1 Introduction

The number of people with low vision worldwide is significant. Globally, in 2020, an estimated 43,3 million people were blind. On the other hand, it is estimated that 295 million people have moderate and severe vision impairment; 258 million have mild vision impairment; and 510 million have visual impairment from uncorrected presbyopia. Globally, between 1990 and 2020, the number of people who were blind increased by 50.6% and the number with moderate and severe vision impairment increased by 91.7% [1]. The same study predicts that by 2050, 61 million people will be blind, 474 million will have moderate and severe vision impairment, 360 million will have mild vision impairment, and 866 million will have uncorrected presbyopia.

Each disability affects visually impaired people differently, resulting in a significant variety of user profiles [2]. However, in all cases, low vision is a visual condition characterized by a substantial reduction in sight that cannot be corrected with lenses, medication, or surgery. Low vision profoundly impacts the daily lives of those who experience it [2] due to the prevalence of visual information in the acquisition of knowledge and daily activities. Visual information is often more detailed and richer compared to auditory or tactile information. As a result, much of our technology has been designed with a focus on sight.

In this context, the total or partial loss of visual perception can lead to a dramatic reduction in autonomy, given its crucial role in fundamental daily tasks such as learning, mobility, access to information, and social inclusion and participation.

Low vision is defined by assessing a person’s visual acuity and field of vision. Visual acuity measures the ability of the visual system to distinguish between two closely spaced points at a specific angle [3]. Field of vision, or peripheral vision, refers to the total visual area in degrees while the central point of focus remains fixed [4]. Generally, an individual is considered to have low vision if, even with the best optical correction, her visual acuity falls below 20/60 [4] or 20/70 [5], or their field of vision is less than 20°.

Low vision encompasses individuals with visual impairments other than blindness which cannot be fully corrected with lenses. This category includes various user profiles resulting from congenital origins or different eye conditions and diseases such as cataracts, glaucoma, macular degeneration, or diabetic retinopathy. Each person may exhibit varying degrees of visual acuity and field of vision, along with specific challenges related to contrast sensitivity, light or glare sensitivity, and color perception.

For example, macular degeneration is an eye disease that can result in blurred or absent vision in the central field, in areas known as scotomas. As a consequence, individuals with this condition often rely on their peripheral vision [2]. Glaucoma leads to the loss of peripheral vision, accompanied by a blurred central area, making tasks like reading, and seeing detail exceptionally challenging [6]. Cataracts can cause vision to become blurred or hazy, especially in bright light [2], and may also affect color perception [7]. Ocular albinism is characterized by reduced visual acuity and heightened sensitivity to brightness and light [8]. Nystagmus is associated with involuntary, uncontrollable eye movements linked to neurological issues [9]. Retinoschisis impacts both central vision acuity and peripheral vision, which can be lost if the inner layer of nerve cells detaches from the outer layer [10]. Multifocal chorioretinitis leads to vision loss, blurry vision and other symptoms [11]. Stargardt’s disease may manifest as gray, black, or hazy spots in the central vision, light sensitivity, slower adaptation to changes between light and dark environments and, in some cases, color blindness [12]. Finally, color vision deficiency (CVD), also known as color blindness, can be acquired but usually is a genetic condition [13] that affect the expression of the full complement of normal cone photoreceptors. CVD presents a wide range of severity: anomalous trichromacy, dichromacy, and monochromacy (complete inability to perceive any colors) [14].

Another challenge associated with low vision is that many of the vision problems mentioned tend to manifest as individuals age, either suddenly or gradually. These visual impairments often occur later in life, making it more challenging to acquire new skills and adapt to assistive technologies that may be unfamiliar.

Furthermore, people with low vision utilize a wide range of assistive technologies, with screen magnifiers being a prominent choice, followed by screen readers, zoom features integrated into web browsers, and high-contrast settings. This diversity in profiles, barriers, and assistive technologies presents a significant challenge in meeting the specific needs of each group through a single, universally effective design.

This research employs a user-centered approach to gain a deeper understanding of the obstacles faced by individuals with low vision when accessing data visualizations. In contemporary society, data permeates nearly every aspect of our lives [15], encompassing information dissemination, education, research, and leisure activities. As Kim et al. aptly stated, ‘our society is becoming data-driven’ [16]. Consequently, the ability to comprehend and manipulate data is crucial for individuals to grasp the world, make informed choices, access scientific findings, grasp abstract scientific concepts [17], and retrieve public health and social care information [18].

In our data-driven era, one of the primary challenges is to elevate data literacy. Data literacy encompasses the ability to proficiently comprehend, analyze, and communicate data [19]. In today’s society, data literacy is an indispensable skill [16, 20, 21]. It involves not only the capacity to read and interpret charts and data tables but also the ability to critically assess data quality, recognize biases, and grasp the implications of findings [22].

