1 .463 .465 .405 .229 -.208 -.476 -.348 -.115 -.185
CS r
1 .403 .409 .084 -.055 -.324 -.136 .027 -.046
OS r
1 .132 .325 -.159 -.054 -.399 -.058 -.115
SO r
1 -.031 .109 -.065 .048 -.005 .010
CO r
1 -.132 -.277 -.380 .029 -.006
ZSPAN r
Correlation is significant at the 0.01 level (1-tailed). Correlation is significant at the 0.05 level (1-tailed).
The Z-score in more detail As an aside, it was decided to examine the processing-storage Z-score in more detail.
First, to find evidence for a trade-off between processing resources devoted to the acceptability judgement against processing those for the recall task, a one-tailed correlation was performed between mean processing time (for both incorrect and correct acceptability judgements) and total number of sentence-final words recalled. A trade-off should produce a negative correlation: longer RTs on the acceptability judgements should produce inferior recall and vice versa. A correlation was performed on those First, to find evidence for a trade-off between processing resources devoted to the acceptability judgement against processing those for the recall task, a one-tailed correlation was performed between mean processing time (for both incorrect and correct acceptability judgements) and total number of sentence-final words recalled. A trade-off should produce a negative correlation: longer RTs on the acceptability judgements should produce inferior recall and vice versa. A correlation was performed on those
correlation was in the hypothesized direction, it was not reliable: r 32 = -0.28, p > 0.05.
In this sample, participants did not trade-off resources.
To refine this point, as a second test for the existence of a trade-off, overall percentage incorrect for acceptability judgements were correlated with overall percentage of sentence-final words recalled. This was intended to determine if participants increased their recall performance by becoming more reckless on the acceptability judgement, which would produce a positive correlation. No such
correlation was found (r 32 = 0.081).
Third, the data were examined for evidence that the Z-score would be distorted by its inclusion of incorrect acceptability judgements – i.e. trials where participants did not correctly interpret the stimulus sentence. Mean correct and incorrect acceptability
judgement RTs were correlated: r 32 = 0.90, p < 0.001. Thus, the Z-score is probably not
distorted by the inclusion of incorrect acceptability RTs.
Thus, (i) though the Z-score measure of reading span was designed to accommodate posited trade-offs (Waters, Caplan, Hildebrandt, 1987), in the present study there seems to be no trade-off, and (ii) including RTs for trials where participants incorrectly read sentences will not distort the Z-score.