Notes on Reading Proficiency and Visual Report Tasks

An Analysis of Reading Proficiency and Its Relationship to Complete and Partial Report Performance

Introduction and Background

• Study Authors: Joseph G.R. Martinez and Peder J. Johnson.

• Source: Reading Research Quarterly, Vol. 18, No. 1 (Autumn, 1982), pp. 105-122.

• Conceptual Framework: Rooted in hierarchical information processing models of reading (e.g., LaBerge & Samuels, 1974), which posit that elementary processing components are integral to the global act of reading.

• Encoding Stage: A central component where the visual code on the page connects with a memory representation, leading to sensory input identification.

  • Hypothesis: Variability in reading ability may be partly explained by individual differences in the speed with which memory codes are accessed by visual inputs.

• Previous Research Supporting Encoding Speed:

  • Jackson & McClelland (1975, 1979): Reported encoding speed related to adult reading performance. Their complete report task showed superior readers reporting more unrelated letters, hypothesizing faster encoding. A subsequent study supported this with faster responses on a same-different letter matching task, concluding letter encoding speed was a common factor.

  • Morrison, Giordani, & Nagy (1977): Used Sperling's (1960) partial report procedure in a delay-of-marker task with sixth-grade children.

    • Delay-of-Marker Task: Subjects see a stimulus array, then after a delay, a marker indicates one element to report. Performance typically declines in the first $200-300\ \text{msec}$, then stabilizes (baseline).

    • Morrison et al. Findings: Average readers performed better than poor readers with marker delays between $250-400\ \text{msec}$. They concluded this difference was due to rate of nonselective readout or short-term memory capacity.

• Current Study's Aim: To further clarify the nature of relationships between letter match reaction time (RT), complete report performance, partial report performance on delay-of-marker, and reading proficiency, all potentially linked by encoding speed. It aims to investigate encoding speed in the context of iconic memory and its relation to adult reading performance.

Experimental Goals

1. More Direct Measure of Nonselective Readout: Address limitations of the delay-of-marker task (memory confound, optional strategy) by adopting Averbach and Coriell's (1961) procedure, which substitutes a masking stimulus for the marker. This forces nonselective readout as the only encoding means.

   • Expected Performance: Chance level at $0\ \text{msec}$ mask delay, improving to an asymptotic level comparable to baseline in delay-of-marker.

2. Investigation of Selective Readout: Employed Averbach and Coriell's (1961) selective readout partial report procedure where the marker is simultaneous with the array, and the mask follows after a variable delay. This measures the time available for encoding a target.

   • Expected Performance: Improves with increased mask delay, asymptoting when encoding is sufficient. Expected selective and nonselective readout to offer converging measures of encoding speed.

3. Replication of Delay-of-Marker for Adults: Determine if Morrison et al.'s (1977) findings with children replicate in average and superior adult readers.

4. Relationship Among Partial Report Tasks: Specifically, if delay-of-marker baseline measures nonselective readout, it should correlate with delay-of-mask performance.

5. Complete Report Task: Similar to Jackson and McClelland (1975), to see if it, nonselective readout rate, and reading performance are related by a common encoding speed factor.

6. Vowel Inclusion: Investigate if the inclusion of vowels in 8-letter strings interacts with reading ability, expecting superior readers to benefit more.

Methodology

• Reading Proficiency Definition: Defined as Effective Reading Rate (ERR), calculated as (percentage comprehension score) $\times$ (words read per minute). This is identical to Jackson and McClelland (1975, 1979).

• Subjects and Reading Test: Eight average and eight superior adult readers from the University of New Mexico (total 42 volunteers screened).

  • Test Material: Isaac Asimov's (1975) "The Trojan Horse" (4,286 words) on asteroids, chosen for probable unfamiliarity.

  • Instructions: Read as fast as possible while maintaining good comprehension, followed by a 10 short-answer question comprehension test.

