Aging and Pattern Complexity Effects on the Visual Span: Evidence from Chinese Character Recognition
, Lin Li 1, Sainan Zhao 1, Jingxin Wang 1,*, Kevin B. Paterson 2, Sarah J. White 2 and Kayleigh L. Warrington 2,*Round 1
Reviewer 1 Report
The authors studied the visual span of younger and older Chinese adults on a trigram reading task. A trigram was presented at different eccentricities in the left and right visual fields, and participants had to read the three trigrams while central fixation was enforced. Trigram complexity was manipulated within subject (low, medium, and high complexity). The results demonstrate an effect of complexity, where visual span was smaller for more complex trigrams. Critically, visual span for high complexity trigrams was even further reduced in older compared to younger adults. The authors also show a difference in reading rate, as measured by eye-tracking, between younger and older adults, where older adults had slower reading rates. There also was a significant correlation between visual span and reading rate, but only when the measures were collapsed across complexity levels.
The authors conduct a well-designed and thoughtful experiment, and do a nice job in presenting their findings in the broader literature. I enjoyed reading this paper and ultimately recommend it for publication, given the following issues are addressed. I have also indicated a couple analyses of interest that I do not think are necessary for the paper, but that they would be interesting to add if the authors also agree.
Issues:
Missing degrees of freedom for the F tests, which are important to verify that the ANOVA was coded properly
I mention this mostly because the authors refer to a 2 (between: age) x 3 (within: complexity) as a two-way repeated measures design, which I hope does not mean that they coded age as a within-subject factor (it should be between!)
I commend the authors for providing their data and processing script. However, I could not verify their analysis since they conducted their tests on model parameters which were based on either 1 or 2 Gaussian fits. They say on Page 6 that “best fit curve selected based on visual inspection and r2 values (for each complexity level, r2 > .96).” I just looked at gaussian fits (1 and 2 gaussians) for one group & condition (older, high complexity), and there were many cases that r2 was not higher than 0.96. More importantly, it looked like 1 gaussian was an adequate fit for most cases, with no clearly objective way described in the methods to decide whether 2 Gaussians provided a meaningfully better fit. I would suggest sticking with 1 gaussian for all subjects, or better describing for which cases 2 gaussians were decided to be a better fit.
Figure 4: add a legend, or describe in caption, what colour is what age group.
Not necessary, but out of curiosity for the authors: are the correlations drastically different if done on each age group separately? The variability in reading rate is much higher in one age group than the other.
Also not necessary, but would add interest to the paper: were there any meaningful asymmetries in span size, and did the asymmetry differ for younger and older adults.
Minor grammatical issues:
Introduction, page 3: “Importantly, however, as crowding effects appear TO be greater”
Same paragraph, “to shed light ON this issue”
Discussion, paragraph 2: “it was particular concern” should be… “it was particularly concerning/interesting”
Author Response
The authors conduct a well-designed and thoughtful experiment, and do a nice job in presenting their findings in the broader literature. I enjoyed reading this paper and ultimately recommend it for publication, given the following issues are addressed. I have also indicated a couple analyses of interest that I do not think are necessary for the paper, but that they would be interesting to add if the authors also agree.
Response: Thank you for your kind comments and thoughtful suggestions. We address each point below. We feel that these revisions have added clarity and interest to our manuscript.
Issues:
Missing degrees of freedom for the F tests, which are important to verify that the ANOVA was coded properly
Response: Thank you for pointing this out. This was an oversight and degrees of freedom have now been added in Table 2a.
I mention this mostly because the authors refer to a 2 (between: age) x 3 (within: complexity) as a two-way repeated measures design, which I hope does not mean that they coded age as a within-subject factor (it should be between!)
Response: Age was coded as a between-subjects variable and complexity as a within-subjects variable (this is detailed on page 4). We have adjusted the name of the test to a mixed design ANOVA (page 7).
