1. Introduction
Fitts’ law has been a successful and broadly applied tool to evaluate human performance in operating specific devices since Card’s work [
1]. In the original Fitts task, the target was a long strip, and the movement of the task could be a reciprocal [
2] or a discrete movement [
3]. Two identical targets were located parallel to a movement amplitude (
A) in front of the participants. Participants moved the pointing device above a horizontal line from right to left or left to right. The short side of the stripe was in the direction of movement, and the long side was perpendicular to the movement direction. We call the length of the short side the width (
W) and the length of the long side the height (
H) of a target. The height was always much greater than the width in the Fitts paradigm. The amplitude and the width of the target derive a particular index of difficulty (
ID) in the Fitts paradigm, which was based on the information theory. Fitts claimed that the movement time (
MT) from a start point to hitting the target was proportional to the index of difficulty [
2].
Extending the Fitts paradigm, researchers investigated the effects of height [
4,
5] and target location [
6]. Accot and Zhai [
5] called this new type of Fitts task, wherein the width is the only constraint to the movement (
H >>
W), the one-dimensional (1D) model. In addition, when the height perpendicular to the movement direction is also a constraint to the movement (
H ≤
W), the target is two-dimensional (2D). MacKenzie and Boxton studied the targets presented at an angle to the horizontal line; this situation is also called a two-dimensional task [
6]. One concerns the dimensions of the target; the other concerns the dimension of the movement. Hoffmann et al. [
7] provided an excellent definition of one-, two-, and three-dimensional targets if the limitation of movement performance is in the direction of movement, perpendicular to the movement, and at the proper depth at the target location.
In the current research, we adopt the definition of the dimension of movement by [
7]. For the dimension of the target, we choose the meaning by [
6]; when the width is the only constraint to the movement (
H >> W), it is a 1D target. In contrast, the target is two-dimensional (2D) when the height perpendicular to the movement direction is also a constraint to the movement (
H ≤ W).
Figure 1 illustrates the 1D and 2D targets; the 2D targets are the equilateral triangle, diamond, circle, and square, all with equal height and width. Murata and Iwase [
8] and Cha and Myung [
9] defined the condition of the start point and the target not being on the transverse or frontal plane simultaneously as a 3D movement. This study defines the tapping task as a one-dimensional movement when the targets (or the start position and the target) are on the line parallel to the X-axis of a plane and which plane is parallel to the transverse or frontal plane. Otherwise, the targets are not on the line parallel to the X-axis; the tapping is a two-dimensional movement. When a movement is neither one-dimensional nor two-dimensional, it is a three-dimensional movement with the targets in the different depths of a human. Accordingly, this research classifies Fitts’ original paradigm as a one-dimensional movement in a one-dimensional target (1D1D) task. Studies on 1D2D [
10], 2D2D [
6], and 3D2D [
8,
9] tasks can be found in the literature.
The Canon model of Fitts’ law is a regression formula derived from empirical data. It regulates the relationship between the dependent variable (
MT) and the independent variable (
ID) with two parameters, the intercept (
a) and slope (
b):
The
ID has various definitions; the two most popular ones were applied in this research. They are as follows:
The W is the width of the used target, and the A is the distance (amplitude) between the two targets.
Meyer et al. [
10] investigated the Fitts task by optimizing the sub-movement during the operation. They thought the phenomenon in Fitts’ law regulates people’s speed–accuracy tradeoff policy in a movement. People move as fast as possible across the amplitude and approach the target in the first stage. Then they slow down and adjust their movement based on visual feedback to hit the target accurately in the second stage. Due to the movement being along the target width, the size of that width dominates the magnitude of the tradeoff in time duration. With a larger width, less tradeoff time is needed to hit the target precisely. Another relationship between the movement time and the index of difficulty was proposed in the form of power law [
10].
Therefore, there is insufficient satisfaction in the statistical principles, like the lack-of-fit or residual normality test, in the information theory-based formula and the power law [
10]. The Canon model using Fitts’ data in 1-oz and 1-lb stylus tapping tasks [
2] fails in the lack-of-fit and the residual normality test [
11]. Fitts claimed that
MT is only related to
ID [
2]. Also, the Canon model implies that
MT is a constant for the same
ID value composed of varied
A and
W. However, our previous 1D1D study showed that Fitts’ law should be considered to underlie a constant amplitude. Different empirical formulations should be developed when a study demands more than one movement amplitude [
11]. In this manner, the not well-fitted cases in
IDFitts make the residual violate the normality assumption.
In our previous research, an SQRT_MT model, fusing findings from physics, psychology, and physiology, was proposed as an option to the Canon model [
11].
The SQRT_MT model has been shown to have a higher R-square, lower prediction sum of squares (PRESS), satisfactions in residual normality, and lack-of-fit test robustly in 1D1D movement compared to the Canon model and the power-law in the historical data in [
2] and experimental data in [
11].
