Linking visual span and reading speed in dyslexia

Abstract

This study examines visual processing in dyslexia using a psychophysical measure linked to reading speed, the visual span, defined as the number of letters reliably identified within a single fixation. We asked whether reduced visual span is associated with slower reading in dyslexia. In a laboratory study, adults with dyslexia and typical readers completed standardized assessments, reading speed, and visual span measurements. Individuals with dyslexia showed significantly slower reading speeds and narrower visual span profiles than typical readers. Visual span size was positively correlated with reading speed across participants. The paradigm was then adapted for online testing. In a validation study with typical readers, online visual span and reading speed test showed patterns consistent with laboratory patterns, including a strong positive relationship between visual span size and reading speed. Applying this paradigm to a larger online sample of individuals with self-reported dyslexia revealed markedly slower reading speeds, reduced visual span sizes, and a robust correlation between visual span and reading speed despite substantial individual variability. Together, these findings provide the first direct evidence that reduced visual span is a reliable characteristic of dyslexia and is closely linked to reading performance. Moreover, they demonstrate that a brief, psychophysically grounded visual span assessment can be implemented reliably online, enabling scalable investigation of visual factors in dyslexia across diverse populations.

Introduction

Dyslexia is a specific learning disorder of neurobiological origin characterized by difficulties with accurate and/or fluent word recognition, poor decoding, and spelling difficulties, despite otherwise typical intelligence and adequate educational opportunity (American Psychiatric Association, 2013). Although dyslexia is often identified using standardized diagnostic criteria, reading ability itself lies on a behavioral continuum, making categorical diagnostic boundaries inherently imperfect (Siegel, 2006). Contemporary accounts increasingly view dyslexia as a heterogeneous, multifactorial disorder arising from interacting risk factors across genetic, neural, cognitive, and environmental levels, rather than from a single underlying deficit (Peterson & Pennington, 2015; O'Brien & Yeatman, 2021).

Among the most well-established cognitive contributors to dyslexia are impairments in phonological processing (Ramus, 2003; Vellutino, Fletcher, Snowling, & Scanlon, 2004). At the same time, evidence for alternative or additional contributing mechanisms does not necessarily conflict with phonological accounts (Share, 2021). In particular, the role of visual processing in dyslexia remains an active area of investigation. Because reading is fundamentally a visually mediated task, disruptions at visual or visual-attentional stages may plausibly constrain reading performance, independent of or interacting with phonological processes (Stein & Walsh, 1997; Vidyasagar & Pammer, 2010). Consistent with this possibility, a growing body of work has reported atypical visual functioning in at least a subset of individuals with dyslexia, including differences in visual attention (Facoetti, Corradi, Ruffino, Gori, & Zorzi, 2010), visual crowding (Martelli, Di Filippo, Spinelli, & Zoccolotti, 2009; Zorzi et al., 2012), and broader high-level visual processing (Pitchford, Devillez, & Sigurdardottir, 2025; Sigurdardottir, Ívarsson, Kristinsdóttir, & Kristjánsson, 2015). Recent reviews have argued that such findings warrant serious consideration of visual contributions within broader multifactorial accounts of dyslexia, rather than being treated as alternatives to phonological explanations (Kristjánsson & Sigurdardottir, 2023). The present study contributes to this ongoing discussion by focusing on one specific visual constraint, visual span, and examining its relationship to reading performance in individuals with dyslexia.

Visual span is an important contributor to reading performance. It is defined as the number of letters that can be recognized accurately during a single eye fixation (Legge, Mansfield, & Chung, 2001). Because visual span constrains how much orthographic information can be extracted at a glance, it has been proposed as an important sensory bottleneck on reading speed (Legge et al., 2007). Consistent with this view, larger visual spans are associated with faster reading, particularly in individuals with visual impairments (Cheong, Legge, Lawrence, Cheung, & Ruff, 2008; Kwon, Liu, Patel, & Girkin, 2017). At the same time, visual span may not be entirely fixed, as developmental and training studies suggest that it can change with experience (e.g., Chung, Legge, & Cheung, 2004; Waugh, Formankiewicz, Álvaro, & Pelli, 2026). For example, visual span increases across school grade levels and is significantly correlated with reading speed in children (Kwon, Legge, & Dubbels, 2007). Accordingly, factors associated with reading experience, such as educational background, could plausibly contribute to individual differences in visual span. However, to our knowledge, there is currently no evidence that reading exposure during adulthood influences visual span size. At the same time, converging evidence indicates that sensory constraints, particularly visual crowding, play a major role in determining visual span size (Legge et al., 2007; Pelli et al., 2007). This raises the question of whether a reduced visual span may contribute to the slower reading commonly observed in people with dyslexia, even in the absence of clinically defined visual impairment.

Despite the close relationship between visual span and reading, visual span has not previously been investigated in the dyslexia population. On the other hand, visual attention (VA) span, which is the number of distinct visual elements that can be processed in parallel in a multielement array (Bosse, Tainturier, & Valdois, 2007), has been widely studied in dyslexia. A substantial body of research has reported reduced VA span in individuals with dyslexia (Bosse & Valdois, 2009; Lobier, Zoubrinetzky, & Valdois, 2012; Chen, Schneps, Masyn, & Thomson, 2016). Although visual span and VA span share some similarities, they are two different concepts addressing different mechanisms. Both measurements require individuals to fixate on a dot on the screen and verbally report the letters presented on the screen without moving their eyes. However, in visual span measurement, the letters are placed side-by-side as in typical text, allowing for crowding interference. In contrast, VA span tasks typically present letters with substantial spacing, minimizing crowding effects. Furthermore, visual span measurement requires accurate report of both letter identity and letter order, whereas VA span tasks impose no order constraint. As a result, visual span measurement captures crowding and letter order errors that are directly relevant to reading performance, whereas VA span tasks aim to isolate spatial attention while minimizing reading-specific constraints (for a detailed review, see Frey & Bosse, 2018).

