Twenty right-handed college students (10 men and 10 women) participated in this experiment. All of them had a normal or correct-to-normal vision and color vision. They were compensated after the experiment with a basic amount of money plus a bonus depending on their performance.

Stimuli were presented on a 19-inch CRT monitor with a refresh rate of 120 Hz and a resolution of 1,024 × 768 pixels via Visual Basic. The viewing distance was 60 cm. A chin and forehead rest was used to reduce head movements. Eye positions were recorded by SMI Hi-speed eye tracker (SensoMotoric Instruments GmbH, Teltow, Germany) at a sampling rate of 350 Hz. The spatial resolution of the eye tracker was 0.1° of visual angle. A standard nine-point calibration and validation were conducted at the beginning of each block to ensure the eye data quality. Memory arrays consisted of four items, three of which were gray disks, while the fourth, a reward cue, was either a 1-Fen or 1-Yuan or blurred Chinese coin (1 Yuan equals 100 Fen in Chinese currency). The locations of reward cues in memory arrays were randomized. All stimuli were in the size of 5.59° visual angle and had the same luminance (Figure 1B). The reward expectation level was assigned to none, low, and high corresponding to reward cues of blurred, 1-Fen, and 1-Yuan Chinese coins (reward cues), respectively.

The screen was divided into 16 grids and equally distributed in four quadrants (four squares in each quadrant). Items (disks or reward cues) were displayed in squares in accordance with the experimental design.

A sample test paradigm was used in the experiment. During the encoding phase, a “+” appeared at the center of the screen and disappeared until the subjects fixated it steadily for at least 800 ms. Four pictures, including three gray disks and a reward cue (blurred/1-Fen/1-Yuan Chinese coins), were presented in each quadrant at one of four grids for 1,000 ms, sequentially. The stimulus positions were described by the sequential order of quadrants (denoted by bigger numbers) and the number of grids (labeled by letters; as shown in Figure 1A) in which the stimuli were displayed. The participants were asked to gaze at the pictures and to remember their locations and sequential order.

There was a delay of 800 ms before the recall phase. During the recall phase, the participants were instructed to reconstruct the spatial location sequence as precisely as possible in 5,000 ms by gazing on a blank screen. The eye scan path was described by two series of positions of different processing levels. One was the quadrant sequence, indicating the sequential order of eye positions in different quadrants, such as “1234” (denoted by letters in Figure 1A). The other one was the sequential positions at grids in each of the quadrants, denoted by letters “ABCD.” The recall performance of spatial working memory was calculated by comparing the sequences of stimulus in the encoding phase and eye scan path in the retrieval phase. Similarities of each trial were calculated according to Brandt and Stark (1997) and Eddy (2004), including global similarity (GS), which was based on the quadrant, and local similarity (LS), which was based on the grid. Both GS and LS refer to degrees of similarity between sequences of stimuli positions in the study phase and eye fixation positions in the recall phase, ranging from 0 (totally different) to 1 (the same). GS and LS indicate the recall accuracy of the global and local positions of stimuli, respectively. LS has a finer spatial scale than GS and reflects a more rigid requirement of spatial resolution.

At the end of each trial, a feedback of reward amount (money in Chinese Yuan) was shown for 1,000 ms. Coin pictures were used to indicate locations to be remembered and reward cues (Hager et al., 2015; Di Rosa et al., 2019; Manga et al., 2020). The amount of reward depended on the performance (i.e., GS and LS) and reward conditions of participants and was calculated by the formula W^*(GS+LS)/2), where W = 0 for No-reward, W = 1 for Low-reward, and W = 10 for High-reward. The unit of it is the Chinese Yuan.

A total of 192 trials in the whole experiment were randomized and assigned into 12 blocks. There were 48 trials in each of the reward expectation conditions (No, Low, and High). The procedure of one trial is shown in Figure 2.

Dependent variables in the present research were derived from the fixation data. Fixations were defined using a temporal threshold of 100 ms and a spatial threshold of 2° visual angle, which were calculated offline. Then, the mean fixation number and the mean fixation duration were generated and grouped by the reward expectation conditions and location of fixation. The percentage of trials in which the cued grid was correctly re-fixed was used as the index of reward-cue memory performance. The spatial working memory task performance was evaluated by the sequence similarity proposed by Eddy (2004). Specifically, locations with the same temporal order of stimuli in the study phase and of eye fixation in the recall phase were compared, and each pair of overlapped locations scored 1 point. Sequence similarity was calculated by dividing the total score of four pairs of locations by 4, with a range from 0 (totally wrong) to 1 (totally correct).