Data visualization, particularly statistical charts and various graphical representations, facilitates the efficient processing of information. Schepers [23] regards data visualization as an inherent assistive technology, a form of cognitive accessibility that leverages our visual system to ease the interpretation of tabular data [24]. Charts enable the visualization of abstract concepts and intricate relationships, which may be challenging to comprehend through alternative data formats [25], and facilitate the identification of patterns and trends in data [26].

In the realm of scientific communication, charts serve as concise and accessible means to convey the primary outcomes of extensive research endeavors [27,28,29]. Consequently, data visualization is recognized by several authors as an essential skill, not only for the general populace [16, 20] but also for future researchers [21].

In this context, obstacles to accessing data can exacerbate societal inequalities, particularly among individuals with disabilities, who already contend with various social and economic disparities. Recent examples have highlighted challenges in accessing public health information [30,31,32], political data [33], preserving professional autonomy [34], and securing quality education [35, 36].

Conversely, accessibility issues that have traditionally affected data visualizations, including statistical charts, also pose challenges for search engines. These engines struggle to crawl, index, utilize, and display these representations on their results pages, primarily due to them being static bitmap images or, at best, vector images with limited accessibility features despite the potential of this format [37].

In large part, this issue arises from content authors (such as designers, journalists, data scientists and researchers) lacking awareness of the accessibility barriers individuals encounter when trying to access their data visualizations. Furthermore, they may be unaware of the available techniques and solutions to address these barriers [38].

The primary objective of this article is to elucidate the needs and perspectives of individuals with low vision, with the aim of identifying key issues related to inaccessible charts. Ultimately, the authors seek to propose specific solutions that can enhance the accessibility of data visualizations.

2 Related research

While the field of data visualization has experienced exponential growth in recent years, research on the accessibility of visual artifacts within this discipline has not kept pace [39]. Presently, there is a rising interest in enhancing the accessibility of data visualizations for individuals with intellectual disabilities and the cognitive barriers caused by conventional design guidelines related to the chart type selection, chart embellishment conventions, and the representation of data through continuous marks versus discrete marks [40]. However, most of the related research on accessible data visualization and charts has predominantly focused on barriers to visual access [16, 39, 41], research methodologies applicable to accessibility [41, 42], practitioner-implemented solutions [43], and the analysis of the impact of elements such as image captions or alternative text [44,45,46,47].

Additionally, considerable attention has been given to the development of specific solutions and techniques aimed at facilitating data access. These include, for instance, the use of 3D printed maps and icons [48], tactile representations such as organic node-link diagrams, grid node-link diagrams, adjacency matrices or Braille lists [49], audio-tactile charts in SVG format optimized for embossers [50]; and the incorporation of natural language descriptions to provide context and insights [51]. Additionally, data sonification techniques [52] employ varying tones in terms of pitch and loudness to guide users through charts, often using a combination of MIDI sounds, synthesized speech, and pre-recorded audio files [53].

Other approaches involve chart image detection and classification to identify the chart-types, generate screen reader-compatible summary descriptions, export data to other formats like data tables, and create new, more accessible data visualizations based on existing ones [54]. Specific image processing algorithms are utilized to extract pertinent information from raster images and generate automated textual descriptions [55, 56], while deep neural network methods are employed to extract data from charts, including chart types, labels and relevant data, and convert them into vector charts format [38].

Some proposals enable users to navigate and interact with line charts using natural language commands and a Text-To-Speech (TTS) engine, facilitating specific queries about the chart’s content [57, 58]. Hybrid systems are designed to convey information through different senses, such as sight, touch, sound, or muscular resistance (haptic interfaces) [59]. There are even structured musical stimuli used to convey simple diagrams [60].

These techniques can be categorized into chart classification, text recognition, data extraction or data summarization [61], all of which aim to create alternative representations or provide users with access to chart information through their assistive technology.

Despite the higher prevalence of individuals with low vision, existing scientific literature has predominantly centered on blind individuals [62, 63], further marginalizing a group that remains relatively unknown to society [64]. Low vision users exhibit notable distinctions from blind individuals, and many within this group use their residual vision in their daily life as much as possible [65, 66], even if it implies continual adjustments to various interface aspects [65] or the adoption of uncomfortable or strained positions in front of screen.

User studies addressing the accessibility of data visualizations and statistical charts have primarily concentrated on blind individuals [38, 41, 49, 52, 57, 60, 67,68,69,70], with only a few including individuals with low vision [48, 71, 72] or color vision deficiency (CVD) [73, 74]. The dearth of studies aimed at identifying the needs and preferences of users with low vision underscores the need for further research in this domain. Notably, research specifically focusing on statistical charts -a content type integral to various key sectors such as education, research, communication, and business, among others- remains largely scarce. To address this gap, the authors have developed a set of heuristic principles for assessing the accessibility of such data visualizations [75] (see Table 1).

