Accessible charts are part of the equation of accessible papers: a heuristic evaluation of the highest impact LIS Journals

Purpose Statistical charts are an essential source of information in academic papers. Charts have an important role in conveying, clarifying and simplifying the research results provided by the authors, but they present some accessibility barriers for people with low vision. This article aims to evaluate the accessibility of the statistical charts published in the library and information science (LIS) journals with the greatest impact factor.

other hand, some incongruities between the technical suggestions of image submission and their application in analyzed charts also emerged. The main problems identified are: poor text alternatives, insufficient contrast ratio between adjacent colors, and the inexistence of customization options. Authoring tools do not help authors to fulfill these requirements.

Research limitations
The sample is not very extensive; nonetheless, it is representative of common practices and the most frequent accessibility problems in this context.

Social implications
The heuristics proposed are a good starting point to generate guidelines for authors when preparing their papers for publication and to guide journal publishers in creating accessible documents. Low vision users, a highly prevalent condition, will benefit from the improvements.

Why the statistical charts accessibility matters
The inclusion of statistical charts in academic research papers is a widespread practice. They have an important role in conveying, clarifying and simplifying the research results provided by the authors (McCathieNevile and Koivunen, 2012). Charts can also save readers time and energy and reduce the word count of the papers (Franzblau and Chung, 2012) On the other hand, other key sectors of society are also characterized by the extensive use of statistical charts as a tool to facilitate the understanding of information. This is the case, for example, of the news media. In this sense, the press has always used charts and infographics to represent data and statistics. The open data movement and the making available of large data sets in open access have only strengthened the socalled data journalism, multiplying this type of graphical representations in the media and increasing its interest among journalists, academics, computer scientists and designers (Meeks et al., 2019). Business intelligence is also another area in which statistical charts serve for exploration, analysis, and communication of data (Cairo, 2017). In the educational field, the knowledge about how to interpret and create statistical charts is present in different subjects and training levels, especially in the disciplines framed under the acronym STEM (Science, Technology, Engineering, and Mathematics), but also in other areas like social sciences or humanities. These are just some examples of key sectors of society that justify the need for accessible charts to guarantee access to information and knowledge for people with disabilities.
Visual representations enable communication of a wide variety of quantitative data, enabling readers to quickly and easily acquire and understand the nature of the underlying information . Although visual depictions are increasingly pervasive in science and social sciences, very little scientific literature is fully understandable because, as of now, critical graphical information is not directly accessible to visually impaired people (Gardner et al., 2009).

Why low vision people (Target group)
Low vision is the loss of sight that cannot be corrected in any form. It includes different degrees of sight loss, poor sensitivity to light or to contrast, color-blindness or color vision deficiency (CVD), night blindness, problems with glare, blurred vision, hazy vision, as well as almost complete loss of sight. There are multiple causes of low vision. Hereditary and congenital conditions are the most common causes of low vision and blindness among children worldwide, cataract among adults and elderly, and in countries in Africa, Asia and South America, infectious diseases such as trachoma and onchocerciasis are the main cause (Oduntan, 2005). Low vision is the visual impairment with the highest prevalence in the world, affecting approximately 217 million people (Bourne, et al. 2017), and this number will increase with the aging trend of the population. It must be emphasized that 86% of people with low vision and 61% of the population with presbyopia are 50 years or older (Bourne et al., 2017).
While the scientific literature published so far is mainly focused on the accessibility of statistical charts for blind people (Alcaraz et al., 2020a), only some of the aspects that improve the accessibility of statistical charts for this collective have benefits for people with low vision.
The solutions that focuses on alternatives other than graphical such as structured data tables, summaries or the use of sounds to communicate trends, do not have the same ability to efficiently show trends or comparisons between variables. They also require a greater use of short-term memory and a higher cognitive load when seeking to obtain answers or conclusions from tabulated data.
We must not forget that a significant percentage of users with low vision still have enough remaining vision to visualize the charts, either simply by resizing them, or by using the support of assistive technologies such as screen magnifiers, and that these people prefer to use their remaining visual capacity in their day to day (Szpiro et al., 2016), a condition that does not take into account the previous alternatives. According to their preferences, solutions such as the possibility to customize color or to increase the size of the chart or the text would better fit this user group and, regrettably, are not included in the current research literature.
In general, there is a significant lack of research focused on analyzing accessibility barriers and adequate technical solutions to guarantee accessibility for people with low vision (Moreno et al., 2020). The fact that many of these people can function independently despite certain limitations, without the help of white canes or guide dogs, makes them go unnoticed on a day-to-day basis. This has led to the description of low vision as an "invisible disability" (Shinohara and Wobbrock, 2011). An invisibility that has also been transferred to the scientific literature in a certain way.  show a lack of common and clearly defined guidelines addressing accessibility issues related to figures in computer science journals, and a high variability in the application of recommendations related to accessibility features, like textual alternatives, the use of safe color palettes and sufficient contrast or the image format, resolution and dimensions. Similar cases are found in mathematics journals; the journals use vector images in most of the cases and yet they do not benefit from the possibilities for accessibility of this format  compared to bitmap images.

