Abstract:

Attentional refreshing is the process of promoting and prolonging the activation of information in working memory (WM) by returning it to the focus of attention. This process can prevent the information in WM from fading over time or being disrupted by distractors. Previous studies have demonstrated that attentional refreshing can be guided by retro-cues or influenced by various experiences, such as reward-related or self-related stimuli. Recent studies have also explored the value effect in WM and found that people tend to prioritize more valuable information in WM, indicating that value may play a role in guiding attentional refreshing during retention. In a groundbreaking study by Atkinson et al. (2022), attentional refreshing was shown to partially explain the value effect in WM. However, the study was unable to determine why high-value information was prioritized for refreshing. It has been suggested that the value effect in WM may be due to a biased attentional refreshing procedure where individuals tend to focus more frequently or for longer periods on the more valuable item during retention, as compared to the other items.

To investigate the value-directed attentional refreshing and its underlying mechanism, this study conducted three experiments. All experiments were designed with a within-subject design, with the independent variable being the value of the item (high or low). The sample size for each experiment was determined using G*power based on prior research. Experiment 1 examined the value-directed attentional refreshing under the condition of simultaneous presentation of information. Twenty-four valid participants took part in the experiment, 7 of them were male, with an average age of 18.92 ± 1.14 years.

The procedure for Experiment 1 was shown in Figure 1. Participants were reminded before the start of the experiment to try their best to obtain the highest score during the recall phase. At the beginning of the experiment, a 500 ms fixation point was presented on the screen, followed by an articulatory suppression task, which required participants to continuously repeat “Fujian Normal University” until the dot detection task was completed. After a 500 ms interval, the memory material was presented, and participants were asked to remember all the letters within 4000 ms. Each letter was associated with a number representing its value. Then, a fixation point lasting 3000 to 3700 ms was presented on the screen, followed by a 300 ms box around the fixation point, reminding participants that a dot detection task was about to be performed. In the dot detection task, participants had 1500 ms to judge whether the two small dots displayed on the screen were vertical or horizontal. If they were vertical, they pressed the “9” key; if they were horizontal, they pressed the “0” key. The keys were balanced among participants, and the probability of the two small dots being horizontal or vertical was 50%. The small dots were randomly presented in one of the positions where the letters were previously presented. The dot detection position was balanced among the values. After the dot detection task, participants had 12 s to input the letters into six boxes displayed on the screen, without inputting the numbers. After completing the input, participants received feedback on the accuracy of the dot detection task and the recall score for the current trial. Each participant completed three blocks of trials, with 48 trials in each block and a total of 144 trials. Before the formal experiment, participants were required to complete five practice trials of the same program to ensure that they were familiar with the experimental procedure.

A paired-sample t-test was conducted on the data from Experiment 1 and revealed that: (1) Participants' memory performance on high-value items (83.50% ± 7.47%) was significantly better than that on low-value items (54.35% ± 15.07%), t(23) = 8.55, p < 0.001, 95% CI = [0.22, 0.36], Cohen’s d = 1.75, indicating that value-guided memory exists in WM processes. (2) Participants' reaction time on the dot probe task at the location corresponding to high-value items (770 ms ± 102 ms) was significantly faster than that at the location corresponding to low-value items (789 ms ± 117 ms), t(23) = −2.41, p = 0.012, 95% CI = [−34.50, −2.60], Cohen’s d = −0.49, indicating that attentional refreshing during the WM maintenance is guided by value when items were presented simultaneously.

Under the condition that items were presented simultaneously, Experiment 1 confirmed the existence of a value-directed attentional refreshing process. However, because the memory items in Experiment 1 were presented simultaneously, there are alternative explanations for this result. Firstly, during the encoding phase, participants may have paid more attention to the location of high-value items, which could lead to a reflective attentional bias. Therefore, the attentional advantage during the maintenance phase may have been caused by location, rather than value. Secondly, since memory items were presented simultaneously, participants may have adopted a strategy of encoding memory items in sequence, starting from high-value items in order to achieve the highest score during encoding. Similarly, during the free recall phase, participants may have prioritized recalling high-value items, leading to higher scores for those items than for low-value items. In order to eliminate these two possibilities, Experiment 2 presented the items one by one, with each item presented for the same amount of time (1000 ms) and randomly appearing in one of six positions. In addition, in order to test whether smaller values can still generate value-directed attentional refreshing, the corresponding pair of numbers for high and low values in Experiment 2 were set to “5” and “1”, respectively.