Table 1 Alcaraz et al.’s [75] list of heuristics with their description
Full size table

Heuristic evaluation (HE) stands as one of the most commonly used and effective usability assessment techniques that do not require direct user involvement. HE is a method within the field of usability engineering, employed to identify issues related to usability within a user interface design. It plays a crucial role in identifying and resolving these issues as part of an ongoing design enhancement process. In HE, a small group of evaluators inspects the interface and evaluates its adherence to established usability principles, often referred to as “heuristics”, “heuristic indicators” or “heuristic principles” [76].

Compared to an accessibility evaluation conducted using the WCAG as a reference, HE offers several advantages, including greater conciseness, memorability, meaningfulness, and comprehensibility of its principles [77].

On the other hand, domain-specific heuristics typically yield more effective and efficient results compared to general guidelines like WCAG, which primarily focus on website analysis [78].

A notable exception to the lack of research in this field is the set of heuristic indicators put forth by Elavsky and colleagues [79, 80], published subsequent to the development of the authors’ own set of heuristics. Elavsky and his co-authors’ recommendations are intended to assist visualization designers, journalists, and other practitioners in assessing the accessibility of data-driven visualizations. Their proposal encompasses a total of 50 heuristics, with 14 of them deemed critical, organized into 7 principles (perceivable, operable, understandable, robust, compromising, assistive, and flexible). These principles draw inspiration from the web content accessibility guidelines (WCAG) but have been extended and tailored to address the unique requirements of data visualizations. While aiming to meet the needs of a broad user spectrum, it is noteworthy that 31 of the 50 proposed principles specifically address barriers that may impact users with low vision. Table 2 provides a comparison between the heuristics proposed by Alcaraz et al. and those suggested by Elavsky et al. concerning individuals with low vision.

Table 2 Mapping between the heuristic principles proposed in previous works and those proposed by Elavsky
Full size table

3 Methodology

The development and validation of the heuristics by Alcaraz et al. followed formal and systematic methodology by Quiñones et al. [81] for creating usability heuristics, comprising eight key steps: (1) exploratory stage (literature review); (2) experimental stage (data analysis to retrieve additional information); (3) descriptive stage (select and prioritize the most important collected information during stage 1 and 2); (4) correlational stage (reconcile the domain features and functionalities with existing heuristic indicators); (5) selection stage (review the list of heuristic principles created); (6) specification stage (formal specification of each heuristic principle); (7) validation stage (experiments to determine the effectiveness and efficiency of the heuristic set); (8) refining stage (refine of the heuristic principles with the conclusions resulting from the previous stage.

Step seven involves the validation of the heuristics set through a series of experiments, assessing their effectiveness and efficiency. This validation process employs the following methods: (a) heuristic evaluation (mandatory): this method is a crucial part of the validation process; (b) expert judgment (optional): experts may be consulted to provide additional feedback, enhancing the validation process; (c) user testing (when necessary): user testing is employed to complement the validation process as needed.

The complete list of heuristics and their definitions is shown in Table 1. These heuristics were validated against WCAG 2.1 [82] in previous research through the analysis of published charts in several contexts: digital media [83], public health information, [31] and scholarly articles [84], with good results. In practical terms, domain-specific heuristics enable the detection of a higher proportion of unique problems, a more even distribution of problems across principles, the identification of a greater number of severe problems, and a more precise identification of specific issues.

3.1 Materials and methods

In this study, a second validation of the heuristic set with users is carried out, because users contribute with a new perspective and identify problems that experts cannot always detect [85, 86]. Special attention is paid to new possible barriers [87], and to the characteristics and needs of every specific profile [88]. According to Brajnik [87], “a barrier is any condition that hinders the user’s progress towards achievement of a goal, when the user is a disabled person”.

For the study, a series of synchronous, moderated, and remote user tests were carried out. The tests consisted of solving tasks for which users had to consult a set of web-based charts that had been created. In total, two different versions of three charts (horizontal bar chart, vertical stacked bar chart, and line chart) were generated: one accessible, created following the abovementioned heuristic guidelines [89], and another non-accessible version. The specific types of charts were chosen based on their popularity and adoption.

Despite the significant diversity among the profiles that participated in the user test, it was decided to utilize a single version for the non-accessible charts. This decision aimed to prevent a substantial increase in the time required to conduct the test and to minimize unnecessary and undesirable fatigue among the users. In the same sense, the color scheme selected for the accessible charts is intentionally kept as “neutral” as possible, utilizing black and white, which is safe for all colorblind profiles. In the case of non-accessible charts, the color scheme and layouts are based on the defaults offered by Microsoft Excel. This choice also enables us to evaluate the accessibility of charts generated automatically by this tool, particularly when they are not customized by their creator.

The non-accessible charts were generated by Microsoft Excel (2019 MSO 16.0.10356.20006 Windows) using the tool’s default options to create a chart of each of the selected types and generating an automatic export in HTML format (Fig. 1). Automated export of charts to HTML format using Excel involves converting the original vector image to a low-quality bitmap image. The export included the chart data table in text format.