I commend the authors for providing their data and processing script. However, I could not verify their analysis since they conducted their tests on model parameters which were based on either 1 or 2 Gaussian fits. They say on Page 6 that “best fit curve selected based on visual inspection and r2 values (for each complexity level, r2 > .96).” I just looked at gaussian fits (1 and 2 gaussians) for one group & condition (older, high complexity), and there were many cases that r2 was not higher than 0.96. More importantly, it looked like 1 gaussian was an adequate fit for most cases, with no clearly objective way described in the methods to decide whether 2 Gaussians provided a meaningfully better fit. I would suggest sticking with 1 gaussian for all subjects, or better describing for which cases 2 gaussians were decided to be a better fit.
Response: For transparency, we now include visual span estimates based on a single-Gaussian fit as a supplement to the analyses based on best-fitting curves (derived from either a single Gaussian or two Gaussians) as we remain concerned that these provide a poorer fit to the data and overestimate the visual span. The mean span sizes are now included in Table 2 and summary statistics in Table 2a and pairwise comparison reported in the following text on P8. The pattern of results for this analysis is the same as that from the analysis of the best-fitting curves.
Note that that r2 values previously referred to group levels averages. This has now been removed.
Figure 4: add a legend, or describe in caption, what colour is what age group.
Response: This information has now been added to the caption of Figure 4.
Not necessary, but out of curiosity for the authors: are the correlations drastically different if done on each age group separately? The variability in reading rate is much higher in one age group than the other.
Response: Across complexity groups, when analysed separately, the r value for the older adults is larger than for the young adults (older; r = .43, young; r = .23). Analyses conducted separately for each age group and complexity were noisy, most likely because of the smaller amount of data in these analyses. In general, while we can see this question is interesting, due to the lack of clarity we didn’t feel these analyses made a useful contribution to our manuscript.
Also not necessary, but would add interest to the paper: were there any meaningful asymmetries in span size, and did the asymmetry differ for younger and older adults.
Response: We have conducted these analyses by examining asymmetry when collapsed across complexity positions (although the pattern is similar within each complexity condition) and found that overall, the span size is asymmetric to the right, such that the span to the right was on average larger by .4 of a character (F(1,122) = 26.98, p < .001 ,ηp2 = .18 ). This asymmetry was numerically larger for the young adults, but this difference did not reach significance (young, M = .5 larger to the right, older, M = .25 larger to the right, F(1,122) = 3.104, p = .081, ηp2 = .03). Note that in the high complexity condition, 2 older adults were not included, as they did not achieve 80% at any position and so it was not possible to assess the asymmetry of the span for these participants. We have added a report of these comparisons (line 295-299).
Minor grammatical issues:
Introduction, page 3: “Importantly, however, as crowding effects appear TO be greater”
Response: Now corrected
Same paragraph, “to shed light ON this issue”
Response: Now corrected
Discussion, paragraph 2: “it was particular concern” should be… “it was particularly concerning/interesting”
Response: Now corrected
Reviewer 2 Report
In this article, the authors found a difference in the visual/attentional span size of trigrams with high complexity between young and older adults. The span size correlated with sentence reading speed. The results are interesting, but due to the nature of the experiment, no causal connection could be still established, and more research must be conducted.
some key points need to be clarified and explained in a better way:
1) The authors must better define how they fit the gaussian at the basis of visual span calculation (p.6, 237-241 “A single Gaussian…. An 80% criterion”. Moreover, I did not find defined how performance in each trial was measured: score of 1 (or 0) for each correct letter of the trigram (for each error), 1/0 for the entire trigram). to date, this method section is not completely clear.
2) Pag 8, 295 “crowding of middle…” the authors suggest that the lack of accuracy in position 0 could be an effect of crowding. Crowding at fixation position do not appear a plausible explanation for the observed low accuracy. Could the black dot act as a temporal mask (Black dot dimensions not described in the article)? This kind ok mask could produce positive effects on adjacent stimuli and negative effects on the central stimuli.
3) I’m not pretty sure that the homogeneity of variance is respected in the analysis in the evaluation of occurrence of the three groups of stimuli, occurrence of high complexity trigram characters appears quite different from the others two groups of stimuli.
