Exploring the Use of Eye Tracking to Evaluate Usability Affordances: A Case Study on Assistive Device Design

Abstract

This study explores the application of Eye-Tracking technology for the ergonomic evaluation of assistive device usability. Sixty-four participants evaluated six jar-opening devices in a two-phase study. First, the participants’ gaze was recorded while they viewed six rendered pictures of assistive devices, each shown in two different versions: with and without rubber in the grip area. Second, the participants physically interacted with the devices in a hands-on usability task. In both phases, participants rated the devices according to six usability affordances: robustness, comfort, easiness to grip, lid slippery, effort level, and easiness to use. Eye-Tracking metrics (fixation duration, number of fixations, and visit duration) correlated with the on-screen ratings, which aligned with ratings after using the physical devices. High ratings in comfort and effort level correlated with more visual attention to the grip area, where the rubber acted as key signifier. Heatmaps revealed the grip area as important for comfort and easiness to use and the lid area for robustness and slipperiness. These findings demonstrate the potential of Eye Tracking in usability studies, providing valuable insights for the ergonomic evaluation of assistive devices. Moreover, they highlight the suitability of Eye Tracking for early-stage design evaluation, offering objective metrics to guide design decisions and improve user experience.

1. Introduction

Assistive devices (ADs) for manipulative activities of daily living can support individuals with impaired hand function due to injury, pathology, or ageing, which often affects grip strength, force, or movement control [1,2,3]. People with hand osteoarthritis—a leading cause of musculoskeletal disability in Western countries—find tasks requiring strong grip and twisting motions, such as jar opening, especially challenging [4,5,6,7,8]. ADs reduce the physical demands of tasks by compensating for limitations in motion, strength, and dexterity, thereby promoting independence and quality of life [1,2,9,10]. However, many users hesitate to adopt ADs due to issues like cognitive requirements, difficulty in learning new devices, and concerns about social stigma [8,11,12,13,14,15].

Providing sufficient information about AD functionality is essential for product understanding [16,17]; elements such as colours, textures, and shapes should naturally suggest the product purpose and usage [16]. The more intuitive a product is, the lower the cognitive demands, resulting in better comprehension and more efficient use [18]. Understanding AD functionality is especially important given the specific needs of potential users, who must grasp how a product operates through clear, effective communication from the device itself. In other words, the use of an AD must be intuitive.

Product communicative functions [19] include indicative functions (technical and practical) and symbolic functions (cultural and historical messages). Norman [20] proposed a theoretical framework for designing products that clearly communicate their functionality, aligning with users’ mental models and prior experiences. Norman’s Fundamental Principles of Interaction emphasise concepts such as affordances, signifiers, constraints, mapping, and feedback, derived from Gibson’s [21] affordance theory but applied within product design contexts to ensure user understanding and usability. Affordances refer to the possible actions a product allows, enabling users to perform desired tasks. In the context of this study, we adopt a broader, operational use of the term affordance, referring to perceivable product characteristics that shape users’ expectations about how easily and effectively a device can be used. To further justify our broader use of the term, the recent literature in product and environmental design recognises affordances as relational properties between product characteristics and user capabilities, extending beyond the original theoretical definitions [22,23,24]. In this sense, aspects such as comfort or ease of grip, when perceived as enabling or facilitating user action, can be interpreted as affordances within an application-oriented framework. When affordances are not immediately clear, signifiers—design elements capable of communicating opportunities for product use—should be employed. Using signifiers is the most effective way to reveal affordances and ensure users understand how to interact with the product [20]. Building on these principles, Kapkın and Joines [25] explored the inter-relationships between perceived product meanings, proposing an ‘interactionist framework’ in which product forms guide user responses along dimensions such as ‘repulsive or inviting for interaction’ and ‘readiness for interaction.’ Their study, using hard drive and soap dispenser prototypes, highlights how product meanings are interconnected and influence usability, suggesting that designers should strategically assign meanings to product forms to enhance user interaction and comprehension.

Several techniques are used to evaluate the functionality conveyed by products, such as semantic differential [26], where participants provide subjective judgments on pairs of opposite adjectives (e.g., classic/modern) using a numerical scale. However, Eye-Tracking (ET) technology offers a more objective way to assess how a product communicates functionality by recording a user’s gaze position and movement to indicate where their attention is focused [27]. Gaze patterns are quantified through various objective parameters, such as fixations, saccadic movements, and scan paths [28].