Fig. 1
figure 1

Non-accessible bar chart created with Excel with an additional table in text format

Full size image

The accessible versions were created using the Highcharts JavaScript library (v. 8.0.0), including many of its accessibility options with the aid of its accessibility module: screen reader support, long description, keyboard navigation, the use of patterns as an alternative to color, a visual indicator when a mark of the chart receives the focus, and a table with the chart’s data, as well as a tooltip functionality that complements the legends, providing information on the value associated with each mark when the focus points to it (Fig. 2). All charts, questions and the results of the test are available online.Footnote 1

Fig. 2
figure 2

Accessible bar chart created with Highcharts

Full size image

3.2 Procedure

We designed five identical tasks for both versions, utilizing fictitious data and scenarios in a within-subject design (the same users participated in both conditions). To minimize learning effects, we modified the values represented in each version of the chart.

In defining these tasks, we drew from the categories proposed by Brehmer and Munzner [90], with a focus on tasks related to information consumption, particularly relevant in the context of public information. These tasks encompassed the following aspects: searching for unknown targets in known locations (browse), searching for unknown targets in unknown locations (explore), comparing multiple subsets of targets (compare), and summarizing targets, including the entire set of targets (summarize).

Tables 3, 4 and 5 show tasks, typology of task [90], objectives and related heuristics.

Table 3 Bar chart tasks, objectives and related heuristics
Full size table
Table 4 Stacked bar chart tasks, objectives and heuristics related
Full size table
Table 5 Line chart tasks, objectives and heuristics related
Full size table

For each task, the moderator read the explanation before starting, asked for questions from the participants, and explained subsequently. We employed the “think-aloud” method as our approach for this study. This method is commonly utilized in user studies to gain insights into the cognitive processes of users as they perform tasks [91]. It has proven to be a valuable and reliable technique due to its minimal disruption of participants’ thought processes [92]. In this method, participants are instructed to verbally articulate their thoughts while engaging with tasks, essentially vocalizing their inner dialogues. Moreover, participants are encouraged to explain what barriers or difficulties they encounter. Another advantage of this method is its avoidance of interpretation by the subjects and its simple verbalization process, making it an objective approach [92].

Metrics related to effectiveness (percentage of completion per task), efficiency (time per task), and satisfaction (measure of expectations, with a simplified 5-point Likert scale from 1, not at all complicated, to 5, very complicated) were collected during the test. Qualitative measures focused on detecting the barriers encountered by users and on analyzing the strategies and workarounds used by users to overcome the barriers they faced. After the test, users were asked for their favorite version of each chart, and informal comments were promoted. This approach integrated the “think-aloud” method with a follow-up interview, during which the moderator specifically inquired about any potential barriers or issues that users may not have verbalized during the tests.

3.3 Participants

To recruit participants, the authors distributed a questionnaire to individuals with low vision who are members of the Asociación Discapacidad Visual de Cataluña: B1 + B2 + B3 (Visual Disability Association of Catalonia, Spain). The questionnaire collected information on several factors, including: (a) age; (b) gender; (c) type and degree of visual impairment; (d) visual field affection and degree of affectation; (e) visual acuity; (f) color blindness; (g) light sensitivity and contrast sensitivity; (h) other disabilities that may impede computer use; and (i) level of education.

A total of 12 users were recruited, and with a snow-ball strategy from the early contacted users. Initially, tests were planned to be held in B1 + B2 + B3 offices, but due to access restrictions during the COVID pandemic, they were repurposed as remote tests with Zoom platform. Because of COVID and also due to the barriers expected to be encountered in the use of videoconferencing platforms, many of the previously contacted users (more than 20) refused to participate after having initially accepted.

On the other hand, remote tests allowed users to answer the tests from their own homes, with their personal computer equipment and assistive technology, with the ideal setup. Consent forms were sent to participants prior to the session so they could read, print, and sign them.

The sample was composed of 58.33% (7) men and 41.66% (5) women; 83.33% (10) of the users had higher studies and only two users (16.66%) had middle school and elementary school studies, respectively. The age of the participants was between 18 and 79 years, the average being 42,3 years. The sample included a variety of conditions associated with low vision: low visual acuity (6 users), reduced central vision (2 users), reduced peripheral vision (2 users), blurry vision (1 user), sensitivity to light (3 users), Nystagmus (2 users) and color vision deficiency (CVD) (4 users). Table 6 shows a detailed description of each user.

Table 6 Information of participants in the user test
Full size table

4 Results

Quantitative (user study results) and qualitative results (observations) are detailed as complementary views of the test.

4.1 User study results

As mentioned, effectiveness was measured dividing the number of completed tasks by the number of attempted tasks (percentage of completion per task). Efficiency was measured with time per task.