  • Selection Criteria: Minimum $70\%$ comprehension.

    • Average Group: Reading rate between $200$ and $300\ \text{WPM}$.

    • Superior Group: Minimum reading rate of $400\ \text{WPM}$.

• Apparatus and Materials:

  • Tachistoscope: Three-channel, Model GB, Sylvania F8T5/CWX fluorescent bulbs, constant $35\ \text{foot-lamberts}$ illumination.

  • Visual Field: $4.0$ degrees vertically, $5.0$ degrees horizontally at $110\ \text{cm}$ viewing distance. Central dark dot for fixation.

  • Stimuli: Upper-case black letters stenciled on $7.5\ \text{x}\ 10$ inch ( $19.05\ \text{x}\ 25.40\ \text{cm}$) white cards.

    • Letter Size: $6\ \text{mm}$ ($.234\ \text{inches}$) height, $4\ \text{mm}$ ($.156\ \text{inches}$) width.

  • Complete Report Stimuli: 8 randomly selected unrelated letters, presented left to right. $25$ strings total, with roughly equal numbers of $0, 1, 2+$ vowels. $5$ practice, $20$ experimental ( $5$ with $0$ vowels, $8$ with $1$, $7$ with $2+$).

    • Visual Angle (1x8 array): $.3$ degrees height, $2.3$ degrees width.

  • Partial Report Stimuli: Similar construction but $2\ \text{x}\ 8$ arrangements of letters.

    • Visual Angle (2x8 array): $.8$ degrees height, $2.3$ degrees width.

    • $128$ unique stimuli for each of the three partial report tasks.

• Procedure and Design:

  • All subjects underwent complete report and three partial report tasks (delay-of-marker, delay-of-mask, selective readout) in random order across three sessions (with several days separation).

  • Trials began with a black dot fixation, initiated by subject button press.

  • Complete Report: Stimuli exposed for $50\ \text{msec}$. Subjects wrote all seen letters in any order, including repetitions. $5$ practice, $20$ experimental trials.

  • Delay-of-Marker: $2\ \text{x}\ 8$ array presented for $50\ \text{msec}$, followed by a bar marker after variable delays ($0, 50, 100, 200, 300, 400\ \text{msec}$). Subjects verbalized the marked letter. $32$ practice trials, then experimental trials (all combinations of $6$ delay intervals and $16$ locations).

  • Delay-of-Mask: Identical to delay-of-marker, but a black circular grid stimulus replaced the bar marker. This forced nonselective readout.

  • Selective Readout: Bar marker occurred simultaneously with the $2\ \text{x}\ 8$ array, and the grid mask occurred after a variable delay. This measures speed of locating and reading a single item.

  • All three partial report tasks used the same $6$ delay intervals and number of trials.

Results and Discussion

Reading Performance

• Average Group: Mean $269.6\ \text{WPM}$ with $80\%$ comprehension; ERR = $215.7$.

• Superior Group: Mean $461\ \text{WPM}$ with $83\%$ comprehension; ERR = $383.9$.

• Comparison: Average group ERR ($215.7$) was close to Jackson and McClelland's (1975) average group ERR ($207$). However, the current study's superior group ERR ($383.9$) was lower than Jackson and McClelland's fast group ERR ($490$).

Complete Report Performance

• Group Difference: Highly skilled readers reported a significantly higher percentage of letters (F(1, 12) = 13.36, p < .01). This replicates Jackson and McClelland's (1975) findings (fast readers ~$64\%$ correct, average readers ~$53\%$ correct).

• Vowel Effect: Inclusion of vowels in letter strings facilitated performance (F(2, 28) = 24.82, p < .01).

• Interaction: No interaction between reading groups and number of vowels (F < 1.0). The hypothesis that skilled readers would benefit more from vowels was not supported.

Delay-of-Mask Performance

• Group Difference: Superior readers performed significantly better (F(1, 14) = 20.1, p < .01).