ET has been used in evaluating product affordances and usability. Yoxall et al. [18] explored the affordance of easily opening food packaging by analysing the cognitive load required under conditions of cognitive distraction. They measured gaze diversion and task completion times, demonstrating a relationship between task difficulty and cognitive load. Burlamaqui and Dong [29] recorded subjects’ gaze while they observed products with innovative designs and used a questionnaire to determine whether participants had perceived the main affordance of each product. They defined the expected interaction area with the product as the area of interest (AoI), such as the bows of a pair of scissors, and analysed the number of fixations, the time to the first fixation, and whether it occurred within the AoI. However, they were unable to demonstrate that participants primarily focused on the expected interaction areas. Instead, their results suggested that affordances could often be perceived by attending only to certain elements of the product and that excessive visual exploration might indicate a lack of clarity in the design. This finding suggests that users do not need to observe the entire product or all its components to perceive its affordance; rather, attention to a few key elements can be enough. Berni et al. [30] explored functional affordances of design elements using ET, while Federico and Brandimonte’s [31] focused on usability affordances of tools, finding that fixation duration indicated greater attention on manipulation areas compared to other parts of the tools. Similarly, Guo et al. [32] investigated the relationship between ET metrics and user experience in product evaluation using smartphone pictures as stimuli. They found that products with a higher perceived user experience elicited a shorter time to first fixation, indicating quicker attentional capture. These studies support the use of ET to explore how users perceive and evaluate product usability. However, they also reveal methodological challenges and highlight that attentional patterns do not always directly reflect affordance perception. Despite these insights, research on affordance analysis through ET remains limited, particularly in ADs for manipulation tasks.

The primary objective of this study is to explore the effectiveness of ET technology as a tool for assessing the usability affordances of a specific category of ADs: those designed for jar opening. This study examines the role of rubber in the grip area as a signifier for various aspects of usability by comparing on-screen assessments of ADs with and without rubber. Heatmaps are used to identify other design elements acting as signifiers. Additionally, correlations between ET metrics and on-screen assessments are used to identify metrics that effectively gauge the usability of jar-opening ADs. Finally, on-screen assessments are compared with evaluations conducted after physically using the real devices to reinforce ET’s potential in analysing usability affordances. By addressing these objectives, this work seeks to contribute to product usability assessment by confirming the potential of ET technology and providing quantifiable metrics. Moreover, it highlights the suitability of ET for early-stage design evaluation, offering objective insights to guide design decisions.

2. Materials and Methods

2.1. Equipment and Participants

The experiment was divided into two parts: (1) evaluating products using ET through the visualisation of AD models and (2) evaluating the same ADs by physically manipulating and using them.

The Tobii Pro X2-60 device (Tobii AB, Danderyd, Sweden) was used for the ET experiment (gaze accuracy of 0.4° and gaze precision of 0.32°). It was positioned beneath a 24-inch monitor, recording data at a 60 Hz, with participants seated approximately 60 cm from the screen. Calibration was performed for all participants at the beginning of the test using a 9-point method, according to manufacturer’s guidelines. The ET experiment lasted approximately 10 min and the physical evaluation 30 min.

The Ethics Committee of Universitat Jaume I approved the experiment (approval number CD/79/2021), and all participants provided informed consent. Initially, 64 subjects (aged 18 to 66) participated, all with normal or corrected vision. Subjects with hand pathology were excluded due to their higher likelihood of prior experience with ADs, as the study aimed to assess users’ first impressions based solely on visual perception. For data analysis, participants with an ET registration percentage below 75% (as recorded by the Tobii Pro Lab software, version 1.145.28180 [×64]) were excluded. This resulted in the exclusion of three participants, with registration percentages of 28%, 46%, and 59%. One additional participant was excluded due to a power outage during the session. The final dataset included 60 participants (28 females and 32 males) with an average age of 40.5 ± 10.7 years.

2.2. Product Selection

Six different commercial ADs designed for opening jars were selected to cover a range of grasp types, usage movements, effort levels required during opening, materials, handle designs, and lid attachment mechanisms (

Table 1. The selected commercial ADs for jar opening, including their identification numbers (IDs), brief descriptions of their functionality, and the required grasp type for use, as defined by the grasp taxonomy proposed by Vergara et al. [33].

ID	ADs for Jar Opening	How to Perform the Opening Action		Grasp Type
1			Grip and twist counterclockwise.	Cylindrical grasp
2			Grip and twist counterclockwise.	Cylindrical grasp
3			Twist the handle counterclockwise until it engages with the lid. Once secured, apply pressure and continue twisting until the lid opens.	Cylindrical and oblique palmar grasp
4			Push and twist counterclockwise.	Oblique palmar grasp
5			Push and twist counterclockwise.	Oblique palmar grasp
6			Apply the non-slip tape to the lid and twist counterclockwise	Oblique palmar grasp

).

2.3. Eye-Tracking Experiment: Stimuli, Questions, and Timeline

To create the stimuli, 3D models of all six ADs were generated using SolidWorks 2022^® and then rendered in Blender 3.0^®. In this work, the term stimulus refers to each individual screen shown during the ET session, which records gaze data and includes a specific arrangement of AD pictures for a given task. The ADs were displayed in their usage position, with additional views highlighting the placement area on the lid. A second 3D model of each device was generated, with rubber added to the grip area to investigate its role as a signifier (Figure 1).