Experiment 2 recruited 23 participants, 7 of whom were male, with an average age of 19.83 ± 1.80 years. The results showed that: (1) participants' memory performance for high-value items (79.41% ± 13.91%) was significantly better than that for low-value items (48.71% ± 11.59%), t(22) = 8.07, p < 0.001, 95% CI = [0.23, 0.39], Cohen's d = 1.68, indicating value could guide WM; (2) participants' reaction time at the location of high-value items (735 ms ± 110 ms) was significantly faster than that at the location of low-value items (751 ms ± 130 ms), t(22) = −1.79, p = 0.044, 95% CI = [−34.96, −2.58], Cohen's d = −0.37, indicating value could guide attentional refreshing during the WM maintenance when items were presented in succession.

Experiment 3 employed the blank screen paradigm in combination with eye-tracking technology to investigate the attentional refreshing advantage of high-value information. Based on the biased attentional refreshing procedures (Atkinson et al., 2018, 2021; Hitch et al., 2018; Hu et al., 2016; Sandry et al., 2014), value-directed attentional refreshing can be achieved in two ways: by frequently refreshing the position of high-value information or by allocating more time to refreshing the position of high-value information. During the WM maintenance, attentional refreshing reactivates the location marker corresponding to the item, thereby causing eye movements to shift towards the corresponding spatial location on the blank screen. Therefore, the number of fixations on the regions of interest corresponding to different value information during the maintenance can measure the frequency of attentional refreshing, and the fixation duration at each fixation point can measure the duration of attentional refreshing.

The procedure for Experiment 3 was shown in Figure 2. Participants were instructed to try their best to obtain the highest score during the recall phase. At the beginning of the experiment, a fixation point was presented on the screen for 500 ms, followed by an articulatory suppression task that required participants to repeat “Fujian Normal University” until the recall test. Then, four numbers corresponding to the value scores of the graphs at each position were displayed on the screen for 1000 ms. After the numbers disappeared for 500 ms, a screen containing 4 grey graphs was displayed for 1500 ms, during which participants were required to remember these graphs. Then, a blank screen lasting 3000 ms was presented, followed by a prompt displayed on the screen for 1000 ms. The prompt was used to instruct participants to recall the graph at that position. Finally, five graphs were presented on the screen. Participants should choose a graph from the 5 graphs that matched the graph corresponding to the cued position in 4000 ms. Each participant completed two blocks of trials, with 30 trials in each block and a total of 60 trials. Before the formal experiment, participants completed 5 practice trials to ensure they were familiar with the experimental procedure. The corresponding pair of numbers for high and low values in Experiment 3 were set to “6” and “1”, respectively.

Experiment 3 recruited 24 participants, 10 of whom were male, with an average age of 21.00 ± 2.38 years. The results showed that: (1) participants’ memory performance for high-value (85.34% ± 12.52%) items was significantly higher than that for low-value (73.33% ± 17.23%) items, t(23) = 3.82, p < 0.001, 95% CI = [0.06, 0.19], Cohen’s d = 0.78, indicating that value could guide WM; (2) the number of fixations on the location corresponding to the high-value (0.32 ± 0.10 times/trial) items during the blank screen period was significantly higher than that on the location corresponding to the low-value (0.27 ± 0.08 times/trial) items, t(23) = 2.63, p = 0.007, 95% CI = [0.01, 0.08], Cohen’s d = 0.54; (3) there is no significant difference in the duration of single fixations on the location corresponding to high-value (313 ms ± 102 ms) and low-value (332 ms ± 125 ms) items during the blank screen period, t(23) = −1.55, p = 0.135, 95% CI = [−44.04, 6.33], Cohen’s d = −0.32. Since the null hypothesis statistical test p-value was greater than 0.05, it failed to support the null hypothesis. Therefore, a Bayesian test was conducted on the duration of single fixations to examine whether it supports the null hypothesis. The result showed that BF₁₀ = 0.61, which did not support the null hypothesis. The above results indicated that value-directed attentional refreshing might be achieved by increasing the refresh rate of high-value information.

The above experiments directly confirmed the value-directed attentional refreshing that high-value information received priority for attentional refreshing in WM retention when compared to low-value information. More importantly, the results indicated that value-directed attentional refreshing might be achieved by increasing the refresh rate of high-value information rather than deploying more time on it. This study contributes to the research on attentional refreshing and provides new insights into how people prioritize information in their daily lives. Moreover, it sheds light on the mechanism of value-directed attentional refreshing and helps develop the time-based resource-sharing model to a certain extent. These findings can aid researchers in developing computational models that simulate people’s attentional refreshing process.