• Interaction: No interaction between group and delay-of-mask (F(5, 70) = 1.13, p > .10).

• Performance Curve: Performance was nearly identical at $0\ \text{msec}$ delay (mask presented immediately), indicating equal mask effectiveness for both groups. Performance improved with initial increases in mask delay and then stabilized.

• Pre-asymptotic Performance: An analysis of performance from $0$ to $100\ \text{msec}$ delay showed a marginally significant interaction (F(2, 28) = 2.58, p < .10). Simple effects analysis revealed group differences at the $50\ \text{msec}$ delay (F(1, 14) = 9.37, p < .05, Bonferroni adjusted).

• Interpretation: The difference prior to asymptote suggests that superior readers had faster nonselective readout rates, allowing them to encode more items within the first $100\ \text{msec}$. This performance partly depends on processing item location, which Mason (1980) showed skilled readers are better at.

Selective Readout Performance

• Delay Effect: Performance improved as a function of mask delay (F(5, 70) = 27.5, p < .01).

• No Group Difference: No significant difference between groups (F(1, 14) = 2.31, p > .10).

• No Interaction: No interaction between groups and delay (F < 1.0).

• Unexpected Findings: The absence of a group effect was surprising, given findings in nonselective readout and Mason's (1980) emphasis on location information. The task is assumed to relate to the speed of locating and reading a single item. This data did not support Mason's thesis regarding reading ability and location information processing.

Delay-of-Marker Performance

• Replication of Morrison et al. (1977): This task was a replication of the prior study with children.

• Pre-baseline Performance (0-100 msec): A significant effect of delay (F(3, 42) = 56.1, p < .01). No significant group effect (F < 1.0) or group x delay interaction (F < 1.0).

  • This is consistent with Morrison et al.'s failure to find differences in the decline of performance reflecting iconic representation decay.

• Baseline Performance (200-400 msec): No significant main effects or interactions (all Fs < 1.0).

  • Contradiction: The failure to find a group effect contradicts Morrison et al. (1977), who found average sixth-graders superior to below-average sixth-graders.

  • Reasoning: The authors argue against the M-S (Morrison et al.) study's longer delay as a reason for non-replication, stating performance was at baseline by $200\ \text{msec}$, and M-S's group differences occurred within $300\ \text{msec}$. They conclude different factors might distinguish these reader populations.

  • Validity Questioned: This task's validity as an index of readout speed is questioned, as nonselective readout is an optional strategy, potentially leading to trial-to-trial variability.

Correlational Analysis

• ERR and Speed/Comprehension: ERR was primarily related to speed ($r = .95$) and less strongly to comprehension ($r = .63$). The comprehension criterion (>70\%) might have restricted its variability.

• Predictors of ERR: Only complete report (CR) and delay-of-mask (Mask) significantly correlated with ERR.

  • CR-ERR Correlation: The correlation between ERR and complete report ($r = .82$ for consonant strings) was surprisingly large compared to Jackson and McClelland's (1975) ($r = .48$). This was not due to vowel inclusion.

  • Delay-of-Mask-ERR Correlation: Delay-of-mask (50-400 msec) correlated significantly ($r = .76$) with ERR, accounting for $58\%$ of the variance. The $0\ \text{msec}$ SOA condition was omitted as it tests mask effectiveness, not reading performance.

• Shared and Unique Variance:

  • CR and Mask correlated ($r = .63$), indicating shared factors related to ERR.

  • Partial correlations showed unique variance: CR-ERR with Mask partialled out ($r = .65$); Mask-ERR with CR partialled out ($r = .52$).

• Relationship to Comprehension:

  • CR correlated significantly with comprehension ($r = .55$).

  • Mask correlated non-significantly with comprehension ($r = .29$), suggesting it's more specifically related to the speed component of reading.

  • Selective Readout (0-100 msec) showed a similar pattern: correlated with CR ($r = .48$) and Mask ($r = .49$), but unrelated to comprehension ($r = -.13$).