An on-screen usability assessment (UA1) was conducted during the ET experiment. Instead of the commonly used System Usability Scale (SUS) [34], the semantic differential method was chosen because it enables a more fine-grained evaluation of specific usability affordances, whereas the SUS offers a general overview of the overall usability of a product. This level of detail was essential to align with the structure of our ET analysis, which focuses on the interpretation of individual usability affordances. Following recommendations for semantic differential studies [26,35], an initial list of 86 pairs of opposing adjectives was refined using affinity diagrams to focus on descriptors relevant to jar opener usability. Ultimately, six usability affordances were identified: robustness (A1), comfort (A2), easiness to grip (A3), lid slippery (A4), effort level (A5), and easiness to use (A6). These dimensions reflect product qualities that may influence users’ perception of how easily the device can be used and were therefore operationalised as evaluative proxies for usability affordance perception. The models were displayed in their usage position (Figure 1b) for all affordances except A4, where the lid placement area was shown to assess slippage (Figure 1a). To minimise bias, affordance assessment was randomised, with the models repositioned on the screen for each affordance. Participants ranked the jar openers on each affordance by selecting the best, second-best, worst, and second-worst models by clicking with the mouse on the corresponding picture (

Table 2. Description of the ranking option for each affordance.

Affordance		Ranking Options
A1	Robustness	(i) The most robust; (ii) the second most robust; (iii) the least robust; (iv) the second least robust
A2	Comfort	(i) The most comfortable; (ii) the second most comfortable; (iii) the least comfortable; (iv) the second least comfortable.
A3	Easiness to grip	(i) The easiest to grip; (ii) the second easiest to grip; (iii) the hardest to grip; (iv) the second hardest to grip.
A4	Lid slippery	(i) The least slippery on the lid; (ii) the second least slippery on the lid; (iii) the most slippery on the lid; (iv) the second most slippery on the lid.
A5	Effort level	(i) The least effort required; (ii) the second least effort required; (iii) the most effort required; (iv) the second most effort required.
A6	Easiness to use	(i) The easiest to use; (ii) the second easiest to use; (iii) the hardest to use; (iv) the second hardest to use.

). After selecting each option (e.g., the best), the stimulus changed, but the picture displayed on the screen remained the same to assess the subsequent choice (e.g., the second best). Thus, while the picture on screen remained unchanged, the recorded stimulus (linked to a new ET dataset and a new selection instruction) was different.

Two timelines (Figure 2) were devised with identical questions, instructions, and stimuli, differing only in the models used. Each participant evaluated six jar openers in one timeline (assigned randomly): timeline A featured models 1, 3, and 5 without rubber and models 2, 4, and 6 with rubber, while timeline B presented the reverse. Each session began with a calibration procedure followed by an introductory slide with instructions. Before each affordance evaluation, a slide displaying a single blue spot at the centre was shown. Participants were instructed to fixate on this spot for 1.3 s before proceeding to the next slide, ensuring consistent starting points for gaze plots across all stimulus groups. The available time was unlimited and controlled by the participants through their mouse clicks when making selections.

Areas of interest (AoIs), defining regions within the observed picture according to specified boundaries, were established for each stimulus to analyse ET metrics and collect data from these areas (

Table 3. List of AoIs and their definitions, accompanied by an example image. In each example image, different AoIs are highlighted in distinct colors for illustrative purposes.

AoI Name	Definition	Example Image
AoI Stimulus	The entire stimulus.
AoI Model Image	The rectangle enclosing each model.
AoI Model	Each model shown in the stimulus.
AoI Grip	The gripping area of each model.
AoI Placement on Lid	The area of each AD that is placed on the jar lid.

). A padding of 4 to 5 mm was applied to the AoIs on the monitor, as recommended by Goldberg and Helfman [36], based on the ET model’s technical specifications and the distance between the participant’s eyes and the ET device.

2.4. Usability Assessment with the Real Products

After the ET experiment, a second usability assessment (UA2) was conducted in which participants physically interacted with the six commercial ADs for the first time. Up to this point, the participants had neither seen nor touch or handled the real devices. Seated in front of the ADs, before any manipulation, they were first asked if they were familiar with or had previously used any of them to verify that the intended participant profile (users without prior experience) was maintained. The participants then used each AD to open a jar with a torque set to 4 N·m, and it was noted whether or not they were able to open the jar. Finally, after using all the ADs, the participants answered the questions listed in

Table 4. Usability assessment questions performed to each participant after using all the ADs.