Current situation
Among the publishers that have incorporated accessibility policies in recent years, Elsevier stands out. The publishing company has recently collaborated with the Highsoft Highcharts company in the creation of a JavaScript library with accessibility features to help improve the accessibility of its web chart library (Ted, 2018). The result is an accessibility chart JavaScript module with integrated screen reader and keyboard support. Moreover, Elsevier is undertaking some initiatives improving the accessibility of its collection, as for example in the journal Research in developmental disabilities (Nganji, 2015). The editorial is focusing the efforts on PDF files.

Related work
Several proposals exist for making statistical charts accessible to people with visual disabilities. However, most approaches focus on blind people or on people with severe low vision (Alcaraz et al., 2020a). Most of these proposals focus on one of the following four approaches: use of textual alternatives, sonification of data, generation of tactile alternatives and creation of multimodal alternatives. Regarding the use of textual alternatives, the Diagram Center (2015) has created guidelines on how to textually describe statistical charts and other types of complex images. Similarly, but oriented to a broader set of image types, the work of Splendiani (2015) focuses on how to textually describe non-text content for scientific articles. On the other hand, authors such as Corio and Lapalme (1999), Chester and Elzer (2005), Elzer et al. (2008), Ferres et al. (2010, Greenbacker (2011), , Nazemi and Murray (2013), Kallimani et al. (2013) or De (2018) propose different methods for the automated generation of textual alternatives from the information available in a chart. For their part, authors such as Elzer et al., (2007), Agarwal and Yu (2009)   have studied the importance of captions for the understanding of a chart as "it often concisely summarizes a paper's most important results" (Cohen et al., 2003).
Regarding the use of data sonification, the mapping of charts to musical tones (Cohen et al., 2005) and vibrations (Evreinova et al., 2008) has been explored, as also has the use of sounds to communicate trends (Alty and Rigas, 2005) (Walker and Nees, 2005) or the use of volume, timbre and position, to represent quantitative and qualitative data (Franklin and Roberts, 2003) (Treviranus et al., 2018). The precision of these techniques has also been analyzed using different combinations of instruments (Brown and Brewster, 2003). For its part, the creation of tactile versions of charts and maps has an important tradition, and there are even specific guidelines for its design (Braille Authority of North America, 2012). In literature we also find different approaches for its semi-automated generation. The works of (Ina, 1996), (Ladner et al., 2005), (Miele and Marston, 2005) (Watanabe et al., 2014) are some examples. Finally, other authors opt for multimodality, combining haptic solutions with data sonification and other stimuli (Kennel, 1996) (Fritz and Barner, 1999) (Yu et al., 2000) (Roth et al., 2002) (Yu and Brewster, 2003) (Iglesias et al., 2004) (McGookin and) (Wall and Brewster, 2006) (Doush et al., 2009) (Goncu et al., 2010).
Among these sources, especially the ones that focus on evaluation, the main references are the Web Content Accessibility Guidelines (WCAG). The WCAG have been adopted by many countries as the minimum legal requirement for public -and in some cases even private-websites to comply. In the case of European countries, the WCAG 2.1 has been integrated into the 301 549: Accessibility requirements suitable for public procurement of ICT products and services in Europe v2.1.2 (ETSI, 2018) a reference standard determining the accessibility of websites and mobile applications of public sector organizations.
The WCAG are organized under four theoretical principles covering every aspect of accessibility: perceivable, operable, understandable and robust. Every principle is detailed in several specific guidelines, which in turn are translated to directly assessable criteria divided in three levels of conformity. The From a business and marketing focused point of view Evergreen and Emery (2018) have created a data visualization checklist, relying on design principles collected by the same authors (2013), which covers many relevant aspects of its accessibility. The checklist has been rigorously tested by Sanjines (2018) and implemented as an online validator more recently (Evergreen, 2020).
On the other hand, in recent years other resources have also been published aimed at collecting accessibility requirements for people with low vision, including some relevant to statistical charts. This is the case of the accessibility requirements for people with low vision published by the Low Vision Task Force of the W3C (Allan et al., 2019), the compilation of adaptation techniques for this same user profile by Moreno et al. (2020) or van Achterberg (2019). In the same vein, but with a more practical orientation, Sorge (2020) has delved into the accessibility not only of statistical charts, but also of the remainder of STEM documents  due to its importance in guaranteeing students' access to these subjects under equal conditions. Finally, in the field of big data and data visualization techniques, Sathi and Sadhasivan (2020) have explored solutions to enable visually impaired users to access the Big data analysis results using Tableau Desktop software. For its part, Snaprud and Velazquez Regarding the field of scientific publication, Simon et al. (2019) results show that the most common accessibility problems with charts and figures in the proceedings published by the Innovation and Technology in Computer Science Education (ITiCSE) are captions that do not adequately describe the figure and the use of font sizes too small to be readable. Our hypothesis is that there are many other accessibility problems present in scientific journal papers. Furthermore, a wide range of barriers to access statistical charts are experienced by the different low vision profiles. These barriers can be overcome by including textual alternatives, high contrast images or with the use of patterns and textures as an alternative to the use of colors, among others, but they are not always required to the authors, or reviewed in sufficient detail by the publishers of these journals before publication.
To fill in the existing low-vision gap for this type of content, this paper aims to evaluate the accessibility for people with low vision of statistical charts in a sample of ten library and information science (LIS) representative science journals through a heuristic evaluation. Artwork submission policies are also reviewed. The results should confirm our hypothesis that there is a significant number of accessibility barriers for people with low vision in articles in scientific journals beyond those detected in other works published by other authors, making it difficult or impossible for this group to access research results presented as statistical charts.