An additional distinction lies in how performance is characterized. Visual span is typically quantified as a profile, reflecting letter recognition accuracy as a function of distance from the point of fixation. This emphasizes the idea of the size of the visual span. As shown in Figure 1, systematically measuring visual recognition at different eccentricities reveals a curve for a particular individual, which corresponds to the span size. In contrast, VA span measures emphasize overall reporting accuracy without capturing how performance varies across visual space. This profile-based approach provides a more direct estimate of how much textual information can be processed per fixation during reading. In the present study, we therefore measured visual span profiles in individuals with and without dyslexia, hypothesizing that visual span would be reduced in dyslexia and that smaller visual spans would be associated with slower reading speeds.

Figure 1.

(a) The demonstration of a trial in the visual span experiment when the stimuli are presented at letter position 4. The gray numbers indicating the letter positions are shown here for demonstration purposes; the participants in the experiment were not presented with these numbers. (b) Diagram of the visual span. A visual span profile is plotted as letter recognition accuracy versus letter position and then converted to information transmitted (bits). The area under the curve is the size of the visual span. (Adapted from He, Legge, & Yu, 2013.)

Our first study examined visual span in typical readers and individuals with dyslexia under controlled laboratory conditions. This experiment was part of a larger investigation that also included measures of visual crowding and standardized assessments of reading-related skills. Due to recruitment and scheduling constraints during the COVID-19 pandemic, the sample size was necessarily limited. Nevertheless, the results provided initial evidence that individuals with dyslexia, on average, exhibited narrower visual spans and slower reading speeds than typical readers, alongside substantial interindividual variability within the dyslexia group.

To address the limitations of small sample sizes and to examine visual span in a more diverse population, we next adapted the visual span paradigm for online testing. Although online data collection is often associated with increased noise, a growing number of vision science studies have demonstrated that carefully designed online paradigms can yield reliable and valid results while substantially expanding sample size and diversity (Yung, Cardoso-Leite, Dale, Bavelier, & Green, 2015). This approach is particularly important in dyslexia research, where high heterogeneity has contributed to mixed findings and competing theoretical accounts (Goswami, 2015). While some variability is expected given the heterogeneous nature of dyslexia, inconsistent results have also been attributed to small samples and selective reporting. As emphasized by Ramus and Ahissar (2012), isolated impairments observed in limited samples are often generalized to the entire dyslexia population, whereas other studies report intact performance on similar tasks. These concerns highlight the need for large-scale, well-controlled studies that treat reading ability as a continuum rather than a categorical distinction.

In line with this motivation, our second experiment served as a validation study for implementing visual span measurement in an online environment. We first tested a typical reading population using the same visual span paradigm and an online version of the MNREAD reading test (Mansfield, Ahn, Legge, & Luebker, 1993). The online results closely mirrored established laboratory findings, with the expected relationship between visual span size and reading speed, while showing slightly slower reading rates overall. Having validated the online approach, we then applied the same paradigm to a larger dyslexia sample in our third experiment. Consistent with both our laboratory findings and prior work linking visual span to reading speed, individuals with dyslexia demonstrated slower reading performance, smaller visual spans, and a strong correlation between the two measures. Together, these findings highlight the association between visual span and reading performance in dyslexia and demonstrate the feasibility of administering psychophysically grounded visual span measurements online to support large-scale, inclusive research.

Experiment 1: Lab study

Methods

Participants

Twenty-one participants were recruited for this study. The participant pool consisted of two groups: dyslexia and control. The control group included 10 participants (age range: 19–51 years, M = 33.2), recruited from staff and students at the University of Minnesota. The dyslexia group included 11 participants (age range: 18–60 years, M = 28.6), all of whom self-reported having dyslexia. Participants in the dyslexia group were recruited through the University of Minnesota, social media announcements, and the Learning Disabilities Association (LDA) Minnesota.

All participants completed assessments of reading and phonological abilities, including subtests from the Woodcock–Johnson IV (Letter–Word Identification, Word Attack, Sentence Reading Fluency, Word Reading Fluency, Spelling) and the Test of Auditory Processing Skills–3 (Phonological Segmentation and Phonological Blending). Group averages for these assessments (reported as percentile ranks) are shown in Table 1, and individual scores are provided in Appendix Table A1.

Table 1.

Characteristics of participants. Notes: Means and standard deviations of the two groups’ scores in phonological and reading measures. *p < 0.05, **p < 0.01, ***p < 0.001.

Characteristic	Control group (n = 10)	Dyslexia group (n = 11)	t value
Age (years)	28.6 (3.59)	33.2 (3.94)	−0.86
Basic Reading Skills (percentile)
Letter–Word Recognition	64.9 (6.8)	36.7 (7.2)	2.89**
Word Attack	59.9 (7)	23.1 (4.5)	4.45***
Reading Rate (percentile)
Sentence Reading Fluency	75.2 (4.5)	33.3 (8.3)	4.42***
Word Reading Fluency	72.7 (4)	27.8 (7.2)	5.43***
Spelling (percentile)	59.1 (7.7)	34.8 (7.8)	2.21*
Phonological skills (percentile)
Phoneme segmentation	48.9 (7.4)	19.8 (3.7)	3.51**
Phonological blending	54.5 (9.9)	27.3 (8.6)	2.07

All participants were native English speakers with self-reported normal or corrected-to-normal vision. Written informed consent was obtained in accordance with protocols approved by the human subjects’ review committee of the University of Minnesota.

Stimuli, apparatus, and procedure

The stimuli were generated using MATLAB with the Psychtoolbox extensions (Brainard, 1997; Pelli, 1997) and displayed on a 21.5-inch iMac (refresh rate: 75 Hz, resolution: 4,096 × 2,304). Participants completed two experimental sessions: (1) reading and phonological assessments, and (2) behavioral visual measurements, which included a visual span experiment and a flashcard-based reading speed test. All visual stimuli were viewed binocularly from a distance of 40 cm in a dimly lit room, with viewing distance stabilized using a chin rest. Letter stimuli were presented with an x-height of 1°.