• Encoding Speed and Short-Term Memory: Complete report, delay-of-mask, and selective readout (0-100 msec) are conjectured to involve an encoding speed component, correlating primarily with reading speed. They show a decreasing relationship with comprehension, corresponding to a decreasing involvement of short-term memory, consistent with theories linking working memory and comprehension (Just & Carpenter, 1980).

• Speed and Comprehension: A positive relationship between speed and comprehension was generally found, though the authors do not claim they are necessarily unrelated. Slow encoding might indirectly interfere with comprehension by demanding more attention (models like Just & Carpenter, 1980; LaBerge & Samuels, 1974).

• Delay-of-Marker Validity: Baseline delay-of-marker (200-400 msec) did not correlate significantly with delay-of-mask ($r = -.36$), nor with ERR or CR performance. This suggests these two tasks are not measuring the same rate of nonselective readout, questioning delay-of-marker's validity as a readout speed index, likely due to nonselective readout being an optional strategy.

Stepwise Regression Analysis of ERR

• Predictors: Complete report performance on trials with $0$ consonants (CR (0)) and delay-of-mask (50-400 msec) were significant predictors.

  • Step 1: CR (0) accounted for $67\%$ of the variance in ERR ($R^2 = .67$).

  • Step 2: Delay-of-mask accounted for an additional $13\%$ of the residual variance. Together, these two variables accounted for an impressive $80\%$ of the variance in ERR ($R^2 = .80$).

• Limitations: The high magnitude of this relationship is likely inflated by the use of an extreme groups experimental design. Further work with more representative samples is needed.

Implications

Implications for Reading Theory

• Encoding Time as a Predictor: This study, along with Jackson and McClelland (1975, 1979), strongly supports encoding time as a reliable predictor of reading proficiency.

• Encoding Time and Comprehension:

  • Is encoding time only related to speed, not comprehension? While a low positive correlation between speed and comprehension was found for average/above-average readers, slower encoding, requiring more attentional resources, could indirectly cause lower comprehension.

  • The link between encoding time and comprehension might be more evident in below-average and problem readers.

• Encoding Stage Validity: The findings, using partial report, provide additional converging evidence that encoding speed is a valid component of information processing, measurable across various tasks, and consistently relates to reading proficiency.

• Nature of Encoding Stage:

  • Hunt, Davidson, and Lansman (Note 1, 1980) showed consistency in subjects' performance across different encoding time measures, indicating it's a general ability partially independent of material type.

  • Jackson (1980) reported superior readers maintained an advantage in matching tasks with unfamiliar characters, suggesting encoding time differences are not due to differential experience.

• Causal Relationship: These findings suggest encoding time is an information processing component that causally influences reading proficiency.

Applied Implications

• Identification of Slow Encoders: Encoding time metrics could be useful in identifying readers with extraordinarily slow encoding times.

  • Diagnostic Value: While not directly leading to remedial procedures, it's a necessary first step for diagnosis and could help evaluate the effectiveness of certain remedial methods, especially for problem readers.

• Future Research: More research is needed on encoding time's role in below-average and younger readers.

• Modifying Encoding Time: Techniques aimed at modifying encoding time should be explored, but first, the specific locus of performance differences and the causal relationship between encoding time and reading must be clarified.

  • Measurement Caveats: Encoding time is based on reaction time differences in identifying briefly presented stimuli. Poor performance might reflect an overly conservative response criterion, requiring analysis of speed-accuracy trade-offs (Pachella, 1974).
    The adjustment for guessing in certain tasks is given by the equation: $P<em>o = P</em>p + (1 - P<em>p) (1 /26)$, where $P</em>o$ is observed percentage correct, and $P<em>p$ is estimated percentage actually perceived after correcting for guessing. This means $P</em>p$ is solved for when $P_o$ is given. This adjustment was used by Averbach and Sperling (1961).