Research method
The research is based on the heuristic evaluation method, one of the most efficient usability evaluation techniques without users. Streamlined, the heuristic evaluation is a usability engineering method to find the usability problems in a user interface design. It involves having a small set of evaluators examining the interface and judging its compliance with recognized usability principles (the "heuristics") (González et al., 2001). This technique has its origin in the work of Johnson et al. (1989) and was widely promoted in the seminal work of Nielsen and Molich (1990). Heuristic evaluations are very widespread in the field of usability and accessibility. On the basis of these works other authors have made methodological proposals for the preparation of new lists of heuristics for the evaluation of both general aspects related to usability, accessibility or user experience, known as "domain heuristics", leading to the emergence of specific and rigorous methodologies focused on how to create new domain heuristics (Rusu et al., 2011;Van Greunen et al., 2011;Hermawati and Lawson, 2015;Jiménez et al., 2017;Quiñones et al., 2018).
In our research, we follow the method by Quiñones et al. (2018), adapted for the creation of a list of heuristic indicators to evaluate the accessibility of statistical charts considering the needs of low vision and CVD users (Alcaraz et al., 2021). The heuristic indicators set proposed is made up of 18 indicators that cover aspects related to the information transmitted by the chart (title, axes, text alternatives, caption...), its visual display (typeface, colors, contrast...) and the behavior and functionalities they offer (personalization, visible focus indicator...).
In some cases, non-compliance to the heuristic affects the user experience of the chart, mildly compromising its accessibility. However, there are cases where the consequences of not complying with the heuristic results in one or more user profiles having serious difficulties to perceive the chart or being unable to do it, completely compromising its accessibility. For that reason, each heuristic has been weighed according to the criteria established in table 1.
The complete list of heuristic indicators is shown in table 2. However, some of the heuristic indicators in the initial list were implemented differently in scientific journal articles compared to news media (Alcaraz et al., 2020b), in particular, it is worth mentioning: -H1 'Title' versus H3 'Caption': most articles do not provide a title but instead they provide a caption. Following the initial evaluation criteria, H1 should be scored with 0 in almost all the charts. After a review, the evaluators decided not to take H1 into account, as not including the title responds to common practices of scientific articles.
-H6 'Data source': most articles presented charts with data from the article itself, thus not explicitly mentioning the source of the plotted numbers. In this case, again, the evaluators decided not to include this indicator on the final score.
-H15 'Without disturbing elements": after evaluating the charts in previous research, the evaluators discovered ads and watermarks for copyright purposes hindering important information from the charts and created an indicator to penalize it; but in the current research such practice is not common at all and the evaluators did not include this indicator in the final evaluation either.