During the study phase, three gray disks and a reward cue were presented sequentially one by one. The subjects were instructed to remember their spatial and temporal locations. We segregated the numbers of eye fixations and their durations by whether they were on the cue or non-cue (gray disk; as shown in Figure 3).

There was no significant effect of reward expectation or fixation position (on non-cue or cue) on the number of fixations. However, the fixation duration was significantly affected by reward expectation [M_No−reward = (362 ± 14) ms; M_Low−reward = (355 ± 13) ms; M_{High−reward} = (410 ± 17) ms; F_{(2, 38)} = 25.736, p < 0.001, ηp2 = 0.575], fixation position [M_cue = (421 ± 17) ms; M_disk = (331 ± 11) ms; F _{(1, 19)} = 91.351, p < 0.001, ηp2 = 0.828], and their interaction [F_{(2, 38)} = 15.087, p < 0.001, ηp2 = 0.443]. Simple effect analysis showed significant differences in fixation duration between the fixations on locations of reward cue and those on locations of non-cue disk under No-reward [t₍₁₉₎ = 8.714, p < 0.001, Cohen's d = 1.949], Low-reward [t₍₁₉₎ = 4.810, p < 0.001, Cohen's d = 1.075], and High-reward expectation conditions [t₍₁₉₎ = 7.005, p < 0.001, Cohen's d = 1.566]. Reward expectation effects were significant for fixation durations of both reward cue [F_{(2, 38)} = 6.020, p < 0.001, ηp2 = 0.241] and non-cue disks [F_{(2, 38)} = 23.120, p < 0.001, ηp2 = 0.549]. For the non-cue disks, fixation durations of High-reward cue were significantly longer than those of Low- [t₍₁₉₎ = 2.174, p = 0.043, Cohen's d = 0.486] and No-reward cues [t₍₁₉₎ = 4.077, p < 0.001, Cohen's d = 0.912]. The difference in fixation durations between No-reward and Low-reward cues was marginally significant [t₍₁₉₎ = 2.057, p = 0.054, Cohen's d = 0.460]. When reward cues were fixated, the same pattern appeared. Fixation durations of High-reward cues were significantly greater than those of No-reward [t₍₁₉₎ = 4.266, p < 0.001, Cohen's d = 0.954] and Low-reward cues [t₍₁₉₎ = 5.596, p < 0.001, Cohen's d = 1.251], and there was no significant difference between No- and Low-reward cues.

During the retrieval phase, the percentage of cued grids that were correctly fixated (as shown in Figure 4A) and the fixation durations of fixated or unfixated cued grids under different reward expectation conditions were calculated (as shown in Figure 4B).

Analysis of variance (ANOVA) of the percentage of trials in which cued grids were fixated (as shown in Figure 4A) revealed a significant effect of reward expectation [M_No−reward = 51.70% ± 1.60%; M_Low−reward = 46.40% ± 1.40%; M_{High−reward} = 29.20% ± 2.50%; F_{(2, 38)} = 44.800, p < 0.001, ηp2 = 0.702]. Multiple comparisons showed that the No-reward cued grid was fixated more frequently than the Low-reward [t₍₁₉₎ = 2.731, p < 0.05, Cohen's d = 0.886] and High-reward cued grids [t₍₁₉₎ = 9.461, p < 0.001, Cohen's d = 3.070], while the Low-reward cued grid received more fixations than the High-reward cued grid [t₍₁₉₎ = 5.702, p < 0.001, Cohen's d = 1.850].

