Keywords

1 Introduction

Information processing is a key component of human performance. As users are faced with the need for more information through the use of complex visual displays, there are increasing demands in developing design strategies to facilitate effective information seeking behavior. Moreover, as users require more and more information to do their jobs, display complexity has greatly increased. This complexity makes it difficult for people to correctly interpret the information displayed and project what is likely going to happen (Salvendy 2012). There are two major ways to represent information, textually and graphically. As computer interfaces have become more complex, the questions of using either graphics or alphanumeric texts to display information (Tullis 1997) has become moot. Instead, combining graphics and alphanumeric texts in an interface to facilitate information seeking behavior becomes an important issue, (Aigner et al. 2011; North 2006; Spence 2007 and Steele and Iliinsky 2010).

As the need for combining graphical and textual elements grows, some researchers compared graphical displays with textual displays to find which display is easier and quicker for information notification. For example, (Ottensooser et al. 2012) reported that subjects with textual representations showed significant improvement in understanding business processes without training beforehand, while training is still required for graphical representations to show significant improvement. Some researchers have found that subjects spent more time and effort in graphical representations, which challenges the general assumption that graphical representations are easier than textual representations, (Cepeda and Guéhéneuc 2010). Moreover, some research emphasizes the importance of combining graphical with textual tasks together. For example, Bederson et al. (2004) used focus and context mechanisms to develop the temporal information presentation of calendars on small handheld devices. If the display space is large enough, calendar entries are shown in textual form. Otherwise, calendar entries are indicated by bars. The present study will explore how to combine graphical and textual information in different display configurations.

Eye tracking is a technique that measures individual’s eye movements. It can help to understand visual and display-based information processing and the factors that may influence the usability of system interfaces, (Poole and Ball 2006). An eye tracker requires calibration before a session is started. Calibration is a procedure in which data are gathered from the coordinates of the subject’s pupils and corneal reflections in the eye video coordinate system, to represent his or her eye fixations in the stimulus space. A common procedure is to have the subject fixate nine points in a rectangular calibration grid, (Kliegl and Olson 1981 and Blignaut et al. 2014) although 16 and 25-point systems are also used. The fact that calibration points consists of equally spaced points extending to the boundaries of the stimulus field provides a structure for an experiment to examine the placement of textual and graphical information.

The configurations of the 9, 16, and 25-point calibration systems occupies 37 different locations on the screen. In the present study, we used an arithmetic task and a gauge task to investigate the effect of locations. The arithmetic task represented the textual task and the gauge task represented the graphical task. In this study, response time, response accuracy and root mean square (RMS) of fixations, representing the magnitude of the distance from the center of the target to the fixation point, were analyzed. The results from this study can be used to help interface designers to locate the different types of information more effectively and also to help eye tracker experimenters to choose the most appropriate calibration points.

2 Methods

2.1 Participants

Thirty participants (Mean age = 22.5, 13 male and 17 female) were recruited from the Pennsylvania State University and were paid $10 each. Because prescription eye glasses affect the accuracy of the eye tracker, recruits were screened through reading 11.25 pt Arial regular font text at a distance of 25 in. without wearing eye glasses (contacts were allowed).

2.2 Apparatus

The study was conducted on a standard PC (Windows 7 installed) and a 20″ monitor (ST2010, Dell Inc., LOCATION), which was set to a resolution of 1600 × 900 pixels and placed 24.5 in away from the participant. An eye-tracker (Mirametrix S2 Eye Tracker, Montreal, QC, Canada) was used for recording the participants’ gaze data at a sampling rate of 60 Hz within 1° of accuracy. A chin rest was used to minimize head movement and the vertical viewing angle and the horizontal viewing angle was controlled to 0° to −24.5° (below eye level) and −21.53° to 21.53°, respectively, to ensure that participants could see the entire screen (Van et al. 1972; Fig. 1).

Fig. 1.
figure 1

Participant position.