4) Have the authors tried to check if the size of the crowding (the accuracy in the sole detection of the central stimulus of the trigram) is not enough to identify a difference between the groups?
Author Response
In this article, the authors found a difference in the visual/attentional span size of trigrams with high complexity between young and older adults. The span size correlated with sentence reading speed. The results are interesting, but due to the nature of the experiment, no causal connection could be still established, and more research must be conducted.
some key points need to be clarified and explained in a better way:
Response: Thank you for your thoughtful comments. We address each point below. We feel that these revisions have added interest and clarity to our manuscript.
1) The authors must better define how they fit the gaussian at the basis of visual span calculation (p.6, 237-241 “A single Gaussian…. An 80% criterion”. Moreover, I did not find defined how performance in each trial was measured: score of 1 (or 0) for each correct letter of the trigram (for each error), 1/0 for the entire trigram). to date, this method section is not completely clear.
Response: We have added the following information to page 6 to clarify how we defined our visual span estimates: “This measure provides an estimate of the number of character positions encompassed by the visual span by estimating the position at which performance reaches 80% accuracy (note that this threshold is commonly used in visual span studies). As performance for a given position was rarely exactly 80%, the Gaussian fitting estimates the intermediate point at which performance would reach 80% (e.g. if performance accuracy is 90% at one position and 70% at the next, the visual span estimate will fall between the two values). To reflect this, span size is reported to one decimal place.”
We now also clearly describe how each response was scored. On page 5 we have added the following text: “Recognition accuracy was scored separately for each character in a trigram (1 for a correct response and 0 for an incorrect response) and this score was assigned to that character position. For instance, for a trigram centered at position -3, the left character would appear at position -4 and the right character at position -2. If a participant incorrectly identified the left character but correctly identified the center and right character, performance for this trigram would be scored as 0 at position -4 and as 1 for positions -3 and -2.”
2) Pag 8, 295 “crowding of middle…” the authors suggest that the lack of accuracy in position 0 could be an effect of crowding. Crowding at fixation position do not appear a plausible explanation for the observed low accuracy. Could the black dot act as a temporal mask (Black dot dimensions not described in the article)? This kind ok mask could produce positive effects on adjacent stimuli and negative effects on the central stimuli.
Response: The reviewer raises an interesting consideration, and it does seem possible that the black dot may act as a kind of forward mask. However, it is not clear to us why such an effect would occur only when the middle character is in position 0 but not when the left or right character is in position 0 (i.e when the dot is in the same position as the left of right character), nor is it clear why the effect would occur only for older adults, as although there is evidence that masking effects are stronger for older adults (Atchley & Hoffman, 2004), this does not explain why they would be entirely absent for the young adults.
Moreover, recent studies suggest crowding can occur within foveal vision (Coates, Levi, Touch & Sabesan, 2018; Lev, Yehezkel & Polat, 2014) and that these effects may be larger for older adults (Yehezkel, Sterkin, Lev, Polat, 2015). Therefore, we consider foveal crowding to be a reasonable explanation for these effects, but we acknowledge that his needs further examination
3) I’m not pretty sure that the homogeneity of variance is respected in the analysis in the evaluation of occurrence of the three groups of stimuli, occurrence of high complexity trigram characters appears quite different from the others two groups of stimuli.
Response: To test for homogeneity of variance, we performed a Levene’s test for each complexity level, this analysis provided no significant results:
Low complexity: F = 3.191, p = .082;
Medium complexity: F = 2.317, p = .136;
High complexity: F = 1.507, p = .227;
4) Have the authors tried to check if the size of the crowding (the accuracy in the sole detection of the central stimulus of the trigram) is not enough to identify a difference between the groups?
Response: This is an interesting point that we looked at in response to the reviewer’s comment. However, this analysis would be based on a small dataset and we are concerned about the reliability of any estimates derived from this analysis. We comment on this on P9 (line 323-328).