A 2 (fixation on cued grid and non-cued grid) × 3 (No-, Low-, and High-reward cue) repeated-measures ANOVA of fixation duration revealed a significant effect of fixation location [M_cued = (441 ± 22) ms; M_uncued = (461 ± 19) ms; F_{(1, 19)} = 6.798, p = 0.017, ηp2 = 0.263], reward expectation [M_No−reward = (470 ± 19) ms; M_Low−reward = (442 ± 21) ms; M_{High−reward} = (441 ± 23) ms; F_{(2, 38)} = 5.808, p = 0.019, ηp2 = 0.234], and their interaction [F_{(2, 38)} = 22.793, p < 0.001, ηp2 = 0.545]. Simple effect analysis showed significant differences in fixation durations between cued and non-cued grids under No- [F_{(1, 19)} = 33.280, p < 0.001, ηp2 = 0.637], Low- [F_{(1, 19)} = 23.640, p < 0.001, ηp2 = 0.554], and High-reward expectation conditions [F _{(1, 19)} = 11.910, p < 0.001, ηp2 = 0.385]. Reward expectation effects were significant for mean fixation durations on both cued grid [F_{(2, 38)} = 11.710, p < 0.010, ηp2 = 0.381] and non-cued grid [F_{(2, 38)} = 15.11, p < 0.001, ηp2 = 0.443]. Duration of fixation on non-cued grid of No-reward expectation was significantly shorter than that of Low-reward [t₍₁₉₎ = −5.051, p < 0.001, Cohen's d = 1.639] or High-reward expectation conditions [t₍₁₉₎ = −3.717, p = 0.001, Cohen's d = 1.206]. There was no significant difference in fixation duration between High- and Low-reward expectations. For the cued grid, the fixation duration of No-reward expectation was significantly longer than that of High- [t₍₁₉₎ = 6.715, p < 0.001, Cohen's d = 2.179] and Low-reward expectation conditions [t₍₁₉₎ = 3.825, p = 0.001, Cohen's d = 1.241], and there was no significant difference between High- and Low- reward expectation conditions.

Spatial working memory performances (both GS and LS between the sequence of stimuli positions and eye fixation positions in the recall phase) under different conditions are shown in Figure 5.

For the GS, a 2 (cued grid fixated and unfixated) × 3 (No-, Low-, and High-reward cue) repeated-measures ANOVA revealed a significant effect of reward expectation [M_No−reward = 0.601 ± 0.012; M_Low−reward = 0.601 ± 0.010; M_{High−reward} = 0.625 ± 0.011; F_{(2, 38)} = 7.078, p = 0.002, ηp2 = 0.271]. The interaction between the cued grid and reward expectation is also significant [F_{(2, 38)} = 4.356, p = 0.020, ηp2 = 0.187]. Simple effect analysis revealed a significant reward expectation effect when the locations of reward were successfully re-fixated [F_{(2, 38)} = 12.095, p < 0.001, ηp2 = 0.389] but not when the locations of reward cue were not re-fixated. Further analysis showed that, when the locations of reward cue were successfully re-fixated, there were significant differences in GS between the Low-reward condition and the High-reward condition [t₍₁₉₎ = 3.788, p = 0.002, Cohen's d = 0.847] and between the No-reward condition and the High-reward condition [t₍₁₉₎ = 4.611, p < 0.001, Cohen's d = 1.031]. However, there was no significant difference between the No-reward condition and the Low-reward condition. The difference of GS between conditions when the cued grid was fixated and when the cued grid was unfixated was significant in High-reward expectation condition [t₍₁₉₎ = 2.813, p = 0.011, Cohen's d = 0.629], but not in the other two reward expectation conditions (p > 0.100).

Local similarity showed a different statistical pattern, that is, 2 (cued grid fixated and unfixated) × 3 (No-, Low-, and High-reward cue) repeated-measures ANOVA revealed a significant main effect of re-fixation [M_unfixated = 0.521 ± 0.012; M_fixated = 0.465 ± 0.010; F_{(2, 38)} = 26.413, p < 0.001, ηp2 = 0.582], showing that when the cued grids failed to be re-fixated, scores of local similarity went up. Other effects on LS were not significant.

The results of the study phase showed that stimuli, reward cues or not, in the High-reward expectation condition were fixated longer than those in No- and Low-reward expectation conditions (as shown in Figure 5). Considering the effect of reward expectation on recall performance in this study and the indecisive relationship between pure fixation duration and working memory performance in previous studies (Saint-Aubin et al., 2007; Oi et al., 2015), the changed fixation duration in the study phase of this study likely reflect differences in attention allocation under different reward expectation conditions. Specifically, the participants paid more attention to the task when reward expectation was high, compared with the No-reward and Low-reward conditions.