In the visual span experiment, the stimuli consisted of black lowercase letters on a white background. The letters were rendered in Courier font. Letter spacing was 1.16 * x-width (standard spacing for Courier).

Participants were asked to fixate at the middle of the screen, on a gap between the two dots. Letter stimuli were presented in one of 11 possible slots, which were horizontally aligned with the fixation gap, as shown in Figure 1a. The slot on the fixation midline is labeled 0, and left and right slots to the fixation are labeled with negative and positive numbers, respectively. Letters were presented either in isolation (single-letter recognition condition) or in triplets (trigrams). Letters were randomly chosen from 26 possible English lowercase letters and were presented for 100 ms.

Participants clicked the mouse to initiate the trial. After the beep sound, the letter stimuli were presented. After the stimulus presentation, participants verbally reported the letters to the experimenter. No feedback was given. For the next trial, participants clicked the mouse again.

This experiment measured the visual span size of participants in three different conditions: single-letter recognition, trigrams with full report, and trigrams with partial report. In the single-letter condition, participants reported the single letter presented on the screen. In the trigram full-report condition, participants reported all three letters. Participants were scored as correct as long as both the letter identity and the location in the trigram were correct. In the trigram partial-report condition, participants were asked only to report the middle letter of the trigram. The difference between the two trigram conditions was included to precisely compare single-letter and trigram conditions with and without the difference in cognitive load in reporting. Each condition consisted of 77 trials, with seven repetitions on each slot from –5 to 5, including 0 (the midline).

Letter recognition accuracy was calculated in each condition. Percent correct for the visual span in each letter position was obtained by summing across all trials as the letters at a given position are sometimes inner, sometimes middle, and sometimes outer letters of the trigram. Proportion correct was plotted as a function of letter position to obtain the visual span profiles for each participant (for an example, see Figure 1b).

A flashcard reading paradigm was used to measure participants’ reading speed. This paradigm is designed to estimate reading speed by measuring the number of words read correctly as a function of exposure time.

In this test, sentences were presented on the screen for fixed exposure times, and participants were asked to read the sentence aloud. After the sentence disappeared from the screen, participants were allowed to complete their verbalization. When participants completed their verbalization, the experimenter noted the number of correct words read for that sentence. Participants clicked the mouse to initiate the next trial.

The sentences were selected from a large pool of MNREAD sentences (Mansfield, Atilgan, Lewis, & Legge, 2019) and were presented in Times New Roman font. These sentences all had 60 characters and an average of 10 words. Each sentence was rendered on the screen in three lines and justified configuration (see Figure 2). None of the sentences was presented twice for a given participant.

Figure 2.

Flashcard reading test—sample stimuli.

Each participant completed a practice run, which included 12 practice sentences at six exposure times. In the main experiment, participants read a total of 36 sentences. Each sentence was presented on the screen in one of the six predefined duration times. Each presentation duration was repeated six times.

The order in which participants were presented with different stimulus durations was randomized across participants. The sentence and the duration pairs were not fixed, meaning that while Sentence 1 was presented for 0.8 seconds to one participant, the same sentence could be presented for 3 seconds to another participant.

The presentation durations were usually 0.4, 0.8, 1.2, 1.6, 2.4, and 3 seconds. Each participant was expected to be able to read no more than 30% correct at the shortest duration and at least 80% correct at the longest duration. For a small number of participants, the range of the duration times was adjusted to make sure they would be able to read the full sentence in the longest duration.

The data were fit with a psychometric function to obtain the maximum reading speed for each participant. More details on this calculation are given in the next section.

Data analysis

The reading and phonology assessments were scored based on the specific instructions of the tests. The scores for the subsets of the Woodcock–Johnson IV test and the Test of Auditory Processing Skills 3 (TAPS-3) were converted to standardized scores.

In the visual span experiment, letter recognition accuracy was calculated for all three conditions, both as proportion correct and as information transmitted in bits. Visual span size was also determined by summing across the information transmitted by the nine slots of the profile. The visual span size analysis was conducted on the 9 slots, instead of the 11 slots that were tested, because the last two slots were never middle or inner members of the trigrams, so they were excluded from the analysis. An independent t-test was conducted to examine the difference between the two groups in their visual span sizes.

Letter recognition accuracy was also converted to information transmitted (in bits), following the standard visual span method used in prior work (e.g., Legge et al., 2001; Legge et al., 2007). This conversion is based on letter confusion matrices reported by Beckmann (1998), which quantify the amount of information transmitted at different levels of letter recognition accuracy using mutual information calculations. Because there are 26 possible letter responses, chance performance (1/26 = 3.8% correct) corresponds to 0 bits of transmitted information, whereas perfect letter recognition (100% correct) corresponds to approximately 4.7 bits. The relationship between proportion-correct letter recognition and information transmitted was well approximated by a linear function, yielding the following conversion formula:

$$\begin{array}{ccc}& & \mathrm{Information}\mathrm{transmitted}\mathrm{in}\mathrm{bits}\hfill \\ & & =-0.036996+4.6761\times \mathrm{letter}\mathrm{recognition}\mathrm{accuracy}\hfill \end{array}$$

The visual span size was quantified by summing across the information transmitted by the nine slots of the profile.

In addition to the standard visual span size metric, we conducted a supplementary analysis to characterize the spatial extent (width) of the span profile. For each participant, we identified letter positions with recognition accuracy exceeding 75%, a threshold commonly used in psychometric analyses to define reliable performance. Span width was then quantified as the number of positions meeting this criterion, as well as the sum of information transmitted (in bits) across those positions. This measure provides an estimate of the spatial range over which letter recognition remains reliable and helps distinguish the breadth of the span from its overall information-based metric.