Criteria Weight
If the chart fails the heuristic, one or more user profiles will not have a satisfactory user experience with the chart, mildly compromising its accessibility.
If the chart succeeds at the heuristic the chart's accessibility slightly improves.

x1
If the chart fails the heuristic, one or more user profiles will have serious difficulties to perceive the chart information, severely compromising its accessibility.
If the chart succeeds at the heuristic the chart's accessibility considerably improves. x2 If the chart fails the heuristic, one or more user profiles will not be able to perceive the chart information, totally compromising its accessibility.
This heuristic is key to provide access to the chart for one or more user profiles.

H1
Does the chart have a brief and descriptive title that helps users identify it among others appearing on the same page, as well as navigate between them? (not included in the final score) With the aim of achieving quantitative results that would later make it possible to compare the means or the level of accessibility with respect to a maximum score, a Likert scale of four points was used for the calculation of the level of compliance of each indicator. The range goes from 0 (worst possible score) to 4 (best possible score). Additionally, the options "Not applicable" and "It is not a problem" have been added, for those cases in which the question is not pertinent, or in which not complying with the indicator does not lead to an accessibility problem, respectively. The Likert scale is shown in table 3. The score in the Likert scale is multiplied by the weight resulting in a weighed value, for every indicator. The final value is multiplied by 10. In parallel, the maximum weighed value of the overall chart is calculated, considering that the maximum score for the "Not Applicable" and "Failure is not a problem" indicators is 0, and 4 for all the other indicators. Finally, the maximum weighed value is used to divide the obtained weighed value. The score formula is shown below: In If the ratios are bigger than 1 then it can be stated that the heuristic indicators set identifies more unique problems.
The results were as follows: ratio of unique problems: 2.54, ratio of problem dispersion: 1.52, ratio of severity: 1.07, and ratio of specificity: 1.27.
Demonstrating that the proposed heuristics find more unique problems, the problems are better distributed, more severe and specific than in the control set, and therefore the new set of heuristics is much more suitable for evaluating the accessibility of statistical charts than WCAG 2.1.

Analysis undertaken and sample of charts
The sample of charts to be evaluated (see Table 4) was taken from under these considerations: 1) Samples were taken from the ten library and information science journals with the highest impact factor according to the ranking of the Journal Citation Report (Science Edition 2018, 6 April 2020).
2) For each journal, 2019 issues were considered, 5 charts for each, among basic charts: bar charts, line charts, scatterplots and pie charts. Issues were reviewed from January to December, only one chart per issue was included (except in the Journal of computer-mediated communication, where 2 charts from the same issue had to be included because there were no more charts available). Preferably, charts that appear alone, not combined with other charts in the same figure, were selected in order to guarantee that the caption, alternative text… refer to the analyzed chart.
In  The complete list of the analyzed charts is included in appendix A.
Additionally, to complement the information gathered from actual charts in each journal, the researchers reviewed their submission policy (as of May 10   2020 He also recorded whether there was an option to change default settings in order to fulfill accessibility requirements. The logic behind this last step is that authoring tools play a key role in terms of the final accessibility of a chart, since it cannot be expected that authors know all the requirements of users and the tool should provide good defaults; in order to grant accessibility to a large scale, the charts created by a tool must meet the accessibility guidelines and the requirements of different users.