Affordance	Questions
A1	E1	Rank the products from the most robust (1st) to the least robust (6th)
A2	E2	Rank the products from the most comfortable (1st) to the least comfortable (6th)
A3	E3.1	Evaluate the grip area of each product: Would you prefer it to be wider or narrower?
A3	E3.2	Evaluate the grip area of each product: Would you prefer it to be longer or shorter?
A4	E4	Did any of the products slip from your hands or from the lid? Which one?
A5	E5	Rank the products from the one that requires the least effort (1st) to the most (6th)
A6	E6	Rank the products from the easiest (1st) to the most difficult to use (6th)

, related to the same affordances considered in UA1.

2.5. Data Analysis

2.5.1. ET Metrics

Table 5. Metrics calculated for each stimulus and participant, along with the reference values used for scaling.

Metric	Definition	Reference Value for Scaling Metrics Into %	Scaled Metric
DoI	Duration of interval (seconds): the total time spent observing the stimulus.	-	-
TDoF	Total duration of fixations in AoI (seconds): a fixation is an eye movement with a velocity of less than 30°/s.	DoI	TDoF%
NoF	Number of fixations in AoI.	NoF in AoI Stimulus	NoF%
TtFF	Time to first fixation in AoI (seconds).	DoI	TtFF%
DoFF	Duration of first fixation in AoI (seconds).	DoI	DoFF%
TDoV	Total duration of visits in AoI (seconds): a visit corresponds to all data (including saccadic movements, blinks, or invalid gaze data) from the start of the first fixation inside an AoI to the end of the last fixation in the same AoI.	DoI	TDoV%
NoV	Number of visits in AoI.	-	-

lists the ET metrics computed for each stimulus and participant. To ensure comparability across participants, some metrics were converted to percentage values (as detailed in Table 5), to account for the variability in decision times. To compare affordance transmission, we calculated the total time each participant spent ranking models for each specific affordance. This total, referred to as the duration of intervals across all stimuli for a given affordance (DoIA), was determined by summing the durations of intervals (DoIs) for the four stimuli shown during the ranking process. The DoIA was then expressed as a percentage of the participant’s overall response time across all affordances, resulting in the DoIA% metric.

2.5.2. Data Curation

Data screening excluded participants who made errors in response selection. Two participants were removed for not selecting a model within AoI Model. Additionally, two more participants were excluded for providing inconsistent ratings for the same affordance (e.g., selecting a model as both the most and second most preferred). These inconsistencies resulted from inadvertent or uncertain mouse clicks recorded through the ET system within the AoIs, reflecting momentary lapses in participant attention or accidental inputs. Ultimately, data from 56 participants (26 females and 30 males, aged 40.2 ± 10.2 years) were retained from the initial 60.

Only the first stimulus for each affordance (selection of the best model) was used to compute ET metrics, as this marked the point when participants formed their initial affordance judgment. In subsequent stimuli, it was assumed that participants had already formed opinions on all the models, leading to faster responses in general.

2.5.3. Comparative Analysis of On-Screen Affordance Transmission and Model Evaluation

The goal is to detect which affordances are easier or more challenging to convey from on-screen evaluation. Mean values for each question were calculated for DoIA% and used as dependent variables in Kruskal–Wallis tests (applied due to violations of variance homogeneity) with Bonferroni correction to check for significant differences (p < 0.05) between affordances. DoIA%, where lower values reflect easier transmission, serves as a clear indicator of affordance transmission.

For on-screen evaluation of models, ratings were assigned to ranks in UA1 as follows: +2 for the best, +1 for the second best, −2 for the worst, −1 for the second worst, and 0 for non-selected models. For each affordance, a non-parametric Scheirer–Ray–Hare test was conducted using ratings as the dependent variable and model, version (with or without rubber), and their interaction as factors (with p = 0.05). Additionally, a non-parametric Kruskal–Wallis test was performed as a post hoc analysis to compare significant differences between versions (with and without rubber) for each model and affordance, using rubber as a factor and ratings as the dependent variable. A separate Kruskal–Wallis test was also applied across models, regardless of the version, to examine significant differences in ratings based on model as the factor.

2.5.4. Analysis of ET Metrics for Evaluating Usability Affordances and Studying Signifiers

The aim is to analyse which ET metrics provide relevant insights for evaluating usability affordances. For each affordance, non-parametric Scheirer–Ray–Hare tests were performed on several ET metrics, following the confirmation of variance non-homogeneity. In the analyses, ET metrics (TtFF%, DoFF%, TDoF%, NoF%, TDoV%, and NoV) from the different AoIs (AoI Model, AoI Grip, and AoI Placement on Lid) were treated as the dependent variable, with model, version (with or without rubber), and their interaction included as factors (p = 0.05).

To further identify useful metrics for analysing usability affordances on ADs (specifically, metrics that provide insights into affordance ratings), Spearman correlations were calculated between the ratings from UA1 and ET metrics (TtFF%, DoFF%, TDoF%, NoF%, TDoV%, and NoV) for each affordance within the AoI Model, AoI Grip, and AoI Placement on Lid.