In the Flashcard reading measurement, the number of correctly read words was plotted as a function of exposure time by fitting a psychometric function. Fitting was accomplished with cumulative Gaussian curves on linear-log axes, and the maximum reading speed in words per minute (wpm) was determined by the exposure time that yielded 80% of the words read correctly. Reading speed was computed according to the following formula:

$$\begin{array}{ccc}& & \mathrm{Reading}\mathrm{speed}\left(\mathrm{in}\mathrm{words}\mathrm{per}\mathrm{minute}\left[\mathrm{wpm}\right]\right):\hfill \\ & & 60/\mathrm{criterion}\mathrm{exposure}\mathrm{time}\mathrm{for}\mathrm{word}\left(\mathrm{in}\mathrm{seconds}\right)\hfill \end{array}$$

The average maximum speeds for both dyslexia and control groups were calculated.

Independent-samples t-tests were used to compare visual span sizes and reading speeds between groups. Pearson correlations were computed to examine relationships between visual span size, reading speed, and standardized assessment scores.

Results and Discussion

Scores on the standardized reading and phonological assessments confirmed the characterization of the dyslexia and control groups. These measures are not discussed further here; a summary of group averages is provided in Table 1, and individual scores are reported in Appendix 1.

The average reading speed measured by the Flashcard reading speed test was significantly slower for the dyslexia group (M = 325.7 wpm, SD = 51.6) than for the control group (M = 538.6 wpm, SD = 103), F(1, 19) = 36.92, p < 0.001. Individual reading speeds and group averages are shown in Figure 3. Substantial variability was observed within both groups, consistent with known individual differences in reading performance.

Figure 3.

Reading speeds measured with the Flashcard reading test in Experiment 1, including individual data points and group averages.

Similarly, individual visual span sizes (in bits), together with group averages, are shown in Figure 4. As with the reading speed data, substantial within-group variability was observed, particularly among participants with dyslexia, as expected.

Figure 4.

Visual span sizes (in bits) in Experiment 1, including individual data points and group averages.

To provide a more detailed characterization of group differences, Figure 5 shows the average visual span profiles for the dyslexia and control groups across the three conditions: single-letter recognition, trigram partial report, and trigram full report. The two groups did not differ in single-letter recognition performance. In contrast, visual span profiles measured with trigrams were narrower in the dyslexia group than in the control group in both partial- and full-report conditions. Independent-samples t-tests comparing visual span size (quantified as the sum of information transmitted in bits) confirmed these observations. Visual span size differed significantly between groups for the trigram full-report condition, t(19) = 2.53, p = 0.021, and for the trigram partial-report condition, t(19) = 2.51, p = 0.022. No significant group difference was found for the single-letter condition, t(19) = 1.45, p = 0.166.

Figure 5.

Visual span profiles (letter recognition accuracy as a function of letter positions) for single letters and trigrams, reported in both full and partial methods, for both dyslexia (blue lines) and control (red lines) groups.

In addition, visual span profiles revealed an asymmetry between positions to the left and right of fixation, with a smaller span on the left and a larger span on the right in both the control and dyslexia groups. The average left visual span sizes (in bits) were 16.1 and 14.4, while the corresponding right visual span sizes were 16.9 and 16.3 for the control and dyslexia groups, respectively. Although the effect size appears relatively large, given the small number of participants in each group, this finding should be interpreted cautiously and considered an observational pattern in the present experiment.

Consistent with the appearance of the span profiles in Figure 5, the width of the visual span analysis revealed that the dyslexia group had a narrower span than the control group, with fewer positions exceeding the 75% threshold (dyslexia: M = 6.7, SD = 1.7; control: M = 7.9, SD = 0.7, t(19) = –1.62, p = 0.06. A similar pattern was observed when span width was quantified as the sum of information transmitted (in bits) across positions exceeding the threshold (dyslexia: M = 28.13, SD = 7.32; control: M = 31.47, SD = 4.13), t(19) = –1.86, p = 0.04. To further examine the spatial characteristics of this effect, and in line with previously reported left–right asymmetry, span width was analyzed separately for letter positions to the left and right of fixation for the positions exceeding the 75% threshold. Group differences were observed on both sides but were more pronounced on the left of fixation (dyslexia: M = 2.3 vs. control: M = 3.2), t(19) = –1.75, p = 0.05, than on the right (dyslexia: M = 3.5 vs. control: M = 3.7), t(19) = –0.38, p = 0.35, in the number of letter positions, as well as the when span width was quantified as the sum of information transmitted in bits (left of fixation dyslexia: M = 9.16, SD = 4.68; vs. control: M = 12.71, SD = 3.58, t(19) = –2.05, p = 0.03, than on the right dyslexia: M = 14.45, SD = 3.05; vs. control: M = 14.65, SD = 2.03, t(19) = –0.72, p = 0.24). This asymmetry indicates that the narrowing of the visual span in dyslexia reflects a steeper decline in recognition accuracy with eccentricity, particularly in the left visual field.

Across all participants, visual span size was moderately correlated with flashcard reading speed, r(19) = 0.45, p = 0.041 (Figure 6). This relationship was weaker when correlations were computed separately within each group, likely reflecting substantial within-group variability in combination with the relatively small sample sizes. Additional correlations among standardized assessments, visual span measures, and reading speed are shown in Appendix 2, Figure A2.

Figure 6.

Correlation between reading speed (words per minute), measured using the Flashcard reading test, and visual span size (bits) for all participants in Experiment 1. The correlation was positive, r = 0.45, p = 0.041.

Experiment 1 provides initial evidence that individuals with dyslexia exhibit reduced visual span sizes and slower reading speeds compared to typical readers. Importantly, the absence of group differences in single-letter recognition indicates that reduced visual span in dyslexia cannot be attributed to basic letter identification deficits or peripheral acuity limitations.

The dissociation between single-letter and trigram performance clarifies the nature of the observed impairment. Participants with dyslexia showed poorer performance when letters were presented in triplets, even in the partial-report condition, where only one letter was required to be reported. This pattern suggests that reduced visual span in dyslexia is not driven by increased response demands but rather reflects difficulties related to multiletter processing, such as increased crowding or imprecise spatial integration.