Evaluators team
Four experienced evaluators assessed the charts using the heuristic evaluation especially the discrepancies. Ideally there should not be any discrepancy between the evaluators, because they agree on the severity of the problems and they fully understand the heuristic principles. However, due to subjectivity affecting the scoring process, and to mitigate its effects, the standard deviation between the different evaluators' score is calculated and a threshold is set. If the standard deviation is higher than this threshold then, the scores are discussed jointly, to better understand the identified problem and the applied heuristic, until the different evaluators' scores are more coherent. After the scores are coherent, the final evaluation given is the average of the different evaluators' scores.
On this research, of the 705 indicators analyzed (15 heuristics for each of the 47 charts evaluated), only in 37 cases (5.248%) the scores differed with a standard deviation greater than 1, and only in 2 cases (0.283%) the different scores presented a standard deviation greater than 2. The threshold was set at 2. These results show a great coherence between the different evaluators' score and they can be perceived as a display of the quality of the heuristic indicators. When deviations higher to 2 were found, the evaluators discussed in depth the specific criteria used to score, and, in both cases, small corrections +-1 were applied after a better understanding of the logic.

Limitations
In this kind of research, sizing the sample is very complicated as bigger sizes imply a time cost difficult to assume. Moreover, the time cost is multiplied by the number of evaluators. On the other hand, information saturation is a good indicator of having covered the many different cases that could appear.
Information saturation signifies that new cases do not add new information to the research, as the results are homogeneous, and at some point, even repetitive. This is the case for this research. The sample of 10 journals, and a total of 47 charts, may be limited for generalization to the broad spectrum of LIS journals; nonetheless, it is representative enough to expose common practices and the most frequent accessibility problems in these contexts, and the results were coherent and repetitive among the sample.

Artwork submission policies
None of the journals analyzed had a specific accessibility policy statement on their websites, but Elsevier's journals link to the accessibility policy of the publisher's website.
Elsevier stands out among the other publishers by including in its Artwork and media instructions different recommendations that help ensure the accessibility of the statistical charts included in its publications. For example, the use of a color-blind safe colors' palette.
All Although optimizing the PDF version is the publisher's duty (H7), Elsevier is the only one that allows authors to decide if the figures in their papers should appear as a single, 1.5 or 2-column fitting image, thus allowing better use of the entire width of the page (see table 5, columns 2-7, row 11).
All journals except MIS quarterly ask authors to use safe colors for people with CVD (H10). In the case of the Journal of computer-mediated communication, they also mention the use of patterns in combination with color so that the differentiation of elements does not rely on color alone. It is precisely this journal the only one that underlines the importance of using images with adequate color contrast (H11) (see table 5, column 11, row 15).
Regarding the aspects related to legibility (H12), all the journals except the

Journal of knowledge management and the Journal of the American Medical
Informatics Association, present different recommendations related to the choice of the font family, its minimum size or line spacing. However, not all of these guidelines coincide with the recommendations of authors such as (Bernard et al., 2001), the recommended 12 pt. or 16 px equivalents for minimum font size (Nielsen, 2002;Kitchel, 2019), or the use of line spacing of at least 1.5 pt. (Rusell-Minda et al., 2007;Calabrese et al., 2010;Blackmore-Wright et al., 2013), or with the preferences of low vision users (WebAIM, 2018) regarding the use of sans-serif fonts. This requirement is included in authors' guidelines Ensure that each illustration has a caption. Supply captions separately, not attached to the figure. A caption should comprise a brief title (not on the figure itself) and a description of the illustration. Keep text in the illustrations themselves to a minimum but explain all symbols and abbreviations used.

All figures (include relevant captions)
The corresponding caption should be placed directly below the figure. As a general rule, the lettering on the artwork should have a finished, printed size of 7 pt for normal text and no smaller than 6 pt for subscript and superscript characters. Smaller lettering will yield text that is hardly legible. This is a rule-of-thumb rather than a strict rule. There are instances where other factors in the artwork (e.g., tints and shadings) dictate a finished size of perhaps 10 pt.
Line weights range from 0.10 pt to 1.5 pt Always include/embed fonts and use the recommended fonts where possible: Arial, Helvetica, Courier, Times, Times New Roman, Symbol