Table 7 shows all users’ average percentage of solved tasks for the accessible and non-accessible versions of each chart (effectiveness), all users’ average efficiency in seconds by type of chart and version and, finally, the median value.

Table 7 Effectiveness and efficiency by chart type and version
Full size table

The accessible version of the stacked bar chart and the line chart present greater effectiveness (88.33% and 93.75%) than the non-accessible versions (81.67% and 87.50%). On the other hand, the non-accessible bar chart shows a higher effectiveness than the accessible one (98.33% vs. 91.67%).

In terms of efficiency, the accessible versions of the bar chart and the stacked bar chart are superior to the non-accessible versions (21.3 and 23.2 s vs. 44.58 and 33.38 s). However, the non-accessible version of the line chart presented greater efficiency compared to the accessible version (23.48 vs. 25.56 s).

As a relevant observation it must be considered that the floating windows of the video conferencing tool sometimes overlapped with the charts, forcing some users to spend part of the time moving them, with a negative impact on the time count.

It must also be taken into consideration that in the line chart, the time required for one of the users, far above the average, has increased the overall time count.

Related to satisfaction the metric was “measure of expectations”, i.e. users were asked to rate their expected task complexity on a scale from 1 (not at all complicated) to 5 (very complicated), and after completing the task, they were also asked to rate the actual complexity they experienced using the same 1 to 5 scale.

The comparison between expectations and experience [93] is clearly favorable, being most charts in the quadrant of “promote-it” (Fig. 3), meaning that the users got better results than expected and as such, were very satisfied, while in the case of non-accessible charts the comparison between expectations and experience puts the experience in the quadrant of “big opportunity” (Fig. 3), meaning that the expectations are so low that small improvements can bring great results. Tables 8 and 9 show the expected and the experienced satisfaction by chart type, respectively.

Fig. 3
figure 3

Measure of expectations with accessible and non-accessible charts

Full size image
Table 8 Expected satisfaction by chart type
Full size table
Table 9 Experienced satisfaction by chart type
Full size table

When asked which version of each chart users found easier to use, most users preferred the accessible version over the non-accessible one (86.11% vs. 13.89%), reinforcing the satisfaction results, except for user 7 (stacked bar chart), user 8 (line chart), user 10 (bar and line charts), and user 11 (both bar charts).

“Interacting with the charts (referring to accessible charts) has allowed me to obtain the data you requested me more quickly. In the case of those that are not accessible, I have had to make an additional effort” (user 5).

4.2 Observations

In this section, we offer a comprehensive explanation of results, strategies, and difficulties experienced by users. This information is not readily available in existing literature but proves invaluable for designing more specialized tests or categorizing users into specific groups for more relevant comparisons. Additionally, this in-depth understanding will greatly inform the design of accessible charts and aid in the development of effective solutions or approaches.

4.2.1 Observations by user

Given the significant diversity in profiles, contexts of use, preferences, and assistive technologies employed by the users, Table 10 presents the details of the observations of users’ interactions with the charts throughout the test. Strategies are marked-up with the style subtle-emphasis and difficulties marked-up with emphasis style, to facilitate skim reading through the table.

Table 10 Qualitative observations per user and chart
Full size table

4.2.2 Other observations

The use of color (H10) in the non-accessible versions of the three charts has been a barrier for users 6, 8 and 9. In these three cases, the accessible version, with greater contrast and with patterns as an alternative to color, has allowed them to complete the tasks in a shorter amount of time. However, some users preferred the use of colors instead of the white, black, and grey version of the accessible version (1, 2 and 7). In particular, user 11 has highlighted that the absence of color and the interactivity (H17) implemented had not benefited him. The same user also highlighted that the use of patterns confuses him. User 6, affected by CVD, and user 9, with achromatopsia, hold a completely contrasting viewpoint on this matter.

“The interactivity of the chart facilitates its use, but it is better in color than in black and white” (user 1).

“I prefer colors than textures or patterns” (user 3).

“Due to the type of vision loss, I have, the color suits me very well” (user 11).

“As I have achromatopsia, I find it very useful that the bars have patterns to better distinguish them” (user 9).

“In the case of stacked bar charts, it is essential to have high contrast colors to be able to differentiate between the two sections. Patterns seem a good solution to me.” (user 6).

Among the magnification options (H14), we find two differentiated strategies depending on the user: (a) use of the operating system’s magnifying glass or screen magnifier; (b) use of the browser zoom. In the first case, resizing means losing certain parts of the chart and, with them, important information to carry out the proposed tasks. This situation has been the case for users 1 and 2 (could not locate the legend) (H2). In those cases, when the task involves making a comparison between data, they are forced to memorize the first value and look for the second by scrolling through the screen. In the second case (magnifying with the web browser), the accessible version adjusts its size to the window width after applying the zoom, allowing users to see the entire chart on the screen, but not certain elements that accompany it, such as the table with the data source (H6) or the legend (H2). Thus, the accessible versions facilitate comparisons within the chart. In this sense, tasks focused on comparing data have been performed better (more efficiently) with the accessible versions of the charts.