The result of fixation duration in the study phase is in line with previous studies that reported that reward expectation prompted encoding of working memory (Wallis et al., 2015; for a review, see Botvinick and Braver, 2015; Klink et al., 2017; Roberts et al., 2017). In their first experiment, Wallis et al. (2015) found that reward improved encoding of reward-associated items, and the reward effect was generalized to other items in the memory list. In this research, fixation durations on gray disks were prolonged in High-reward conditions as well. It seems that the mechanism underlying the reward effect in the encoding phase is more general than the specified encoding of the reward cue. A plausible explanation is that reward expectation enhances arousal, which in turn provides more cognitive resources for working memory encoding (Murray, 2007; Murayama and Kitagami, 2014; Unsworth and Robison, 2015), and participants are willing to make more efforts to obtain a higher reward. The null result of fixation number and the longer fixation duration of reward cue in the study phase are predictable. The design of the task in this research restricted the saccadic patterns of participants in the study phase and led to equivalent fixations under all reward expectation levels. These reward cues contained additional reward information compared with gray disks, requiring additional processes.

The main aim of this research was to explore the reward effect on spatial working memory at different processing levels. As predicted, we found a significant reward effect on GS, the indicator of global processing, but no reward effect on local similarity, the indicator of local processing. Furthermore, the percentage of trials in which the cued grid was correctly re-fixated during the retrieval phase, a relatively local indicator, decreased with the increment of reward. High reward improved spatial working memory performance at the global level but undermined the precision of spatial representations at the local level.

Mean fixation duration in the recall phase is in line with this conclusion. The mean fixation duration in memory-guided saccade tasks is affected by the strength of the memory trace, and it requires more time and effort to recall a weak memory trace (Burke et al., 2012; Haj et al., 2017; Dang et al., 2021). Therefore, when the memory of the next saccade target is weak, prolonged fixation duration is required to generate the following saccade (Meghanathan et al., 2020). Eye-tracking data of this research showed that fixation on uncued locations in the recall phase, sometimes followed by a saccade to the cued location, prolonged as the reward expectation was higher. A potential explanation is that, in this research, the memory for the cued location was degraded when the reward expectation was high, and more time was required before the saccade to the cued location could be generated.

The results of similarities at different processing levels and the negative reward effect on the representation of cued location are consistent with the hypothesis made by Ahmed and Fockert (2012), that is, the authors suggested that working memory load modulates selective attention to different levels of the same stimulus. When working memory load is high, information at a more global level is easily selected, while local-level information is ignored. It is difficult to constrain attention to the local level with a high working memory load. As a result, reward expectation prompted global similarity but did not affect local similarity in this study.

An interesting finding of this research is the trade-off between different indicators. As mentioned above, the mean fixation duration of the recall phase reflects degraded representations of cued locations and promoted representations of uncued locations with the increment of reward. Local similarities of trials in which the cued location was falsely re-fixated were higher than local similarities of trials in which the cued location was successfully re-fixated. It seems that the improvement of representations of uncued locations and local similarities comes at the cost of representation of cued locations.

These results of cued-location memory are consistent with the proposal of flexible attention theory (Sandry et al., 2014; Sandry and Ricker, 2020), which suggests that the cognitive system assigns attention resources flexibly among items in working memory, and the elevation of working memory performance of a certain item is at the cost of performance of other items (Sandry et al., 2014; Allen and Ueno, 2018; Sandry and Ricker, 2020). In this research, to obtain a higher reward, goal-directed cognitive control may have inhibited the maintenance of reward-cue location and allocated saved resources to the maintenance of other locations in the delay phase. Moreover, higher local similarities were observed in the trials in which the cued location was falsely re-fixated.

During the study phase of this study, all three conditions (No-, Low-, and High-) of the reward cue were randomly presented at different temporal positions. It may lead to a confounding effect. The different intervals of processing the reward cue mean that the levels of processing or the motivational states may vary accordingly. Specifically, motivational states might be identical in the three reward conditions until the presentation of the reward cue. Moreover, according to Sandry and Ricker (2020), the temporal position does affect performance in working memory tasks. Hence, the temporal position of reward cues should be considered in future studies.