font size
As a general rule, the lettering on the artwork should have a finished, printed size of 7 pt for normal text and no smaller than 6 pt for subscript and superscript characters. Smaller lettering will yield text that is hardly legible. This is a rule-of-thumb rather than a strict rule. There are instances where other factors in the artwork (e.g., tints and shadings) dictate a finished size of perhaps 10 pt.
Line weights range from 0.10 pt to 1.5 pt Always include/embed fonts and use the recommended fonts where possible: Arial, Helvetica, Courier, Times, Times New Roman, Symbol Words within the image part of a figure may be singlespaced, oneand-a-halfspaced, or double-spaced, depending on which is the most effective layout for the information.
Align the text of an APA Style paper to the left margin Formats Regardless of the application used, when your electronic artwork is finalized, please 'save as' or convert the images to one of the following formats (note the resolution requirements for line drawings, halftones, and line/halftone combinations given below): EPS (or PDF): Vector drawings. Embed the font or save the text as 'graphics'. TIFF (or JPG): Color or grayscale photographs (halftones): always use a minimum of 300 dpi. TIFF (or JPG): Bitmapped line drawings: use a minimum of 1000 dpi. TIFF (or JPG): Combinations bitmapped line/half-tone (color or grayscale): a minimum of 500 dpi is required. Please do not: • Supply files that are optimized for screen use (e.g., GIF, BMP, PICT, WPG); the resolution is too low. • Supply files that are too low in resolution. • Submit graphics that are disproportionately large for the content.
Color artwork Please make sure that artwork files are in an acceptable format (TIFF (or JPEG), EPS (or PDF), or MS Office files) and with the correct resolution.

Evaluation results
All the charts have a legend (H2) if they require it and the majority of them (70.22%) have received scores between 2 and 4 (acceptable and excellent compliance). Only in 4.79% of the cases (9 out of 188 evaluations) the heuristic H2 has been scored with a 1 (low compliance), while the score of zero has only been given in 2.13% of the cases (4 out of 188 evaluations).
The H3 heuristic (axes) has also been evaluated positively in most cases, with scores of 2, 3 or 4, in 46.28% (87 out of 188 evaluations), 37.23% (70 out of 188 evaluations) and 4.26% (8 out of 188 evaluations) of cases, respectively. Even though, as mentioned above, most authors were not offered specific guidance in this aspect.
In those charts in which abbreviations (H5) were used, 17.02% of the cases (32 out of 188 evaluations) corresponded to standardized abbreviations, and the evaluators considered the lack of text expansion not a problem. In 35.11% of the cases (66 out of 188 evaluations), the charts showed abbreviations that were not expanded in the same chart, but instead they were expanded within the body of the article, and thus received a low score.
All publishers offered an optimized version for printing in PDF format. However, in many cases (34,57%) the two columns layout of the article make the charts too small to be readable.
Only 10  In 72,34% of the charts (136 out of 188 evaluations), color is used as a visual means of conveying information or distinguishing a visual element, and in 67,65%% of these cases a safe color palette or a pattern is used to facilitate differentiation. In all other cases, the colors used are not safe for one or more CVD profiles.
32.45% of the charts (61 out of 188 evaluations) has a text or non-text contrast ratio sufficient or higher than required. In the rest of the charts (65.55%, 127 charts) one or more color combinations do not reach the minimum required ratios. In general, the charts enjoy a good score for the legibility heuristic (H12), which was rated like "acceptable compliance" in 37.23% of cases (  compliance" or "low compliance" in heuristic H13 (image quality).
In 95.21% of cases (179 out of 188 evaluations), the resize heuristic (H14) was scored with a 4, corresponding to "excellent compliance". This is explained, in part, by the methodological choice of the best version available of the image (HTML, PDF, PPTX or the high-resolution JPEG or PNG version linked from the HTML version).
The heuristics related to the visibility of focus (H16) and device independent navigation (H17) have been rated as not applicable in all cases, because charts are raster images in which their marks (lines, points or bars) cannot be accessed. It is worth pointing out that we have not found any chart made with Highcharts (see introduction) in Elsevier's journals.
If the images used are raster images, they prevent or greatly hinder the personalization of charts through assistive technologies, which automatically scores 0 (no compliance) in heuristic H18 (customization). Figure 4 shows the average score by journal and the total average score and table 6 shows the average score of all the evaluators by chart.   All analyzed charts have a caption. However, in most cases, these are limited to function as replacements for the title. In most cases, the text alternative is limited to repeating the caption, therefore, far from being useful for users.
Regarding abbreviations, although Elsevier clearly indicates in its guidelines for authors that all the symbols and abbreviations used should be explained, the truth is that in the five journals from this publisher: International journal of information management (3 out of 3 cases), Government information quarterly (1 out of 2 cases), Information & management (4 out of 4 cases), Journal of strategic information systems (2 out of 2 cases) and Information processing & management (2 out of 2 cases), a common practice is that the abbreviations are explained in the main body of the article and not in the same chart. Thus, despite the publisher's requirement is met, the reader is forced to search for the meaning of the abbreviation in the text even if he or she only wants to consult the results of the research available in the charts.
In two out of the three journals that do not include technical requirements related to image quality (resolution, dimensions, etc.), the Journal of knowledge management and Journal of computer-mediated communication, this is not an obstacle to high quality images, which is a similar outcome to that of the publishers who include it in their guidelines. The exception is MIS quarterly journal, in which we find a chart that does not meet the indicator and two other charts in which it can be significantly improved.
Despite having evaluated statistical charts of the journals with the greatest impact in the area of library and information science, the results show a considerable number of accessibility problems and several inconsistencies with the editorial policies of the publishers. This observation showcases that even the largest publishers, which are motivated by increasing the quality of their publications and possess a larger budget and a larger editorial team, do not always guarantee quality aspects of their publications, such as accessibility.
The results of the evaluation confirm our initial hypothesis that there is a significant number of accessibility barriers for people with low vision in the charts included in papers of scientific journals beyond those detected in other works published by other authors, making it difficult or impossible for this group to access research results. In comparison with the results collected by Simon et al. (2019), the evaluation carried out has allowed finding a greater number of accessibility problems on the set of statistical charts evaluated. Unlike this other work, in our case, the captions in general have overcome the related heuristic. However, we have also encountered various legibility problems related not only to the font size but also to the font family used, the line height, or the contrast.
Unlike our previous work of evaluation of a set of statistical charts published in digital newspapers (Alcaraz et al., 2020b), the problem of the lack of text alternatives has not occurred in most of the charts analyzed. However, other problems coincide. In particular, the common problems in both types of publications are a poor non-text contrast ratio, a too small font size, the non-systematization of the use of color palettes appropriate for people with CVD, poor use of indicators to highlight the elements that receive focus -a functionality present only in certain vector charts-, or the inaccessibility through a keyboard interface.
It is difficult to compare these results with the related work, because there are no other similar evaluations apart from the one by Simon et al. (2019) and those made by our group.
Finally, Microsoft Excel, a very widespread tool in creating charts, offers default options that do not help authors in creating accessible charts. Significant changes need to be implemented to reach a high degree of accessibility, but simple improvements in color palettes and legibility would clearly improve the results.