With the ubiquity of computers and mobile devices in daily life, the icons of applications may have many reasons to be shaped as rectangles, squares, ellipses, or other shapes [
12]. Grossman’s work illustrated the non-rectangular interface in daily life [
13]. The state-of-the-art programming languages also have a drawing tool capable of producing various graphic-user-interface design shapes (
Figure 2). In the range of 2D targets, objects of equal width and height are in a particular class [
5]. Most studies in this class have employed square or circular targets [
14].
Figure 2 also presents the most frequent icon shapes in real life: equilateral triangle, round, and square. Consequently, this research chose the three shapes as the treatments. Furthermore, the diamond shape applied by Seikh and Hoffmann [
15] was taken for comparison.
However, the effect of a 2D target is seldom investigated in the literature. The area restrained by equal width and height is utilized unequally in different shapes. Therefore, the magnitude of width in the movement direction is no longer the primary factor that impacts people’s speed-accuracy tradeoff. Sheikh and Hoffmann’s work might be the first research on this issue [
15]. They posteriorly calculated the standard deviation of the hits on the target in the movement direction to define a new effective width for each shape. Grossman et al. [
13] applied a probabilistic Fitts’ law model to study 10 arbitrary shapes. These shapes included a complex mixture of a convex and concave graph, like a star. A traditional formula like Equations (2) or (3) was not recommended for such a complex shape. According to three varied target center definitions, Grossman’s model also needs posterior calculation for the probabilistic index of difficulty from a bivariate normal density function of endpoints’ coordination. Although the R-square is superior in the Canon model in both studies after using their specific
IDs, the posterior calculation of the endpoints is time-consuming and complicated. In advance, the researchers do not know the calculated difficulties of the shape targets in their studies. Therefore, the shapes in daily life applications are convex, like the illustration in
Figure 2; the current research is limited to the convex shape target applying Fitts’ law.
This study was inspired by Sheikh and Hoffmann’s work [
15], which was the first research about the geometric shape of the target on the effect of movement time. The information-theory-based formula for the geometric shape of targets did not perform excellently as in the traditional Fitts paradigm. Figures showed that the relationship between the movement time and the index of difficulty was not linear as expected. Moreover, it was evident that the different shapes with the same
IDFitts value significantly differed in the movement time (
Figure 1 and Figure 4 in [
15]). The reported R-square was only 0.919 and 0.918 in Experiments 1 and 2 of their research. Sheikh and Hoffmann concluded that the effective width of the target varied with the shape and used a bivariate-normal distribution to predict the shape’s effect accurately. The endpoint desperation related to the geometrical center was recorded to calculate the deviation in the direction of movement and perpendicular to the direction of movement. They proposed a modified
IDFitts applying the standard deviation in the direction of motion (called
IDSDH) and perpendicular to the direction of movement (called
IDSDV) instead of the target width (
W) in Equation (2). By applying
IDSDH, the R-square was increased to 0.962 and 0.963 for Experiments 1 and 2, respectively.
Despite the improved performance of the modified
ID in the endpoint distribution proposed in [
15], there is still space to improve the R-square. Furthermore, the inconvenience in the data collection of the endpoints and the post-calculation of the standard deviation might discourage the researchers from applying the proposal. Therefore, the first motivation of this study is to extend the success of the SQRT_MT model in the Fitts paradigm [
11]. The second motivation is to develop an easy-to-use
ID contrast to the post-calculation
ID for the 2D geometric shapes. Then, we expect the SQRT_MT model with the particular
ID could be applied in the geometric shape of the target to conquer the mentioned cons, like the model fitting adequacy insufficiently and the complicated post-calculation in
ID.
In our pilot study using the data in [
15], we found a strong relationship between the movement time and the area of the geometric shape. The first purpose of this research was to show that the area of a shape of equal width and height is a factor that impacts the prediction formula of movement time. The second purpose is that the area of a 2D shape affects the perceived difficulty in 1D2D movement. In this research, we hypothesized that the magnitude of the area might be a proper factor instead of the target width, as it impacts people’s speed-accuracy tradeoff policy when shapes are of equal width and height.
Consequently, researchers can use the target’s area instead of the width in the logarithmic term. Thus, we propose an optional
IDarea for Fitts’ law in this specific 1D2D class. The
ID could be in the Fitts or the Shannon form.
Here, area, in the logarithmic term, is the geometric area of the target shape. Finally, we extend the SQRT_MT model from 1D1D to 1D2D movement. Thus, the IDs in Equations (1) and (4) could be IDFitts, IDShannon, IDarea_Fitts, or IDarea_Shannon. The third purpose was to extend our work in a 1D1D task to validate that MT is related to IDFitts when the amplitude is constant but not mixing the varied amplitudes with the same IDFitts value in a 1D2D task.
