Abstract:

Visual working memory (VWM) and selective attention are two essential topics of investigation in the field of cognitive psychology. Previous studies have suggested that object-based attention selection modes may be present during the VWM encoding stage, and feature-based attention selection modes may be present during the maintenance stage. Nonetheless, these conclusions are based on different research paradigms, object feature dimensions, and response indicators, so it is prudent to exercise caution when inferring the existence of distinct attention selection modes during different stages of VWM processing. The aim of the present study is to evaluate this hypothesis and provide empirical support.

In Experiment 1a, thirty college students were recruited to complete a change-detection task. Participants were instructed to memorize the features of the objects presented in the memory display by means of a pre-cue or retro-cue presented prior to or following the memory display. Specifically, in pre-cue trials, participants were asked to memorize only the cueing task-relevant feature while ignoring the task-irrelevant feature. In retro-cue trials, participants needed to memorize the entire object so that they could select the task-relevant feature according to the retro-cue. The present study examined the “irrelevant-change distracting effect” by comparing memory performance between the condition of task-irrelevant feature changes and no-changes on the memory probe test display. Experiment 1b had a similar procedure, except that the cue types were block designs. Based on the design of Experiment 1b, Experiments 2 and 3 increased the number of memory items to test whether the memory load would modulate the attention selection modes. Twenty-eight participants were recruited for Experiment 1b, Experiment 2, and Experiment 3. All experiments were 2 (cue types: pre-cue, retro-cue) × 2 (change types: irrelevant change, irrelevant no-change) within-subjects designs, participants’ reaction times (RTs) and correct rates were recorded, and the sensitivity and criteria of the participants were calculated by signal detection theory (SDT).

The purpose of Experiment 1a and Experiment 1b was to investigate attentional selection modes in the VWM coding and maintenance stages under low memory load. The results of Experiment 1a showed that the main effect of change types in RTs [701 ms vs. 668 ms, F(1, 29) = 34.48, p < 0.001, η²_p = 0.54] and criteria [−0.15 vs. 0.15, F(1, 29) = 47.93, p < 0.001, η²_p = 0.62] was significant. In addition, the interaction between cue types and change types in criteria was significant, F (1, 29) = 19.98, p < 0.001, η²_p = 0.41. Pairwise comparisons showed that in the pre-cue trial, the criteria under the condition of irrelevant change was smaller (−0.19 vs. 0.25), t (29) = 9.62, p = 0.001, Cohen's d = 1.42, 95%CI = [1.12, 1.72]; In retro-cue trials, the criteria under irrelevant changes was also smaller (−0.11 vs. 0.05), t (29) = 2.55, p = 0.016, Cohen's d = 0.50, 95%CI = [0.10, 0.89] (see Table 1 and Figure 1). The results of Experiment 1b showed that the main effect of change types in RTs [739 ms vs. 722 ms, F (1, 27) = 10.14, p = 0.004, η²_p = 0.27] and criteria [−0.12 vs. 0.07, F(1, 27) = 27.87, p < 0.001, η²_p = 0.51] was significant (see Table 2), but the interaction in RTs [F (1, 27) = 2.55, p = 0.122] and criteria [F (1, 27) = 2.28, p = 0.143] was not significant (see Figure 2 A/B).

The purpose of Experiment 2 was to investigate attentional selection modes in the VWM coding and maintenance stages under middle memory load. The results of Experiment 2 showed that the main effect of change types in RTs [956 ms vs. 921 ms, F (1, 27) = 18.18, p < 0.001, η²_p = 0.40] and criteria [0.19 vs. 0.33, F(1, 27) = 16.23, p < 0.001, η²_p = 0.38] was significant (see Table 3). In addition, the interaction between cue types and change types in RTs was significant, F (1, 27) = 8.29, p = 0.008, η²_p = 0.24. Pairwise comparisons showed that in the pre-cue trial, the criteria under the condition of irrelevant change was smaller (915 ms vs. 860 ms), t(27) = −6.07, p < 0.001, Cohen’s d= −0.61, 95% CI = [−0.82, −0.40]; In retro-cue trials, however, there was no significant difference between irrelevant change and no-change conditions (998 ms vs. 983 ms), t(27) = −1.24, p = 0.227, Cohen’s d= −0.17, 95% CI = [−0.45, 0.11] (see Figure 2 C). The interaction between cue types and change types in criteria was significant, F (1, 27) = 14.10, p < 0.001, η²_p = 0.34. Pairwise comparisons showed that in the pre-cue trial, the criteria under the condition of irrelevant change was smaller (0.16 vs. 0.42), t(27) = 5.23, p < 0.001, Cohen’s d= 0.95, 95% CI = [0.58, 1.32]; However, there was no significant difference between irrelevant change and no-change condition in retro-cue trials (0.23 vs. 0.24), t(27) = 0.24, p = 0.816, Cohen’s d= 0.04, 95% CI = [−0.30, 0.37] (see Figure 2 D).