In the accessible versions of the charts, a tooltip functionality has been implemented to provide the value of the selected mark (bar or point) as an alternative to legends (H2). Tooltips have been useful for all users, except for user 9 who preferred to use the data table (H6). This functionality, used by almost all users, has been highly valued by users 1, 3 and 5, while users 6 and 8 in the interaction with the line chart have highlighted the fact that the tooltips obscured the chart preventing them from following the lines and seeing the marks, especially after magnifying the screen. In this case, the accessible chart does not meet the dismissible requirement associated with the success criterion 1.4.13 (Content on hover of focus) of the WCAG 2.1 [82], which could solve the difficulty mentioned by users.

“In the case of interactive charts, you have all the information at your fingertips. You consult a point and see all the related information without looking at two places at the same time.” (user 11).

“Tooltips are nice, but with a zoom applied they cover too large a part of the chart. It would be interesting if they were optional, for example, that they only appear after clicking on them.” (user 6).

“With the second type of charts (referring to the accessible charts) I have not needed to use zoom at any time because the data is near the bars and points. I simply needed to approach a little more on the screen.” (user 12).

All users have followed the strategy of following the axes (H3) and the marks with the cursor pointer.

When bitmap images (Microsoft Excel exports) were resized, the problem of their low quality was more pronounced, creating legibility problems (H12) for users 2, 3, 4, 5, 6, 7, 8 and 9 (Fig. 4). For user 2 there was even a problem differentiating the bars of the first chart due to the poor quality of the image. Specifically, he stated “it seems to be missing pixels”. Insufficient contrast (H11) between the text color used by default by Microsoft Excel and the background has also been a barrier for these users even after being resized. In all these cases, the users have solved the tasks using the data table (H6) available, not the chart.

“In the first versions (referring to the non-accessible charts) I noticed fewer sharp charts and worse contrast. The poor sharpness of the text also affected me a lot.” (user 8).

Fig. 4
figure 4

Detail of the low quality of the non-accessible bar chart resized

Full size image

In all cases, users have initially used the chart to solve the tasks. Only when they have been unable to find the answer, they have used the data table to find it, or they have used it to confirm their answers before verbalizing it (users 5, 6 and 7). User 6 was the only one to recognize that he preferred to consult the table rather than the chart in all cases. Users 7, 8 and 9 (the two last due to poor color perception) found that their efficiency improved when using the data table after the first task and have used it more frequently since then.

“It is easier to see the data in a chart than in a table” (user 1).

“In some cases (referring to non-accessible charts) I’d rather go to the table and deal directly with data than consult the chart, because the lines and dots are too thin for me” (user 5).

“When I read a scientific paper (the user is a researcher in the field of genetics), I never consult the charts because they are totally inaccessible. At most, they worry about color blindness, but not about other issues that affect people with low vision. I always prefer data tables than charts”. (user 6).

“For all the questions I needed to consult the table. I always use the tables to solve this kind of situations.” (user 9).

Every user, except for user 9 (who exclusively relied on the data table), frequently used legends to interpret the data (H2). Throughout the test, we observed difficulties in locating the legend when it was not positioned at the bottom of the chart or when it went off the screen due to applied zoom. In this regard, the suggestion put forth by Evergreen and Metzner [94] to label data directly, in close proximity to data points (such as on top of or beside bars and next to lines), can not only reduce cognitive load and facilitate more efficient information processing but also aid users with low vision in comprehending data series without the need for constant scrolling through the interface. We have also observed difficulty in differentiating the data series if the color was not sufficiently distinguishable (H10) or the size of the legend was not sufficient (fonts set to Calibri, 9 pt. In not-accessible charts).

In accessible versions of the bar chart and the stacked-bar chart, when a data series receives the focus (H16), the rest of the bars are displayed with less contrast to highlight the active element. This has been a barrier for user 10, who has expressed that it has confused him.

5 Discussion and limitations

The paper describes the results of an ongoing study that aims to verify a list of heuristics with users. The relatively small number of users does not allow to statistically validate the results nor to generalize them to the whole population, but the authors consider that the insights collected with this first approach are relevant and give light to barriers and priorities.

The observations made during the test heightened the authors’ awareness of the diverse preferences and strategies within the low vision users’ group, emphasizing the necessity to gain a deeper understanding of these interactions. Consequently, the authors have chosen to incorporate a thorough description of each user’s results, strategies, and preferences in the article. This information is deemed invaluable for future research in the field.

Due to COVID lockdowns, the tests were conducted remotely, and the interface of the video conferencing platform occasionally disrupted the efficiency of users. Some users had to spend part of their time minimizing the platform’s interface. This is a significant consideration for the authors and may also be a determining factor to address in future tests.

