This experiment received approval from the Ethics Committee of Hunan Normal University in China, and all participants provided written informed consent before taking part. The sample consisted of 52 college students in Changsha (Reani et al., 2017), including 20 males and 32 females, with average ages of 25 and 20 years, respectively. All participants were second-year students with no background in medicine and no prior knowledge of Bayes’ theorem, Chinese native speakers and normal vision or corrected to normal vision.

This experiment was a 3 (hit rate: high: 80% / middle: 50% / low: 10%) × 2 (cognitive style: field-dependent / field-independent) between-subjects design.

The dependent variables were a series of eye movement indexes, including total fixation time, number of fixations, average fixation time, pupil diameter changes and various eye movement indexes in the interest area. Among these variables, total fixation time refers to the sum of all the fixation times of a reader in a certain area, the number of fixation times refers to the number of fixations falling in a certain area, and the average fixation time refers to the average time spent by a reader each time he or she makes a fixation. In this study, the change in pupil diameter is the operational definition, and it is also the most commonly used index in the study of eye movement. Pupil diameter distinguishes the psychological load of the subjects, but the pupil diameter of an individual varies greatly. Only investigating the difference in pupil diameter may have an impact on the analysis due to noise related to individual differences. Moreover, this study aims to explore the psychological load change in the whole reasoning process; thus, the whole reasoning process is divided into 20 equal parts, on average, and the average pupil diameter in the first part is used as the baseline. The variation in pupil diameter in this study refers to the difference between the average pupil diameter and the baseline.

The EFT was administered to classify participants as field-dependent or field-independent. The test required participants to identify simple geometric shapes (e.g., triangles, rectangles) embedded within complex visual patterns within a 10-min time limit. Each correct identification earned 1 point, with a maximum score of 20. Participants were classified as field-independent if their scores fell above the sample mean (mean score = 12.3, SD = 3.1) and field-dependent if below. This approach aligns with established protocols (Salmani Nodoushan, 2007; Zhang, 2004), where higher scores indicate stronger ability to disembed details from contextual distractions.

The task utilized a modified breast cancer screening problem [adapted from Tang and Shi (2011)].

The context paradigm problem was employed to conduct Bayesian inference, with a base rate of 1% and a false alarm rate of 9.5%. Consistent with previous studies (Tang and Shi, 2011), the high hit rate was set at 80% while the low hit rate was set at 10%. The problem materials utilized identical wording and sentence structure.

The entire material was segmented into five distinct areas of interest (AOIs) (see Figure 1). The sentence pertaining to the base rate was designated as AOI1 (area: 61642), while the hit rate sentence was identified as AOI2 (area: 61642), and the false alarm rate sentence was labeled AOI3 (area: 68112). The sentence describing the problem itself was categorized as AOI4 (area: 124936), with the remaining regions falling under AOI5.

The eye movement instrument utilized in this study was the Eye-Link II model, manufactured by SR Research company. This instrument has a data acquisition rate of 250 Hz, ensuring high precision in tracking eye movements. The average fixation position error is less than 0.5°, indicating accurate measurement of eye fixation positions. The spatial resolution of the instrument is less than 0.005°, enabling detailed tracking of eye movements. The reasoning materials were converted into an 800 × 600 pixel image, with a black background and white text. The text was presented in 21-point SimSun font (1.5 line spacing) on a black background. The task was displayed as an 800 × 600 pixel image centered on a 17-inch screen.

The subjects sat in a chair 70 cm away from the monitor, the instructions for the test were explained, and subjects were briefly introduced the experimental process. Participants were given the response apparatus and were told how to operate it. Then, they put on the eye cap, and their head was fixed in a relatively stable position; and we conducted calibration, validation calibration (nine o’clock) and drift calibration.

After drift calibration, the subjects were presented with instructions (all the instructions and problem materials were converted into pictures for presentation, and the text was presented in the center of the picture), practice exercises, confirmation instructions, Bayesian questions, and concluding remarks; the entire process is shown in Figure 2. The Bayesian reasoning task was presented in a randomized across. The duration of the eye movement task was approximately 10 min, and after the task, the subjects needed to complete the Embedded Figure Test (EFT) and fill in the basic data; the duration of the task was approximately 20 min. At the end of the experiment, the subjects were given a small gift.

Eye-movement allow for real-time recording of a viewer’s visual fixations, enabling exploration of the various processes involved by analyzing the overall duration of fixations during the inference process. In the problem-solving process, inference unfolds across three distinct stages (Shi et al., 2010).

The first stage entails the representation of the Bayesian reasoning problem. Reasoners’ attention toward each area of interest undergoes a turning point across the five designated areas. During the initial 25% of the inference process, problem information is preliminarily characterized. As the reasoning process unfolds, attention toward the base rate diminishes while greater focus is placed on the hit rate. Participants allocate comparable attention to the false alarm rate and the problem description information (Shi et al., 2010).

The second stage involves the integration of probability information and the selection of strategies. The middle 65% of the time is dedicated to establishing the relationship between probability information and selecting a probability estimation strategy. In the early part of this stage, emphasis on the problem description information and hit rate may lead to a deeper problem analysis, albeit at the expense of neglecting the base rate. In the middle of this stage, attention is relatively evenly distributed across different types of information, suggesting the subject’s intent to plan and integrate the information effectively. Toward the later part of this stage, the base rate tends to be relatively overlooked, with increased value placed on the hit rate (Shi et al., 2019).