The purpose of Experiment 3 was to investigate attentional selection modes in the VWM coding and maintenance stages under high memory load. The results of Experiment 3 showed that the main effect of change types in RTs [1044 ms vs. 1009 ms, F (1, 27) = 11.17, p = 0.002, η²_p = 0.29] and criteria [0.27 vs. 0.41, F(1, 27) = 20.05, p < 0.001, η²_p = 0.43] was significant (see Table 4). In addition, the interaction between cue types and change types in criteria was significant, F (1, 27) = 16.90, p < 0.001, η²_p = 0.39. Pairwise comparisons showed that in the pre-cue trial, the criteria under the condition of irrelevant change was smaller (0.31 vs. 0.56), t(27) = 5.71, p < 0.001, Cohen’s d= 0.83, 95% CI = [0.53, 1.12] (see Figure 2 E); However, there was no significant difference between irrelevant change and no-change condition in retro-cue trials (0.23 vs. 0.25), t(27) = 0.38, p = 0.705, Cohen’s d= 0.05, 95% CI = [−0.21, 0.31] (see Figure 2 F).

We further conducted a mixed-design analysis of variance on the RTs and criteria of retro-cue in Experiment 1b, Experiment 2, and Experiment 3, combining them into 3 (memory load: low, middle, high) × 2 (change types: irrelevant change, irrelevant no-change) conditions. Memory load was treated as a between-subjects variable and task-irrelevant feature change type was treated as a within-subjects variable. The results revealed a marginally significant interaction effect on criteria, F(2, 81) = 3.10, p = 0.051, η²_p = 0.07. Pairwise comparisons showed that under low memory load condition, the difference between the irrelevant change condition and no-change condition was significant (−0.16 vs. −0.01), t(81) = 3.34, p = 0.001, Cohen’s d= 0.64, 95% CI = [0.26, 1.02]; However, under middle memory load (0.23 vs. 0.24), t(81) = 0.25, p = 0.807, Cohen’s d= 0.05, 95% CI = [−0.33, 0.43], and high memory load(0.23 vs. 0.25), t(81) = 0.35, p = 0.728, Cohen’s d= 0.07, 95% CI = [−0.31, 0.45], the difference between the irrelevant change condition and no-change condition was not significant (see Figure 3 A). In addition, we also calculated memory capacity K values for each condition, and conducted a mixed-design analysis of variance on them with 3 (memory load: low, middle, high) × 2 (cue type: pre-cue, retro-cue) conditions. The results indicated a significant interaction effect, F(2, 81) = 23.34, p < 0.001, η²_p = 0.37. Pairwise comparisons showed that under low memory load, there was no significant difference between pre-cue trials and retro-cue trials (1.91 vs. 1.92), t(81) = 0.17, p = 0.869. Under middle memory load, K values under pre-cue trials were significantly larger than those under retro-cue trials (2.77 vs. 2.28), t(81) = 4.73, p < 0.001, Cohen’s d= 0.91, 95% CI = [0.53, 1.29]. Under high memory load, K values under pre-cue trials were also significantly larger than those under retro-cue trials (3.00 vs. 2.02), t(81) = 9.50, p < 0.001, Cohen’s d= 1.82, 95% CI = [1.44, 2.20] (see Figure 3 B).

The results of the three experiments showed that the change in task-irrelevant features had an impact on task performance in the pre-cue trials, with longer RTs and lower criteria in the task-irrelevant feature change condition than in the no-change condition. This distracting effect was not modulated by the memory load. This suggests the existence of robust object-based attentional selection during the encoding stage in VWM. In contrast, in the retro-cue trials, the distracting effect was present only in the low memory load condition (Experiment 1a/1b) and disappeared when the memory load increased (Experiment 2/3). This suggests that during the maintenance stage, task-irrelevant features are processed only under low memory load conditions, and insufficient resources lead to their inability to be processed as the demand for attentional resources for task-relevant features increases.

In summary, the present study provides further evidence for the hypothesis that different modes of attentional selection exist in the encoding and maintenance stages of VWM, specifically that the attention selection mode during the VWM encoding stage is object- based, while the attention selection mode during the maintenance stage is feature-based and regulated by memory load. This study has important implications for resolving the controversy surrounding the attention selection mode of multifeature objects in VWM.

Key words: visual working memory, encoding stage, maintenance stage, object-based attention, feature-based attention