Full size image

2.3 Stimuli and Task

To evaluate the performance of textual and graphical tasks and to give advice on how to display two types of information on an interface, two types of tasks were prepared, the arithmetic and gauge task. The arithmetic task consisted of presenting participants with textual stimuli, which were arithmetic equations (Arial font, regular, 11.25 point, white) placed inside black circles (80 × 80 pixels; Fig. 2). Participants were required to determine whether the equations were incorrect or correct by pressing ‘F’ or ‘J’, respectively.

Fig. 2.
figure 2

Examples of the equations presented in the arithmetic task.

Full size image

The gauge task consisted of presenting participants with graphical stimuli, which were semicircle gauges (80 × 40 pixels; Fig. 3). Participants were required to determine whether the indicator arrow was on the left or right by pressing ‘F’ or ‘J’, respectively.

Fig. 3.
figure 3

Examples of the semicircle gauges presented in the gauge task.

Full size image

Both tasks had three conditions, which were configuration points of 9, 16, or 25, and all tasks were presented on a white background (Fig. 4).

Fig. 4.
figure 4

The arithmetic and gauge task with 9, 16, or 25 configuration points.

Full size image

For both tasks, stimuli were presented one-by-one on the screen (according to the task and configuration point setting) and participants were required to press ‘F’ or ‘J’ based on the displayed stimulus. Response to the current stimulus triggered the next stimulus to appear.

A combination of task and configuration point were defined as trials. Each trial required participants to respond to either 9, 16, or 25 stimuli. Stimuli did not appear more than once on the same location per trial.

Combining all three configurations, there were 37 possible locations for each task. The possible locations for 9 points configurations were 1, 4, 7, 17, 19, 21, 31, 41, 47. The possible locations for 16 points configurations were 1, 3, 5, 7, 13–16, 22–25, 31, 33, 35, 37. The possible locations for 25 points configurations were: 1, 2, 4, 6, 7, 8–12, 17–21, 26–30, 31, 32, 34, 36, 37 (Fig. 5).

Fig. 5.
figure 5

The locations of points.

Full size image

2.4 Experimental Design

The experiment was a 2 (arithmetic and gauge task) × 3 (9, 16, or 25 configuration points) repeated measures factorial design measuring accuracy, response time, and RMS for each configuration. Root Mean Square (RMS) is the magnitude of deviation from the target, which reflects the vision requirement to process the information in the current experiment. A higher RMS reflects a lower requirement of the eye to process the information, while a lower RMS reflects a higher requirement of the eye to process the data and more focus on the target.

Half of the participants completed the arithmetic task first, while the other half completed the gauge task first. For each task, participants performed 24 trials (8 trials per configuration point setting) in a fully randomized sequence. For each configuration, the number of correct/incorrect or left/right stimuli were equally divided and the sequence was also fully randomized. Combining all three configurations, there were 37 possible locations for each task.

2.5 Procedure

Prior to arrival, the participants completed an online demographic questionnaire to provide background information to make sure that they could read 11.25 pt Arial regular font text at a distance of 25 in without wearing eye glasses (contacts were allowed). After an introduction, participants provided informed consent. Each experiment started with instruction of each task. Next, they were instructed to put their chin on the chin rest with the thighs roughly parallel to the floor, and were asked to maintain a roughly upright sitting posture (Woo et al. 2015). Then, participants started the experiment by pressing the SPACE button, and the white board showed up. When they pressed the SPACE again, the first target stimulus appeared and the participants were asked to respond as quickly and accurately as possible. After they finished the first part (either arithmetic and gauge) of the experiment, each participant was required to take a five-minute break, and then continue to the second part with the same procedure as the first part. After they finished all parts, they were given a post-experiment questionnaire asking them for feedback about their experience and additional suggestions. Finally, they were compensated.

2.6 Statistical Analysis

Regression models were fitted for each of the response variables: Correctness (proportion of correct answers), Response time, and RMS. The Mixed model was used for RMS and also for the log of the response time, while GLIMMIX (Generalized Mixed Model) was used for correctness. Pvalues were adjusted to allow for the multiple comparisons, using the Sidak correction.