Finally, a qualitative analysis explored additional design elements (e.g., handle and placement areas on the lid) and characteristics (e.g., material, shape, and size) as potential signifiers for each affordance, supported by heatmaps of the first stimulus of each affordance. These heatmaps provide a chromatic representation of fixation duration across all participants and highlight areas that attracted the most attention.

The analyses were conducted using the statistical software SPSS v29 (IBM Corp., Armonk, NY, USA). For non-parametric analyses of multiple factors [37,38], MATLAB^® 2023b (MathWorks, Natick, MA, USA) was used.

2.5.5. On-Screen vs. Physical Assessment of Usability Affordances

Data from the UA2 questions (E1–E6) were analysed and compared to ratings from UA1. For questions E1, E2, E5, and E6, the model ranking (from 1 to 6) was used as the rating for the corresponding affordances. For question E3, the following ratings were assigned: wider (1) or narrower (−1) and longer (1) or shorter (−1). In question E4, the models were rated based on how much they slipped during use: significantly (2), slightly (1), or not at all (0). The ratings from E1, E2, E5, and E6 were analysed through non-parametric Kruskal–Wallis tests, using rating as the dependent variable and model as the factor, with a Bonferroni correction applied. Finally, the ratings from UA2 were compared with the ones obtained in UA1.

3. Results

3.1. Comparative Results of On-Screen Affordance Transmission and Model Evaluation

Table 6. Mean DoIA% values by affordance and homogeneous groups from the non-parametric test (p < 0.05). In homogeneous groups, each row represents a group with no significant differences between its affordances, ordered from the lowest to the highest value of DoIA%.

Affordance
Robustness	Easiness to grip	Easiness to use	Comfort	Lid slippery	Effort level
15.06	15.26	15.45	16.18	18.96	19.09
Homogeneous group
Robustness	Easiness to grip	Easiness to use	Comfort
				Lid slippery	Effort level

presents the results of the analysis of DoIA% to compare the easiness of affordance transmission.

The results of the Scheirer–Ray–Hare test for UA1 ratings, considering model and rubber factors (and their interactions), indicate significant differences (p < 0.05) between models for all affordances. Differences were also found between the versions without and with rubber in the affordances of robustness, comfort, easiness to grip, and effort level, which consistently favoured the models with rubber. However, the model–rubber interaction was not significant in any case. Figure 3 shows the 95% confidence interval plots for the mean ratings of each affordance for each model, both with and without rubber. The asterisk in the plots indicates significant differences (post hoc Kruskal–Wallis test) between the versions with and without rubber for each model and affordance. The results of the Kruskal–Wallis test applied across models to examine significant differences in ratings from UA1 are presented along with the heatmap results.

3.2. Results of ET Metrics for Evaluating Usability Affordances

Table 7. Significance levels for the model and rubber factors in the non-parametric tests of ET metrics in the AoI Model, AoI Grip, and AoI Placement on Lid. Significant values (p < 0.05) are underlined in the table.

Metrics	Affordances
	Robustness	Comfort	Easiness to grip	Lid slippery	Effort level	Easiness to use
AoI Model
	Model factor
TtFF%	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
DoFF%	0.435	0.753	0.496	0.848	0.705	0.818
TDoF%	<0.001	<0.001	0.001	<0.001	<0.001	0.001
NoF%	<0.001	<0.001	0.001	<0.001	<0.001	<0.001
TDoV%	<0.001	<0.001	0.003	<0.001	<0.001	<0.001
NoV	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
	Rubber factor
TtFF%	0.416	0.847	0.897	0.670	0.925	0.208
DoFF%	0.681	0.550	0.697	0.987	0.859	0.203
TDoF%	0.008	0.001	<0.001	0.007	0.194	0.273
NoF%	0.019	0.006	<0.001	0.029	0.093	0.060
TDoV%	0.004	0.001	<0.001	0.008	0.135	0.220
NoV	0.156	0.147	0.019	0.070	0.750	0.261
AoI Grip
	Model factor
TtFF%	0.691	0.003	0.504	0.532	0.013	0.105
DoFF%	0.836	0.897	0.964	0.999	0.888	0.986
TDoF%	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
NoF%	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
TDoV%	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
NoV	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
	Rubber factor
TtFF%	0.941	0.772	0.863	0.649	0.864	0.871
DoFF%	0.772	0.617	0.342	0.673	0.899	0.941
TDoF%	0.036	0.020	<0.001	0.055	0.007	0.497
NoF%	0.043	0.035	<0.001	0.091	0.001	0.460
TDoV%	0.036	0.015	<0.001	0.059	0.004	0.498
NoV	0.031	0.122	<0.001	0.089	0.025	0.536
AoI Placement on Lid
	Model factor
TtFF%	<0.001	0.054	0.007	0.006	<0.001	<0.001
DoFF%	0.809	0.994	0.785	0.919	0.842	0.565
TDoF%	<0.001	<0.001	<0.001	0.006	<0.001	<0.001
NoF%	<0.001	<0.001	<0.001	0.034	<0.001	<0.001
TDoV%	<0.001	<0.001	<0.001	0.004	<0.001	<0.001
NoV	<0.001	<0.001	<0.001	0.181	<0.001	<0.001