From the results obtained, we can derive some insights: a global view of the chart is very informative and users rely on it for comparative evaluations and trends; they look at it with a size that fits on screen, with not much zoom; instead the use of zoom is very important to read text, axes, labels, legends or tooltips, with many users using levels of zoom much bigger than the 200% level established by WCAG. A mechanism specific for these elements not affecting the chart should be developed and tested with users. Insight (a): a zoom option specific for text elements not affecting the chart should be devised, and it shall be flexible for zooms over 200%. Tooltips have been an important source of information but also created some problems obscuring parts of the charts. Insight (b): offer zoomable tooltips (see insight a), but callable on demand. Black and white charts did not satisfy some users, emphasizing the need to redundantly encode categories with both colors and patterns to cater to diverse user profiles and their preferences. Insight (c): offer color categories with high contrast, plus patterns, better than black and white coloring. The data table emerged as a useful alternative, particularly in non-accessible versions, enhancing task efficiency for many users who have solved the tasks by combining both the chart and table. Insight (d): always offer data in a table as a complement. On the observations many users follow axes to find a specific value in the chart. Insight (e): include axes in the charts as guides for locating specific points. Challenges were identified regarding the legend’s position and size, and addressing these aspects is essential to eliminate potential barriers. Insight (f): provide a legend to understand the data. Ensure that its position, size, colors and contrast do not impose any barrier to users.

The test is proof that in most cases, users prefer to solve tasks using the chart, even if it is not accessible, instead of using the data table. This confirms the results of other studies in which the use of the residual vision was preferred over other strategies [65, 66].

The tooltips, which, as we have highlighted previously, have been highly valued by all users, have been shown to be useful for: (a) giving direct access to the data associated with each mark, avoiding forcing users to consult the data table; and (b) serving as an alternative or complement to the legend. However, tooltips generated by Highcharts library do not comply with the accessibility recommendations of WCAG 2.1 and part of the literature [95, 96], as it is not possible to hide them in case of overlay with other elements.

The order of the bars was key to interpret the time series data for users 3, 4 and 6. The recommendation of sorting the axes chronologically is also cited in the literature [97, 98].

Users 1, 2, 4, 5, 7, 10, 11 and 12 solved the tasks using the browser zoom set between 110 and 200%. In the accessible version of the charts, this means that content needs to reflow to avoid horizontal scrolling, clipping, or overlapping of elements. This functionality associated with the responsive web design technique is implemented by the Highcharts library, but in some cases, it has presented some unexpected behavior that has involved usability problems that of course affects accessibility, such as some labels disappearing (see Fig. 5).

Fig. 5
figure 5

Detail of the accessible bar chart showing the absence of some labels

Full size image

Currently, Microsoft Excel does not provide accessible defaults for creating a new chart. However, it is possible to create fairly accessible charts. Exporting charts to non-Microsoft formats like HTML is also very problematic in terms of accessibility properties. Only an expert author will be able to create a moderately accessible chart.

The tasks were primarily centered on visual perception rather than testing data literacy and chart comprehension. Consequently, we designed the charts to be sufficiently clear and comprehensible for all users, regardless of their educational level. Notably, we found no significant differences in the results between users with the lowest educational attainment and the rest of the users.

At the outset of this publication, we conducted a theoretical comparison between the heuristic indicators proposed by Elavsky et al. [80] and our own set of heuristics. Although Elavsky and colleagues’ work is highly relevant, our user tests were conducted prior to its publication, preventing a direct comparison. However, the theoretical analysis revealed that Alcaraz et al.’s set of heuristic indicators encompasses all the principles proposed by Elavsky et al. [80]. Notably, Elavsky’s list comprises a larger number of more specific heuristics and aims to address a broader range of disabilities, whereas Alcaraz’s heuristics are tailored to the specific needs of users with low vision.

The list of heuristic principles associated with this research also includes certain principles that provide advantages or help in overcoming accessibility challenges for individuals with various disabilities, including those who are blind, have motor or cognitive impairments. These principles, considered as a universal set of best practices, offer benefits to the wider public, irrespective of their disability status. Table 11 provides a summary of these principles.

Table 11 Summary of the advantages linked to adhering to the suggested heuristic indicators for different user profiles
Full size table

The results of the user tests underscored the importance of incorporating tooltips or directly labeling data on the chart’s marks as an alternative or complement to using legends (H2). Tooltips and direct data labeling provide users with immediate access to data associated with each mark, eliminating the need to consult the data table or scroll through the interface. Additionally, they reduce cognitive load and promote more efficient information processing [94]. Data labels can also draw attention to specific data points, making them valuable when data values are essential [98]. However, when implementing tooltips, the following considerations should be taken into account: (a) tooltips should be hidden by default; (b) their use should be restricted to situations where concise and useful information is provided; (c) consistency in their usage across all charts is crucial; (d) ensure compatibility with both mouse and keyboard interactions; (e) use arrows, akin to comic bubbles, to guide users to the relevant element; (f) maintain sufficient contrast for readability; (g) avoid obstructing or concealing other related elements with the tooltip [99].