In all analyses, p-values less than 0.05 were considered statistically significant.

3 Results

For each dependent variable, the model was fitted first to find whether configuration points was accompanied by elevated calibration accuracy, shorter response time and higher response accuracy. Then, pairwise comparisons among different locations were conducted to test which locations caused the significant results. The points that located at the four corners were examined first, followed by the examination of the points located at the marginal places. Finally, the points located at the center place were examined. Points that differed from only two or fewer other points were excluded, because it did not form a pattern. These statistically significant differences were interpreted as random effects.

3.1 Accuracy

The generalized mixed model showed that the tasks and configuration points had a significant effect on accuracy (F(1,121.5) = 57.26, p < .0001). Post hoc analysis with Sidak correction showed that the accuracy in the gauge task was significantly higher than in the arithmetic task for all three configuration points (all p < .05). The accuracy in the 25 configuration point of the arithmetic task was significantly lower than in both the 9 configuration point (p < .001) and the 16 configuration point (p < .001; Table 1). Two trials were discarded due to high error rate (> 3 standard deviations from the mean).

Table 1. Mean (SD) accuracy for each task and each resolution.
Full size table

The fitted model showed that the interaction between task and location was significant (F(36,1044) = 8.608, p < .001). Thus, pairwise comparisons of locations were conducted for each task separately. The significant results were summarized below (Tables 2 and 3).

Table 2. Significant location results for the arithmetic task.
Full size table
Table 3. Significant location results for the gauge task.
Full size table

3.2 Response Time

According to the fitted mixed model, the tasks and configuration points had a significant effect on response time (F(1,38.8) = 681.3, p < 0.0001). Post hoc showed that the mean response time for the gauge task was significantly lower than the arithmetic task for all three configuration points (all p < .05). For the gauge task, a significantly small mean was found for the 25-point configuration compared with the other 2 configurations (Table 4). One trial was discarded due to high error rate (>3 standard deviations from the mean).

Table 4. Mean (SD) response time (msec) for each task and each resolution.
Full size table

It was also found that the interaction between task and location was significant (F(36,1044) = 3.205, p < .001). Pairwise comparisons of locations were conducted only for the gauge task (Table 5). The location effect was not significant for the arithmetic task.

Table 5. Significant location results for the gauge task.
Full size table

3.3 RMS (Root Mean Square)

The fitted mixed model showed that the tasks and configuration points had a significant effect on RMS (F(1,155.7) = 116.8, p < 0.0001). Post hoc analysis showed that the gauge values were significantly larger than the arithmetic values for all resolutions. For the gauge task, a significant large mean was obtained for the 9 points (p adjusted < 0.001; Table 6).

Table 6. Mean (SD) RMS (pixel) for each task and each resolution.
Full size table

It was also found that the interaction between task and location was significant (F(36,684) = 2.890, p < .001). Thus, the pairwise comparisons of locations were conducted for each task separately (Tables 7 and 8). Because the point 20 and point 35 are too close, they were considered as one location point for the arithmetic task.

Table 7. Significant location results for the arithmetic task.
Full size table
Table 8. Significant location results for the gauge task.
Full size table

4 Discussion

In this experiment, task type × configurations analysis was first conducted to find whether more configuration points are accompanied by elevated calibration accuracy, shorter response time and higher response accuracy. Next, task type × locations analysis was conducted to find the influence of locations on task performance.

When looking at the results of accuracy of different configurations (Table 1), the accuracy in the 25-point configuration of the arithmetic task was significantly lower than in both the 9-point and the 16-point configurations with 9-point the highest. Moreover, although the configuration results for the gauge task was not significant, the accuracy in the gauge task was the highest in the 9-point configuration, followed by 16-point and 25-point. This indicated that more configuration points did not facilitate the information processing accuracy, which may lead to the suggestion that fewer configuration points would be more accurate. Finally, A more detailed analysis on locations of the arithmetic task was conducted and summarized in Fig. 6 (accuracy of red points is significantly smaller than many of the other points). Results of the arithmetic task showed that the accuracy was more influenced by the points that were located at the right side of the screen, indicating that in the interface design, textual information should be placed at the left side of the screen. Results of the gauge task (Fig. 7) showed that the accuracy of the 19-point was significantly smaller than many of the other points. This result was counterintuitive because the participants seated in front of the center of the screen, and they were supposed to see the stimuli in the center much quicker than the stimuli in the corners. A further analysis of the eye search pattern of the participants is needed to further explore the eye behaviour that affected the accuracy.