Conclusions
From the point of view of publishers, accessibility is important for three reasons. First and foremost, to reach more readers making library and information science journals accessible to researchers with disabilities; second and equally important, to fulfill the accessibility regulations of many countries affecting public administration purchase policies (European Union, 2019). Finally, regarding brand image, accessibility helps comply with corporate social responsibility.
To help improve the accessibility of the statistical charts included in academic journals, publishers could do the following, amongst other actions: − Include a clear and complete policy on accessibility based on WCAG 2.1 for authors to adhere to when preparing their papers for publication and to guide their staff in producing accessible documents. This policy must include specific requirements so that the statistical charts included in the papers are accessible. The heuristics proposed in this research are a good starting point to generate these guidelines.
− Encourage authors to use authoring tools that conform to accessibility standards and help in producing accessible charts, fostering the use of vector charts.
− Comply with accessibility requirements for HTML version of the papers and adopt the accessible PDF/UA-compliant file format for the downloadable content.
This work showed the first stage of statistical charts accessibility evaluation, through a set of heuristic indicators, currently the researchers are working on a second stage, including users on the evaluation, as they are key to the final validation (Power et al. 2012, Lechner 2013.
The results of our research show that there is still a long way to go to achieve full accessibility of graphical content in academic journals, especially for people with low vision. This work contributes to solving this problem in two ways. First, our evaluation serves to get an idea of the current situation and show the main existing accessibility problems. Second, the proposed heuristics are also useful as a guide for creating accessible charts that could be easily incorporated into the style guides of any journal.