Consistent with prior findings in low-vision research, visual span size was positively associated with reading speed. Although the sample size was modest, the observed correlation supports the hypothesis that visual span constitutes a visual constraint relevant to reading performance. These findings motivated the extension of the visual span paradigm to an online environment in order to test its reliability and scalability in larger samples and to further examine its relationship with reading behavior.

Experiment 2: Validation of online visual span measurement

Methods

Participants

Twenty-five participants recruited through online platform Prolific (Palan & Schitter, 2018). To make sure the study covered all ages, the age groups were controlled, ranging from 18 to 75 years (M = 44.16). All participants were native English speakers with self-reported normal or corrected-to-normal vision and reported no history of learning disabilities. Written informed consent was obtained electronically in accordance with protocols approved by the University of Central Florida Human Subjects Review Committee.

Stimuli, apparatus, and procedure

The stimuli were generated using Python with the PsychoPy package (Peirce et al., 2019), and the experiment was hosted online via Pavlovia (pavlovia.org). The experiment consisted of two sequential components: an online visual span measurement, followed by an online reading speed assessment. Given the uncontrolled nature of participants’ testing environments, particular care was taken to preserve stimulus scaling and task structure as closely as possible to the laboratory version used in Experiment 1.

Participants were instructed to maintain an approximate viewing distance of 50 cm, which was described as roughly equivalent to an arm's length from the screen. They were reminded throughout the experiment to keep this viewing distance as constant as possible. To standardize letter size across different screen resolutions and physical display sizes, a credit card calibration procedure was employed. Participants adjusted an on-screen rectangle to match the size of a physical credit card, allowing accurate conversion from pixels to visual angle. All letters were presented with an x-height of 1°, matching the laboratory study.

The visual span task closely followed the laboratory protocol, with the primary difference being the response modality. Instead of verbally reporting letters to an experimenter, participants typed their responses on a response screen following stimulus presentation. Trials were self-paced, with participants initiating each trial by pressing the space bar. To reduce participant fatigue and overall task duration in the online setting, visual span size was measured only using trigrams in the full-report condition. This condition was selected because it most directly captures crowding effects and reading-relevant constraints. The task consisted of 77 trials, with seven repetitions at each of the 11 letter positions ranging from –5 to +5 relative to fixation.

After completing the visual span task, participants completed an online adaptation of the MNREAD reading test. This test was modeled closely on the standard MNREAD procedure, which measures reading speed as a function of print size. As in the standard MNREAD test, sentences were presented one at a time in the Times Roman font, with print size decreasing in fixed logarithmic steps of 0.1 log units. A total of 12 sentences were presented, with the largest print size corresponding to 1.1 logMAR, equivalent to 1.045° of visual angle at a viewing distance of 50 cm, and an x-height of approximately 0.91 cm. Participants were instructed to read each sentence aloud. To assess reading accuracy and comprehension in the absence of an experimenter, participants completed a brief fill-in-the-blank response after each sentence. The blanked words were selected to be content-critical and varied across sentence positions. For very small print sizes, participants could select a “could not read” option, allowing the task to terminate appropriately. Practice trials were provided for both the visual span and MNREAD tasks. The total duration of the experiment was approximately 15 minutes.

Data analysis

Visual span size was computed using the same procedures as in Experiment 1. Letter recognition accuracy was converted to information transmitted in bits, and visual span size was quantified by summing across the nine central letter positions.

Reading speed was calculated following standard MNREAD scoring procedures. Because verbal responses could not be directly scored online, incorrect words were identified based on participants’ fill-in-the-blank responses. Maximum reading speed was computed in words per minute.

Results and discussion

The average reading speed measured by the online MNREAD test was 163.9 words per minute. This value was lower than reading speeds obtained using the flashcard paradigm in Experiment 1 and slightly lower than typical in-lab MNREAD values, which was expected given differences in task format, environmental control, and response requirements in the online setting. Importantly, reading speed showed a clear effect of age: Younger participants (≤45 years) read faster on average (M = 191.5 wpm) than older participants (M = 145.6 wpm). This pattern is consistent with well-established age-related declines in reading speed and provides an important validity check for the online implementation.

Similar to the literature in low vision and consistent with our lab experiment, visual span sizes in our study were correlated with reading speeds (r(23) = 0.49, p = 0.014; Figure 7). Participants with larger visual spans tended to read faster, even in the noisier online testing environment. Age was negatively correlated with reading speed (r(23) = –.41, p = 0.043).

Figure 7.

Correlation between reading speed (words per minute), measured using the online MNREAD test, and visual span size (bits) for all participants in Experiment 2. The correlation was positive, r = 0.49, p = 0.014.

In addition, as in Experiment 1, visual span profiles revealed a significant asymmetry in visual span size between positions to the left and right of fixation, with a smaller span on the left and a larger span on the right of fixation (t(23) = 5.54, p < 0.001); the average left visual span size in bits was 15.03, while the right visual span size was 16.8. The effect size was quite large, with a Cohen's d of 1.1. Such asymmetries have been linked to uneven crowding and may contribute to individual differences in reading efficiency. The observation of this effect in the online data further supports the sensitivity of the paradigm.

As in Experiment 1, to further characterize the spatial extent of the visual span, we conducted the width of the visual span analysis, which showed that, on average, participants exhibited a span width of M = 7.4 (SD = 1.4), corresponding to positions with reliably identifiable letters and bit-based span measure (M = 31.71, SD = 6.68). Consistent with the asymmetry observed in the bit-based span measure, span width was also greater on the right side of fixation (M = 3.7) than on the left (M = 2.7) and also in the bit-based span measure for letter positions exceeding 75% recognition accuracy (left of fixation dyslexia: M = 11.28, SD = 4.95; right of fixation dyslexia: M = 15.88, SD = 2.69). These results indicate that the spatial extent of reliable letter recognition is asymmetric in the online setting, mirroring patterns observed in the visual span profiles.