Furthermore, based on the test results, we will introduce two new requirements to enhance the H2 heuristic concerning legends: (a) The legend must be of sufficient size to enable users to distinguish the colors or patterns associated with each mark effectively; (b) The legend must be positioned either at the bottom of the chart or in a standardized and highly visible location.

6 Conclusions and future work

The user test aimed at validating heuristic indicators for assessing the accessibility of statistical charts has provided valuable insights. The study involved 12 users with different low vision conditions, and the accessible versions of charts demonstrated superior efficiency, effectiveness, and user satisfaction. The evaluation conducted using the think-aloud method allowed us to visualize, from the users’ perspective, those elements that constitute barriers to task completion, as well as the strategies that each user employs to overcome them. Another crucial aspect of these user tests has been the opportunity to comprehend specific preferences that may not necessarily align with the best practices commonly acknowledged in the existing literature. A notable example is the preference for color charts, even among users with CVD.

From a qualitative point of view, the heuristics related to legends (H2), axes (H3), data source as a data table (H6), safe colors (H10), contrast (H11), legibility (H12), image quality (H13), resize options (H14), focus visibility (H16), avoid disturbing elements (H15), and independent navigation (H17) proved to be crucial for task performance.

Legend (H2) is essential to understand the data. Its position and size, as well as the colors (H10) and contrast (H11) used, can negatively influence the effectiveness and efficiency if they are not designed following accessibility guidelines. On the other hand, labelling the values directly in the chart marks or implementing tooltips are even better alternatives. For users with CVD, it is essential to use safe color combinations or patterns to differentiate the marks. However, the combinations based on white, black and grey produce an effect of visual saturation in certain users, especially in those who preserve the perception of color. Considering the suitability of color to encode categories [100] and that some users prefer it over monochrome interfaces, a possible conclusion of the test is the need to redundantly encode categories with colors and patterns as well, to target all profiles.

Of equal importance to the legend are the titles of the axes (H3). Both have been used by all users to understand the data. Using vertical text on the y-axis does not seem to have been a problem for any user. On the contrary, low-quality images of text hinder the legibility of the legend and axes text (H13).

Providing access to the data source as a table (H6) allows users to have a highly efficient, fully text-based alternative when the task involves searching for a particular datum. Also, as observed during the test, it is useful to verify an answer before delivering the task. For this reason, and considering the challenges faced by individuals with more severe low vision in accessing charts, the presence of an equivalent table becomes indispensable.

Another common barrier has been insufficient image quality (H13) of non-accessible charts to cope with demanding resizes (up to 500%) (H14). In such cases, legibility (H12) is compromised and the use of charts in vector format is the best alternative because they can be enlarged as much as necessary without losing quality [37]. Another of vector charts’ advantages is their complete integration with the Document Object Model (DOM), that grants the ability to manipulate and customize them as any other HTML element and makes them compatible with assistive technology [101, 102].

Other works highlight the difficulties that users with low vision experience when interacting with screen magnifiers [103,104,105], because they only have a partial view of the page they are interacting with, and this can cause loss of context since not all the elements necessary to interpret or interact with the content are displayed on the screen. This is a common issue when interacting with a chart whenever the task requires comparing data. This requirement seems to lead to designs with reflow, to avoid horizontal scrolling, clipping, or overlapping of elements (H14), but this only worked for users using browser zoom and not for those using screen magnifiers with magnifications much greater than 200%.

The heterogeneity of needs and preferences among participants leads to test personalization techniques (H18) as a key factor to ensure the best accessibility in the greatest number of possible situations. However, as other works point out [63] one single method of adapting the presentation of the charts may not be sufficient to meet all the requirements for people with low vision.

With the user test conducted in this research, we have successfully followed all the steps outlined in the methodology by Quiñones et al. [81], affirming the validation and reliability of our heuristic set.

In these tests, the authors decided to start with simple charts. With more complex charts it might be possible to find a larger number of barriers (this was even mentioned by users 2 and 8).

As a future research direction the authors aim to test how complexity affects the barriers encountered by the users and also the effect of customization options (H18), to allow users to change mark colors, font style and font size, among others.

The main line of future work is trying to recruit new users, to cover most low vision profiles to continue reviewing the list of heuristic indicators and improve it by refining the guidelines and doing a new iteration in the definition and scoring of the heuristic set. Further work is required to plan other types of tasks that allow validating some of the heuristics not contemplated in this study (H1, Title; H4, Caption; H5, Abbreviations; H7, Print version; H8, Short text alternative; H9, Long description; H18, Personalization).