The third and final stage, comprising the last 10% of the reasoning process, corresponds to the probability judgment stage. Overall analysis reveals that reasoners allocate equal attention to probability information during the decision-making stage and do not disregard the base rate (Shi et al., 2010; Shi et al., 2019).

The utilization of eye movement technology presents a significant advantage in that it allows for the recording of immediate information collection during inference, thereby enabling the reconstruction of the entire reasoning process based on real-time data acquired by the eye movement instrument.

Through the analysis of various eye movement indexes, it was found that the subject pays the most attention to the hit rate, followed by the false alarm rate and the problem description information, and it pays the least amount of attention to the base rate. The reasoner’s fixation time on the hit rate exceeds that on the base rate by approximately 50%, and the hit rate is indeed the most important information in the process of solving the problem. The subject also pays more attention to the false alarm rate, only less than the hit rate, perhaps because the problem description information contains two characteristics: illness|positive (p|h). These two characteristics are specifically included in the hit rate (p|h), and the false alarm rate is closely connected to these two characteristics of the system (−p|h). It is possible that the degree of attention paid to the information is affected by the degree of fit between them and the task information, and the higher the degree of fit is between them, the higher the level of attention the reasoner will allocate to them (Zhang et al., 2003).

Although the base rate exhibits the shortest fixation duration and frequency, it is noteworthy that participants do not entirely disregard it. However, the analysis of fixation duration and frequency merely indicates the extent to which the reasoner attends to each type of information, visually processes the information, and engages with it. This analysis does not address the extent to which the subject utilizes this information in making inference decisions or calculating the results. The relationship between the utilization of information and computation in decision-making is not directly examined. Therefore, further research utilizing Event-Related Potentials (ERPs) should be conducted to investigate the specific information processing mechanisms in the future (Shi et al., 2019).

Complex Bayesian reasoning and the hit rate play crucial roles in problem-solving, and variations in the hit rate level have a certain impact on the process.

Regarding the indices of total fixation duration and fixation count, there is a significant main effect of the hit rate. When solving Bayesian reasoning problems, the reasoning process exhibits the longest fixation duration when the hit rate is 80%, and the fixation duration is longer for reasoning problems with a hit rate of 1%. Conversely, when the hit rate is 50%, the fixation duration is the shortest. The fixation count follows a similar pattern, with high, low, and medium orders. However, there is no difference in the average fixation duration. Inference demonstrates the fewest fixations and the lowest number of guesses when the hit rate is 50%. This might be attributed to the ease of understanding inference at the 50% hit rate, as it aligns with common everyday experiences. Consequently, reasoners can express it quickly, recognizing that the hit rate itself constitutes the most critical information required for Bayesian reasoning. By simplifying the representation of the hit rate, the overall reasoning time is reduced (Tang and Shi, 2011).

The variation in hit rate levels influences the attention allocated to the base rate, primarily during the second and third stages, while no difference is observed in the first stage. When tackling problems with a high hit rate, there is an early-stage peak of attention directed toward the base rate. In contrast, when individuals solve Bayesian problems with a high hit rate, the peak attention toward the base rate emerges in the middle of the second stage. However, in scenarios involving low hit rates, the attention peak toward the base rate occurs in the third stage (Tang and Shi, 2011).

Cognitive style refers to individuals’ preferred and habitual ways of organizing and interpreting information, encompassing their perceptual, memory, thinking, and problem-solving preferences. Field-dependent individuals rely on external environmental cues during information processing, whereas field-independent individuals rely on internal perceptual cues for information processing.

Analyzing the overall fixation duration during inference revealed that field-dependent subjects required more time to solve the problem compared to field-independent subjects. The index of the number of fixations also indicated a higher number of fixations among field-independent individuals compared to field-dependent individuals. Further analysis showed that the fixation frequency of field-dependent subjects was influenced by the hit rate, while field-independent subjects were not affected by it (Wang and Ouyang, 2004).

When examining eye movement indexes within each area of interest, it was found that fixation duration in each area of interest varied for field-independent individuals and was unaffected by the hit rate. However, for field-dependent individuals, there was no difference in fixation duration across each area of interest, but it was influenced by the hit rate. Overall, field-dependent participants exhibited shorter fixation times, and their fixation duration varied across different areas of interest, indicating unequal attention allocation during problem-solving and a strategy distinct from that of field-independent individuals. Field-independent individuals effectively allocate their resources and are not easily influenced by external cues such as changes in the hit rate.

In contrast, field-independent participants demonstrated longer overall fixation times with no difference in the distribution of time across each area of interest. This suggests that field-independent individuals take more time to solve the problem than field-dependent individuals, and they allocate equal attention to information within each area of interest without a distinction of priority. Additionally, the performance of field-dependent individuals is also influenced by changes in the hit rate, indicating that external environmental changes truly affect the reasoning time allocated by field-dependent individuals (Shi et al., 2010; Wang and Ouyang, 2004).