Taken together, these results demonstrate that the visual span measurement can be reliably implemented online. Despite increased variability relative to laboratory testing, the online paradigm preserved known relationships among visual span, reading speed, and age. Moreover, the combined visual span assessment required less than 5 minutes of active task time, suggesting its potential utility as a rapid, scalable tool for assessing visual constraints on reading. Having established the validity of the online approach in typical readers, Experiment 3 applied the same methodology to a large sample of individuals with dyslexia.

Experiment 3: Online visual span study in dyslexia

Methods

Participants

Seventy participants were recruited through the online platform Prolific (Palan & Schitter, 2018) with ages ranging from 19 to 59 years (M = 34.3). All participants were native English speakers with self-reported normal or corrected-to-normal vision. All participants reported having dyslexia, either through self-identification or a prior clinical diagnosis. Written informed consent was obtained electronically in accordance with protocols approved by the University of Central Florida Human Subjects Review Committee.

Stimuli, apparatus, and procedure

Stimuli, task design, and procedures were identical to those used in Experiment 2. Participants completed the online visual span task, followed by the online MNREAD reading assessment.

Data analysis

Data were analyzed using the same procedures as in Experiment 2. Visual span size was quantified based on trigram full-report performance, and reading speed was calculated according to standard MNREAD conventions. Four participants were excluded from the analyses because they did not complete the full experiment.

Results and discussion

The average reading speed measured by the online MNREAD test in the dyslexia group was 123.3 wpm (SD = 43.6), reflecting substantial interindividual variability. This variability is expected given the heterogeneous nature of dyslexia and the continuum of reading ability within the population. Reading speeds were markedly slower than those observed in typical readers in Experiment 2, confirming the sensitivity of the online MNREAD adaptation to known group-level differences. This pattern is also consistent with findings reported by O'Brien, Mansfield, and Legge (2005), who compared children with dyslexia and typically developing children matched for reading age, and similarly observed slower maximum reading speeds in the dyslexia group using printed MNREAD charts.

Visual span size was also smaller than that observed in the typical reader sample from Experiment 2, averaging 34.3 bits of information transmitted, compared with 36.4 bits in Experiment 2. Unlike in Experiment 2, correlation between age and reading speed (r(64) = 0.29, p < 0.022) was weak and not statistically reliable. This attenuation of age effects is consistent with the idea that dyslexia-related factors contribute more strongly to reading performance than age within this sample.

Critically, visual span size remained significantly correlated with reading speed (r(64) = 0.45, p < 0.001; Figure 8). The magnitude of this relationship closely matched that observed in the laboratory study (Experiment 1) and in the online validation study with typical readers (Experiment 2). Thus, despite substantial variability in both visual span and reading speed, the association between these measures was robust across testing environments and populations. To account for age differences within the dyslexia group, we conducted partial correlations controlling for age. The relationship between visual span size and reading speed remained significant (partial r = 0.41, p < 0.001), indicating that age did not account for the observed association.

Figure 8.

Correlation between reading speed (words per minute), measured using the online MNREAD test, and visual span size (bits) for all participants in Experiment 3. The correlation was positive, r = 0.45, p < 0.001.

As in Experiment 2, visual span profiles showed a significant asymmetry between left and right visual field positions, with a larger span on the right than on the left (t(64) = 8.12, p < 0.001); the average left visual span size in bits was 14.15, while the right visual span size was 15.8. The effect size was quite large, with a Cohen's d of 1.0. This finding suggests that asymmetric crowding or spatial processing may contribute to variability in reading performance in dyslexia, consistent with the notion that visual constraints play a role alongside phonological factors.

Similar to Experiments 1 and 2, looking at the analysis of span width, on average, participants exhibited a span width of M = 6.1 (SD = 2.0), corresponding to positions with reliably identifiable letters (75% correct criterion), and a bit-based span measure of M = 26.03 (SD = 8.85) across those positions. Consistent with the asymmetry observed in the overall span profiles, span width was greater on the right side of fixation (M = 3.0) than on the left (M = 2.2). A similar asymmetry was observed in the bit-based measure restricted to positions exceeding the 75% threshold (left: M = 9.02, SD = 5.03; right: M = 12.64, SD = 4.90). These results indicate that the spatial extent of reliable letter recognition is asymmetric in dyslexia, mirroring the pattern observed in the visual span profiles.

Taken together, these findings provide important insight into the factors underlying reading performance in dyslexia. The absence of a correlation between age and reading speed is expected, given that dyslexia is influenced by multiple interacting factors, with age not being a primary predictor of reading performance. What is particularly noteworthy, however, is that the relationship between visual span size and reading speed remains robust. Dyslexia is known to be highly heterogeneous, both in terms of severity and in the specific characteristics of the reading difficulties observed. In this context, it is an important finding that the visual span measure captures meaningful variation, even within such a diverse sample. The observed variability and wide range in both visual span size and reading speed are therefore expected and reflect the natural distribution of reading ability. Because reading skill lies on a continuum, this range is also expected to overlap with that of typical readers, which is indeed what we observe. A median split analysis, discussed further in the General Discussion, showed that the lower half of typical readers had an average visual span size of 35.3 bits, closely resembling the upper half of the dyslexia group (35.6 bits), while the corresponding reading speeds (106 and 154 wpm, respectively) were also substantially closer than the overall group averages.

Importantly, the fact that this brief, approximately 5-minute assessment reveals systematic patterns related to reading performance suggests that the visual span task may serve as a useful tool for the early identification of reading-related difficulties.