shows the significance levels for the rubber and model factors in the Scheirer–Ray–Hare test for each affordance, metric, and AoI. No significant results were found for the rubber factor within the AoI Placement on Lid (p > 0.05), so these results have been omitted in Table 7. The interaction between model and rubber was significant (p < 0.05) in only one case: for the effort level affordance within the AoI Model, specifically for NoF%, with a p-value of 0.042.

Table 8. Spearman correlation coefficients between the ratings and the ET metrics. A colour gradient by affordance, ranging from light to dark according to increasing values, is used, with non-significant values shown in white. All correlations are significant at p ≤ 0.01, except for those marked with an asterisk (p ≤ 0.05).

Metrics	AoI Model	AoI Grip	AoI Placement on Lid	AoI Model	AoI Grip	AoI Placement on Lid
	Robustness			Comfort
TtFF%	−0.122 *	−0.067	−0.116	−0.027	−0.045	0.086
DoFF%	0.096	0.089	0.114	0.114	0.125	0.142 *
TDoF%	0.514	0.240	0.448	0.510	0.414	0.284
NoF%	0.483	0.235	0.427	0.508	0.412	0.265
TDoV%	0.511	0.238	0.448	0.504	0.406	0.278
NoV	0.356	0.223	0.346	0.390	0.352	0.234
	Easiness to grip			Lid slippery
TtFF%	−0.051	−0.047	−0.065	−0.012	0.095	0.042
DoFF%	0.014	0.112	0.007	0.071	0.082	0.093
TDoF%	0.501	0.340	0.192	0.506	0.271	0.424
NoF%	0.489	0.328	0.176	0.482	0.274	0.374
TDoV%	0.497	0.341	0.190	0.503	0.270	0.420
NoV	0.357	0.285	0.134 *	0.290	0.249	0.234
	Effort level			Easiness to use
TtFF%	0.012	0.055	−0.007	0.263	0.161 *	0.261
DoFF%	0.086	0.075	0.080	0.012	0.007	0.069
TDoF%	0.530	0.397	0.307	0.491	0.308	0.224
NoF%	0.507	0.386	0.293	0.483	0.310	0.198
TDoV%	0.530	0.402	0.300	0.494	0.310	0.221
NoV	0.313	0.314	0.258	0.282	0.278	0.150

presents the Spearman correlation values between the UA1 ratings and the ET metrics for each affordance in each AoI. Non-significant values are shown with a white background, while significant values are indicated with a colour gradient by affordance, with darker shades representing higher correlations.

Figure 4 presents the heatmaps for each affordance. The results of the Kruskal–Wallis test show a significant effect in model rankings across all affordances (p < 0.05), with homogeneous groups highlighted in Figure 4: models with identifiers in red and green represent the worst-rated and best-rated groups, respectively.

3.3. Results of On-Screen vs. Physical Assessment of Usability Affordances

None of the participants failed to open the jar with models 1 and 6, whereas 4% were unable to do so with model 2, 5% with models 3 and 4, and 11% with model 5.

Figure 5 shows the 95% confidence interval graphs for mean ratings obtained from the usability assessment after using the real devices (UA2). In addition, for those affordances where the models were ranked from best to worst (E1, E2, E5, and E6), the significance level of the non-parametric Kruskal–Wallis test and the homogeneous groups of models (from best to worst) are displayed next to the graph. In order to compare results of both usability assessments, the model ranking obtained from the on-screen usability assessment (UA1) is also included in this figure.

4. Discussion

This study aimed to explore the effectiveness of ET technology for assessing the usability affordances of jar opening ADs. Interestingly, we have identified TDoF%, NoF%, and TDoV% as key metrics for assessing usability affordances in ADs. We also found that the variation in evaluation times across affordances suggests that some affordances are less effectively communicated through the models’ pictures and would benefit from clearer signifiers. Specifically, lid slippery and effort level are less effectively communicated, whereas robustness and easiness to grip are conveyed more effectively. In this sense, we have analysed the potential impact of adding rubber to the grip area as a signifier. Non-parametric multi-factor testing revealed that some models were rated more favourably than others depending on the affordance, confirming that the selected models effectively represent different levels of each affordance. Additionally, the presence of rubber consistently enhanced affordance clarity across all models. The lack of significant interaction effects indicates that rubber has a uniform positive impact, enhancing the evaluation of the affected affordance regardless of the model.