Since the distance between targets (amplitude) was a constant of 320 mm for all conditions in Sheikh and Hoffmann’s work [
15], their data cannot be applied in the pilot study to testify to this research’s third purpose. There were two similar experiments in Sheikh and Hoffmann [
15]. The difference between the two experiments was the 2D target shapes they applied. Experiment 1 used rectangle, square, circle, and diamond shapes. One more equal width and height target, an equilateral triangle, was involved in Experiment 2. Nevertheless, the rectangular target’s data were excluded in the pilot study since it was a 1D target, not the 2D target that is the focus of this research. A modified
ID (
IDSDH) in Sheikh and Hoffmann [
15] was also applied in the pilot study to compare the performance of various
IDs in the Canon and the SQRT_MT models.
Although the width and the height of the 2D shapes might be the same, people could perceive the difficulty of a tapping task with different shapes varied intuitionally. Chan and Hoffmann applied the subjective assessment of difficulty for an ongoing visual movement [
16]. Participants’ self-reported evaluation is a popular method in psychophysical research. The Borg CR-10 scale [
17] is an easy-to-use and prevalent questionnaire in Psychologysics, which might be applied in a 1D2D Fitts task to assist the objective measurement of movement time.
2. Materials and Methods
Twelve students (six males and six females) served as the participants. They were 27.3 ± 2.4 (mean ± standard deviation) years old and 165.8 ± 6.9 cm in height. All of them are right-handed with no upper arm injury history.
A 24-inch full-HD-resolution projected capacitive touch monitor (model: Nextech NTSP240) and an active capacitive stylus (model: PenPower Pencil Pro) were applied in the experiment. The experimenter developed specific software to show the targets and record hit positions and duration between two hits. The hitting outliers were removed by a conservative and straightforward rule proposed by Zhai et al. [
18].
The environmental setting and the procedure were the same as in our previous study. The touch monitor showed the two shape targets horizontally after the experiment was initiated. The screen was located on a 66 cm high desk and angled backward; the top edge of the screen was 35 cm from the desktop in the vertical direction, and the bottom edge of the screen was 34 cm from the desk edge near the participant. Participants sat on a chair and adjusted the seat height until they felt comfortable operating the apparatus. Participants sat before the screen as regular daily activity and continually moved the stylus to hit the two targets reciprocally. Every participant had to hit the target 25 times in each treatment. The same four target shapes in [
15] were applied: equilateral triangle, diamond, circle, and square. To validate the third purpose of this research, three distance settings were used (
A) in [
11], 256, 512, and 1024 device-independent pixels between the two targets. One device-independent pixel is equal to a square with 0.265 mm in width. A wide range of the indexes of difficulty, 2, 3, 4, 5, and 6, defined by Equation (2) in [
11], was utilized in this study. The width of the target is determined by the movement amplitude and the index of difficulty. Each target’s height is the same as its width.
The target shape, the movement amplitude, and the index of difficulty are cross-design factors. Each participant conducted his/her treatments in different random order. After 25 treatments are finished, he/she could have a rest of 5 minutes or until he/she feels a recovery from the fatigue. When he/she completed each treatment, every participant evaluated the perceived difficulty by a subjective difficulty scale. This study applied the Borg CR-10 scale [
17] to assess the perceived difficulty. However, the original scale was enlarged from 0 to 100 (multiplied by 10) but kept the scale description to the responded point.
The dependent variables in this study were the mean movement time (unit: milliseconds) and the participant’s perceived difficulty. To achieve the first purpose of this study, we would like to show that tasks with the same
IDFitts values in varying amplitudes, called
ID(
A), are not identical in the mean movement time. Fifteen planned prior contrasts by Dunn-Šidák simultaneous test could validate the hypothesis [
19]. They are:
This hypothesis was extended to the effect of perceived difficulty. After we validated our first purpose, the IDFitts was broken down to the amplitude (A) and the target width (W) as factors of an exploring study. Another factor of the exploring analysis was the geometric shape of the applied target. If the factor was significant in the ANOVA, Tukey pairwise post-hoc test was applied in the exploring analysis. All the hypothesis and model adequacy tests were performed with a significant level set at 5%.
For the second purpose of this study, we would like to show that the area of a shape of equal width and height is a factor that impacts people’s speed–accuracy tradeoff policy. If the second hypothesis were valid, the proposed IDarea would perform better than other IDs in the fitted regression formulation for predicting the mean movement time. The proper formulation would be judged by the satisfaction of the essential assumptions in a simple linear regression and the performance of the quality indexes.
The lack-of-fit test is the diagnostic tool for adequately fitting a regression model. The Anderson–Darling test [
20] is used to test if a sample of data came from a population with a specific distribution. We applied this method to validate the essential assumption, the residual normality requirement, in a simple linear regression formula. An excellent linear regression formulation must pass both tests simultaneously.
Besides the R-square, this study applied the PRESS/SSE ratio as another quality index. Both PRESS and SSE could be used to evaluate the fitting quality of a regression model. However, the unit of the dependent variable in the Canon and the SQRT_MT model is different. The PRESS and SSE of the two models cannot be compared directly. Therefore, the PRESS/SSE ratio is a normalized metric with a unit free that could conquer the issue.