General discussion

The present work examined whether visual span, a well-established predictor of reading speed in low vision, is also related to reading performance in dyslexia. Across a controlled laboratory experiment and two complementary online studies, we investigated visual span characteristics and reading speed in both typical readers and individuals with dyslexia. In the laboratory study, individuals with dyslexia, on average, exhibited narrower visual span profiles and slower reading speeds than typical readers. A positive association between visual span size and reading speed was observed, most clearly when considering the full sample, with weaker and more variable patterns within individual groups. A similar overall pattern was observed in the online studies, where visual span and reading behavior were positively related despite differences in testing format and where larger samples allowed this association to be observed within groups as well. Together, these findings indicate that visual constraints at the level of multiletter processing are reliably linked to reading performance in dyslexia, even in the presence of substantial individual variability. To our knowledge, this is the first study to directly characterize visual span profiles in individuals with dyslexia and to demonstrate their systematic relationship with reading speed across both laboratory and online contexts.

To further clarify the nature of the observed reduction in visual span, we conducted an additional analysis to separately quantify the spatial extent (width) of the span profile. For each participant, we identified the letter positions at which recognition accuracy exceeded 75% and derived two complementary measures: the number of such positions and the sum of information transmitted (in bits) across those positions. Across all experiments, this analysis revealed a consistent pattern. The dyslexia group in Experiment 1 and the dyslexia sample in Experiment 3 exhibited reduced span width compared with the control groups in Experiment 1 and Experiment 2, reflected in both fewer positions exceeding the threshold and lower information transmitted within those positions. Importantly, this pattern indicates that the reduced visual span in dyslexia is driven primarily by a narrowing of the span profile (i.e., a steeper decline in recognition accuracy with eccentricity), rather than a general reduction in recognition accuracy, particularly near fixation. Although variability was greater in the online experiments, as expected given less controlled testing conditions, the same pattern was observed consistently across laboratory and online settings. These findings provide a more precise characterization of the visual span limitation in dyslexia, emphasizing a spatial constraint on multiletter processing rather than a uniform deficit in letter recognition.

Another key contribution of this work lies in extending the visual span framework to dyslexia. Visual span has traditionally been studied in the context of peripheral vision and low vision, where it is thought to reflect a sensory bottleneck that limits the number of letters that can be processed in parallel (Cheong et al., 2008; Legge et al., 2001). The present findings suggest that a similar constraint may be relevant for dyslexia. Importantly, reduced visual span was observed only when letters were presented in multiletter contexts and not in single-letter identification, ruling out explanations based on basic visual acuity or letter confusions. Performance was impaired both when participants reported all letters in a trigram and when they reported only the central letter, indicating that group differences cannot be attributed to response demands or memory load. This dissociation points to difficulties in integrating information across adjacent letters, consistent with accounts emphasizing crowding, spatial uncertainty, or imprecise letter position coding.

These findings also help clarify the relationship between visual span and the more extensively studied construct of VA span. Prior work on VA span has focused on the number of distinct visual elements that can be processed simultaneously under conditions designed to minimize crowding and letter position uncertainty, isolating attentional capacity. Deficits in VA span have been reported in a subset of individuals with dyslexia and have motivated influential subtype accounts based on attentional limitations (Bosse et al., 2007). While visual attention span and visual span are related, they are not equivalent. Visual span measurements explicitly incorporate crowding and letter order constraints, both of which are directly relevant to reading continuous text. This distinction has important implications for interpretation: While VA span impairments have been proposed to characterize only some individuals with dyslexia, reduced visual span was observed more consistently across individuals in the present study, despite substantial variability in magnitude. This pattern raises the possibility that visual span reflects a more general perceptual constraint on reading performance in dyslexia.

Although dyslexia is often characterized primarily in phonological terms, the present results underscore the importance of considering visual factors as complementary contributors to reading difficulty. Dyslexia is highly heterogeneous in both severity and underlying characteristics, and no single mechanism is expected to account for all cases. The persistence of a visual span–reading speed relationship across heterogeneous samples suggests that visual span captures a dimension of reading-related processing that cuts across individual differences. The observed overlap between dyslexic and typical readers further supports a dimensional, rather than categorical, view of visual constraints on reading.

To further examine group composition, we conducted a median split within each group based on reading speed. The lower half of the typical readers showed an average reading speed of 106 wpm and a visual span size of 35.3 bits, resembling the upper half of the dyslexic group (154 wpm; 35.6 bits). These results highlight substantial overlap between slower typical readers and higher-performing individuals with dyslexia in both reading speed and visual span size. Age may partly contribute to this overlap: the lower half of typical readers tended to be older (M = 51 years) than the upper half (M = 38 years), whereas no comparable age-related pattern was observed within the dyslexia group. Overall, these findings support the view that reading ability varies along a continuum, with group differences reflecting overlapping distributions rather than clearly discrete categories. The variability observed in both visual span size and reading speed is therefore not a limitation of the paradigm but a reflection of the continuous nature of reading ability, with overlap between dyslexic and typical readers.

The larger online dyslexia sample further revealed asymmetries in visual span profiles across the left and right visual fields, which were associated with additional variability in reading speed. Such asymmetries are consistent with prior findings linking perceptual and attentional biases to reading performance and may be especially relevant in dyslexia, where spatial and temporal processing differences have been reported. While the present data do not allow firm conclusions about the mechanisms underlying these asymmetries, they highlight the value of visual span profiles for capturing fine-grained characteristics of visual processing that may be obscured in global performance measures.

An additional strength of the visual span task is its efficiency and flexibility. The task can be administered in a relatively short time and does not require reading proficiency beyond basic letter knowledge. Moreover, prior work has shown that visual span can be measured using nonletter stimuli, such as faces, with similar profile shapes, suggesting that visual span may reflect a more general sensory bottleneck underlying pattern recognition (He et al., 2015). To our knowledge, this has not yet been tested in individuals with dyslexia. However, given the proposed underlying mechanisms, such as visual crowding and peripheral visual attention, we speculate that visual span measured with nonletter stimuli may also differ in this population. This possibility remains empirical and should be examined in future studies. Given this possibility, this property may make the visual span paradigm particularly promising for assessing children at risk for dyslexia, potentially before formal reading instruction and even before letter knowledge is fully established. In this sense, visual span may serve as a low-level visual indicator that complements phonological and attentional measures and is applicable across different stages of reading development.