4.1. About the Use of ET Metrics for Evaluating Usability Affordances

The study sought to identify ET metrics that effectively gauge the usability of jar opening ADs. In this sense, significant differences in metrics aligned with differences in on-screen assessment ratings, along with significant correlations between these metrics and affordance ratings. Specifically, fixation and visit metrics, such as TDoF%, NoF%, TDoV%, and NoV, demonstrated significant correlations with on-screen assessment affordance ratings, suggesting their relevance for assessing usability affordances. Although the correlation coefficients were not high—a common feature of Spearman’s coefficients—they were significant and consistent with the hypothesis that higher affordance ratings correspond to more frequent and prolonged visual attention. In particular, TDoF%, NoF%, and TDoV% showed the highest correlations within the AoI Model. This result aligns with prior research on product selection suggesting that products receiving higher attention are more likely to be selected [39,40]. In contrast, metrics related to the first fixation—such as TtFF% and DoFF%—provided limited information, with low and non-significant correlations with on-screen assessments. This limitation may be due to other factors unrelated to affordance perception. ET records revealed that participants’ initial fixations often centred on the screen due to the gaze restriction blue dot used before affordance assessment, potentially distorting these metrics for centrally located models. Future studies should ensure that models are not placed in the central region of visual stimuli.

The differences observed in fixation and visit metrics across AoIs align with the findings of Burlamaqui and Dong [29], who argued that affordances can be perceived through visual inspection of critical product elements. For affordances such as comfort, easiness to grip, effort level, and easiness to use, the AoIs Grip and Placement on Lid were the most observed, as these regions convey essential information for the evaluation of these specific affordances. The results also corroborate findings by Federico and Brandimonte [31], who emphasised that fixation times are prolonged in areas associated with product manipulation, a pattern they observed in usability affordances of tools, where the manipulation zones attracted significantly more attention compared to other parts of the tools.

In summary, this study has identified TDoF%, NoF%, and TDoV% as key metrics for assessing usability affordances in ADs, while first fixation metrics offer limited utility due to contextual bias in this study. The addition of rubber to the grip area significantly improved on-screen assessment affordance ratings, as reflected in higher fixation and visit metrics. These findings underscore the importance of targeted AoI analysis in uncovering the visual strategies underlying affordance perception. Future research should refine experimental designs to minimise biases and further explore the interplay between ET metrics, product characteristics, and perceived affordances.

4.2. About Signifiers of AD Usability

A detailed examination of rubber’s role as a signifier for each affordance (Figure 3) reveals several positive effects. Models with rubber generally received higher ratings from on-screen assessment, particularly where rubber’s physical properties were most advantageous. For instance, rubber enhanced robustness of the grip area, especially in model 6, by providing a sense of increased thickness. Comfort improved in the rubber versions (significantly in models 1, 5, and 6), likely due to the perceived softness, reducing pressure points and providing greater comfort. Rubber also clarified the intended grip area, as in models 1, 2, and 6, resulting in perceiving an improved ease of grip. In model 1, the rubber version was rated higher for lid slippery, as the enhanced grip is perceived to reduce slippage and facilitate tighter sealing. Additionally, the rubbered model 1 was also rated more positively for easiness to use, suggesting that the defined grip area clarified how to operate the model. Finally, no significant differences were observed between rubber and non-rubber versions for the affordance of effort level, which was one of the least effectively conveyed.

To conduct a more detailed study of other design elements that function as signifiers, a qualitative analysis by affordance has been carried out, supported by participants’ visual attention from heatmaps (Figure 4). Comments follow:

Robustness. Participants focused primarily on the placement area on the lid, highlighting its importance in evaluating this affordance. Material and thickness notably influenced perceptions: in models 2 and 3, the steel material served as a positive signifier, while in model 4, the thick plastic conveyed solidity. Additionally, the solid and somewhat bulky shape of model 4 may have further reinforced the perception of robustness.

Comfort. The grip area received the most attention, particularly in models 1 and 6, which are elongated, plastic, and feature large grip areas, characteristics that likely act as comfort signifiers, especially with rubber reinforcement. In contrast, model 2, rated lowest for comfort, showed fewer fixations in its rubber version, possibly due to its slim design and steel material, which may have been perceived as uncomfortable when pressing the hand.

Easiness to grip. Attention across AoIs varied by model: in model 1, the grip area received most attention; in models 5 and 6, focus centred on the placement area on the lid, while in model 3, both areas were equally observed. Models 2 and 4 showed similar pattern, with attention distributed between both areas. The handle design in model 3, with its familiar shape and material, likely acted as a signifier of easiness to grip. In contrast, model 2, perceived as the hardest to grasp, received less attention, possibly due to its wire-like shape and limited support surface, leading participants to dismiss it early.