Fig. 6.
figure 6

The accuracy result for the arithmetic task. (Color figure online)

Full size image
Fig. 7.
figure 7

The accuracy result for the gauge task.

Full size image

As for the mean response time, it was found that for the 25-point configuration it was significantly smaller than for the other two configurations in the gauge task, while there were no significant findings in the arithmetic task (Table 4). Although the mean response time for the 25-point configuration was the smallest in the gauge task, the overall time was the largest. Moreover, this result was only for the gauge task. Thus, it seems reasonable to suggest that the 9-point configuration was the most efficient choice since it requires less time to calibrate.

The main result about the effect of locations on response time is displayed in Fig. 8. The green points denote the locations where the mean response time was significantly larger than in the other locations. As we see, these points are located at the corners and the bottom right margins. Thus, in an interface design, the graphical information should not be placed at the corners and bottom right margins to avoid longer response time.

Fig. 8.
figure 8

The response time result for the gauge task.

Full size image

As for the RMS, the gauge values were significantly larger than the arithmetic values for all resolutions (Table 6). In the gauge task, the RMS for the 9-point configuration was significantly larger than the other two configurations. The purpose of calculating RMS was to find whether more calibration points would lead to elevated calibration accuracy. However, it seemed that the RMS was significantly influenced by the task type. In the real calibration process, the subjects only needed to look at specific points without any textual or graphical information. With such huge differences between the two types of tasks, it was very difficult to conclude which configuration would facilitate the calibration accuracy. The results on locations are displayed in Figs. 9 and 10. The accuracy which is denoted by red points is significantly smaller than many of the other points and the accuracy marked by green points is significantly larger than many of the other points. An analysis of the locations of the arithmetic task (Fig. 9) found that the corner points have larger RMS. Larger RMS was also significant for the corner points in the gauge task, (Fig. 10). These results indicate that the participants could process the corner textual and graphical information correctly without a high vision acuity. Also, the participant could process the margin graphical information correctly without a high vision acuity. Thus, when an interface includes both textual and graphical elements, graphical information could be placed at margin and corner areas and textual information could be placed at corner areas to facilitate the efficiency of information processing.

Fig. 9.
figure 9

The RMS result for the arithmetic task.

Full size image
Fig. 10.
figure 10

The RMS result for the gauge task.

Full size image

5 Conclusions

The present research found that from the perspective of response accuracy and response time, the 9-point configuration was most appropriate for the calibration process. The RMS values failed to offer valid advice on configuration point choice because of the significant differences between the textual and graphical tasks.

Moreover, the present research found that the locations of targets have impacts on task accuracy, response time and the RMS of participant fixations. Specifically, if the accuracy was the highest priority, the textual information should be placed at the left side of the screen and the graphical information should not be placed at the center of the screen. If a quicker response time was the highest priority, the graphical information should not be placed at the corners and bottom right margins. Finally, since the RMS at the marginal and corner areas of the gauge task was higher and the accuracy was not lower, the RMS at the corner areas of the arithmetic was higher and the accuracy was not lower, it could be concluded that the graphical and textual information could be processed without a high visual acuity. Thus, if an interface includes both textual and graphical elements, graphical information could be placed at corner and margin areas to facilitate the efficiency of information processing and textual information could be placed at corner areas to facilitate the efficiency of information processing.

One limitation of the current experiment is that the RMS values failed to offer valid advice on configuration point choice because of the significant differences between the textual and graphical tasks. Future research could use specific points which do not include textual or graphical information to calculate the RMS values.