Several limitations should be considered when interpreting the present findings. Dyslexia status in the online sample was determined based on self-report using Prolific's prescreening tool. This measure includes both individuals with a prior clinical diagnosis and those who self-identify as having dyslexia but does not distinguish between these groups. As a result, analyses separating formally diagnosed and self-identified individuals were not possible. Our operationalization of dyslexia in the online sample differs from approaches that rely on formal diagnostic assessment. Rather than requiring documented evaluation, self-report captures a broader group of individuals who experience persistent reading difficulties, including both those who have received a diagnosis and those who have not. This approach reflects both practical constraints of large-scale online data collection, where verification of diagnosis is not feasible, and a theoretical motivation to avoid restricting the sample to formally diagnosed individuals, which may bias it toward more severe cases. At the same time, this approach introduces additional sources of variability. Self-identification may be imperfect, and the inclusion of individuals without formal diagnosis may increase heterogeneity within the dyslexic sample. This limitation is particularly relevant when comparing across experiments. In Experiment 1, participants were recruited primarily through the LDA of Minnesota and thus likely represent individuals with prior formal evaluation. In addition, participants completed a battery of standardized Woodcock–Johnson assessments in the laboratory, providing further characterization of reading ability. In contrast, the online sample (Experiments 2, 3) relied on Prolific's prescreening tool, which may yield a more heterogeneous group. Furthermore, comparable standardized diagnostic measures could not be administered in this setting. Differences in recruitment context and reporting may therefore affect the composition of the dyslexic sample across experiments.

These considerations should be taken into account when interpreting the results, particularly with respect to the magnitude and consistency of group differences. Future work would benefit from combining scalable online methods with more detailed characterization of reading ability and diagnostic history, allowing for a more precise mapping between self-reported and clinically characterized dyslexia.

In addition, although the online visual span and reading tasks closely mirrored established laboratory paradigms, several factors likely limit direct comparability across experiments. First, the change in testing format (flashcard vs. MNREAD) introduces differences in task demands, stimulus presentation, and response dynamics, which may affect measured reading speed. Second, differences in sample composition, particularly the inclusion of a more heterogeneous, self-reported dyslexic group in the online experiments, may have attenuated group-level differences. Finally, the self-administered nature of the online procedure likely introduced additional variability due to less controlled viewing conditions and task engagement. Together, these factors likely increased measurement noise and reduced the apparent separation between groups across Experiments 1 and 3.

The present work also focused on adult readers and English orthography, leaving open questions about generalizability to children, preliterate populations, and readers of other writing systems. Finally, the correlational nature of the data does not permit causal conclusions regarding whether reduced visual span contributes to dyslexia or reflects downstream consequences of reduced reading experience. Future work combining longitudinal designs and targeted manipulations of visual parameters will be necessary to address these causal questions.

Beyond the specific findings, this work demonstrates the feasibility of conducting rigorous psychophysical assessments of visual processing in online environments. Dyslexia research spans multiple disciplines, yet the literature remains marked by mixed and sometimes contradictory results, due in part to small samples and reliance on single-task approaches. Online testing provides an opportunity to address these challenges by enabling access to larger and more diverse populations, supporting repeated testing within individuals, and allowing systematic manipulation of visual parameters such as font type, letter size, spacing, and contrast. By combining psychophysically grounded tasks with scalable methodologies, future work can move beyond isolated findings toward more integrated accounts of reading difficulty.

In summary, the present findings show that visual span is reliably related to reading speed in dyslexia across laboratory and online settings. This convergence supports the inclusion of visual constraints in models of dyslexia and highlights visual span as a promising tool for studying individual differences in reading. By bridging psychophysical methods and scalable online testing, this work contributes to a more comprehensive understanding of the visual factors that shape reading performance in dyslexia.

Appendix 1. Appendix 1: The individual scores in the assessments and reading speed test in experiment 1

Table A1.

Individual scores in all the standardized assessments and Flashcard reading speed test in Experiment 1. Note: N/A, not available.

		Assessments
			Test of Auditory Processing (in percentile rank)		Woodcock–Johnson IV Test of Achievement (in percentile rank)
Group	Age	Flashcard test Reading Speed (words per min)	Phoneme Segmentation	Phonological Blending	Letter–Word Recognition	Word Attack	Sentence Reading Fluency	Word Reading Fluency	Spelling
Dyslexia	60	370.6	16	16	79	35	38	45	66
Dyslexia	27	343.2	37	91	50	35	19	3	33
Dyslexia	54	264.0	1	2	8	3	6	5	1
Dyslexia	29	252.2	16	16	14	25	14	12	17
Dyslexia	34	374.6	2	25	N/A	N/A	N/A	N/A	N/A
Dyslexia	32	272.7	25	25	50	42	66	66	68
Dyslexia	36	365.4	37	9	30	4	9	9	N/A
Dyslexia	28	405.5	9	9	45	12	70	66	53
Dyslexia	18	295.7	25	75	18	18	19	45	18
Dyslexia	19	342.3	25	16	18	19	18	19	19
Dyslexia	28	297.0	25	16	55	38	19	63	38
Control	20	653.6	50	75	50	30	88	91	50
Control	32	428.4	50	75	63	60	50	53	33
Control	51	652.8	50	25	63	45	83	91	50
Control	29	452.8	75	91	72	76	68	70	72
Control	22	561.6	25	75	75	72	78	78	75
Control	29	624.0	75	95	94	76	78	75	91
Control	19	595.5	50	27	76	75	75	83	72
Control	20	495.2	37	16	55	63	53	50	24
Control	45	571.8	75	50	84	84	84	86	91

Appendix 2. Appendix 2: Correlation map including all assessments

Figure A2.

Correlation heatmap including all the standardized assessment scores and the values from visual span and Flashcard experiments.