Lid slippery. Attention primarily focused on the placement area on the lid, with minimal focus on the grip area. Higher-rated models displayed features enhancing adhesion perception: models 2 and 3, with metal contact areas and protruding tabs, were likely perceived as less slippery due to these elements (Figure 6). Similarly, models 4 and 6, featuring soft, rubber-like plastic in the contact area, acted as a positive signifier of adhesion. In contrast, model 1 was perceived as more slippery due to its smooth hard plastic and model 5 due to its toothless metal surface (Figure 6).

Effort level. Model 2, one of the best rated, attracted more attention to the grip area, while the lower-rated models were more observed in the placement area on the lid. Regarding signifiers, model 2’s metallic and lever-shaped design suggests reduced effort. However, model 1, despite a similar lever shape, was rated as requiring more effort, possibly due to its plastic material, perceived as a negative signifier. For model 3, the rating may reflect the short handle length, limiting leverage and acting as a negative signifier.

Easiness to use. Attention patterns revealed that in model 1, focus was directed towards the grip area, while in models 4 and 6, greater attention was given to the placement area on the lid. Higher-rated models like model 1, with its longer grip and larger palm area, drew more attention to the grip. In model 4, the lid area could also serve as a grip, and model 6 had multiple placement zones, explaining the attention on these areas. In contrast, lower-rated models 2 and 3, perceived as harder to use, had complex mechanisms that acted as signifiers difficult to interpret. Overall, the placement area on the lid was observed more than the grip area, indicating that easiness to use is better understood by recognising how to place the opener on the lid rather than how to hold it.

Table 9. Key areas in evaluating affordances and signifiers influencing perception of ADs.

	Affordances
	Robustness	Comfort	Easiness to grip	Lid slippery	Effort level	Easiness to use
Key areas		AoI Grip	AoI Grip		AoI Grip	AoI Grip
	AoI Placement on Lid		AoI Placement on Lid	AoI Placement on Lid	AoI Placement on Lid	AoI Placement on Lid
Signifiers	Rubber	Rubber	Rubber	Rubber		Rubber
	Material	Material	Material	Material	Material
	Thickness
	Shape	Shape	Shape	Shape	Shape	Shape
						Number of elements

summarises the key areas in evaluating affordances and the signifiers that influence perception of ADs.

The results from the usability evaluation with real products (UA2) largely aligned with the on-screen evaluation performed during the ET recordings (UA1), reinforcing the reliability of on-screen presentation of products and, consequently, of ET for assessing usability affordances. Consistent findings were observed for comfort, easiness to grip, and easiness to use, indicating that these affordances were adequately assessed through ET. However, discrepancies emerged for robustness, where model 1, initially perceived as fragile in ET due to its plastic appearance, was rated higher upon physical interaction. Similarly, for lid slippery, models 2 and 3, considered non-slippery in ET, were found to be slippery during real use due to material properties, such as the lack of adhesion from metal tabs. For effort level, although model 2 was highly rated in on-screen assessment, its ranking during real use was lower, likely due to a less functional lever design that complicates gripping far from the jar. In contrast, model 1, which received a negative rating from on-screen assessment, performed best during real use, as its soft material reduced perceived effort compared to the hard plastic observed visually. Some of these differences highlight the limitations of visual-based methods in capturing material properties and physical interactions. Further research should be conducted on how to use images whose assessments align with real-use evaluations.

5. Conclusions

In conclusion, this study demonstrated the utility of Eye-Tracking technology for analysing the usability affordances in assistive devices, identifying design elements that act as signifiers and highlighting relevant ET metrics for evaluation. Some usability affordances, such as lid slippery and effort level, proved more challenging to assess visually, emphasising the need for clear signifiers to improve user understanding.

The rubber in the grip area of the assistive devices acted as a signifier of robustness, comfort, easiness to grip, and effort level, drawing more focused attention and showing higher values in fixation and visit metrics. However, early fixation metrics did not provide useful information for evaluating the effect of rubber. The hypothesis that higher-rated affordances attract more visual engagement was confirmed, with the models rated higher in comfort and effort level receiving more attention.

The analysis, based solely on Eye-Tracking data from the first stimulus (the question about the best model), revealed significant correlations between Eye-Tracking metrics and usability ratings in the areas of interest. Specifically, metrics such as total fixation duration, number of fixations, and total visit duration showed a positive correlation with better usability perception. The number of visits also exhibited a positive, albeit weaker, correlation. However, the first fixation metrics were not informative in this context.

Heatmap analysis highlighted the grip area as crucial for assessing comfort and easiness to use, while the lid area was more relevant for robustness and lid slippery. In general, design elements such as rubber, material, thickness, model shape, and number of elements serve as signifiers that help convey each affordance.

Overall, the after-use assessment aligned with the Eye-Tracking findings, except for models with material modifications. This reinforces Eye-Tracking’s value in studying assistive devices design and usability, offering a solid foundation for future ergonomic and functional improvements.