Can divided attention at retrieval improve memory? Effects of target detection during recognition

ABSTRACT

The attentional boost effect (ABE) is an improvement of memory under divided attention conditions in which stimulus encoding is enhanced when a target is detected in a simultaneous target-monitoring distracting task. Here we asked whether memory is similarly improved when the target-monitoring task occurs at the time of retrieval. In four experiments, participants encoded words under full attention then completed a recognition test under either divided attention, during which participants made recognition judgments while performing the target-monitoring task, or full attention, in which the target-monitoring task was not performed. Relative to distractor rejection, target detection increased hits and false alarms under divided attention with no net effect on discrimination. Targets and distractors had no effect on recognition under full attention. The target-related increase in hits and false alarms occurred regardless of whether the target-monitoring material matched or mismatched the test material and regardless of the target-to-distractor ratio and the target response. A change in bias accounts for the phenomenon, in which participants adopt a more lenient criterion for target-paired words than for distractor-paired words. The same divided attention manipulation that enhances memory at encoding does not similarly enhance memory at retrieval. Theoretical explanations are discussed.

KEYWORDS:

• Divided attention
• attentional boost effect
• recognition
• episodic memory

Distractions during encoding typically compromise memory. For example, when people try to learn information while simultaneously engaging in a second, unrelated task, attention is said to be divided and memory suffers relative to conditions in which the unrelated task is not performed and attention is undivided. The information to be learned could consist of verbal material such as individual words or word pairs, or it could consist of nonverbal material such as scenes or unfamiliar faces. Similarly, the secondary task could employ a range of materials such as numbers, letters, words, or tones, which may be processed in any number of ways. Additionally, the memory task could be free recall, cued recall, or recognition. Although there are some conditions under which the deleterious effect of divided attention (DA) on memory is negligible (e.g., Middlebrooks et al., 2017), it seems safe to say that encoding-phase DA impairs memory relative to full attention (FA) in most configurations of materials, secondary tasks, and explicit memory tests (see Kilb & Naveh-Benjamin, 2014; Long et al., 2018 for reviews).

This investigation focuses on the effects of DA during memory retrieval. Surprisingly, the negative effect of retrieval-phase DA is much more modest than DA at the time of encoding. For example, in a landmark study, various secondary tasks carried out during retrieval had small and often nonsignificant effects on word recall and recognition accuracy. In contrast, those secondary tasks dramatically reduced recall and recognition when carried out at the time of encoding (Baddeley et al., 1984). That finding led to the idea that retrieval is automatic insofar as it did not require attentional resources. Subsequent research demonstrated that the memory task did, in fact, require attentional resources because the memory task slowed performance on the secondary task relative to performing the secondary task alone (Craik et al., 1996; Naveh-Benjamin et al., 2000). In general, retrieval processes appear to be obligatory or protected from the deleterious effects of retrieval-phase DA (Craik et al., 2018; Kellogg et al., 1982). Nevertheless, many studies do report recall and recognition impairments under retrieval-phase DA (e.g., Clarke & Butler, 2008; Guez & Naveh-Benjamin, 2013; Park et al., 1989; Prull et al., 2016; Sarhane & Daurat, 2019). Impairments to memory during retrieval-phase DA appear to be most pronounced when retrieval is dominated by recollective processes (Hicks & Marsh, 2000; Lozito & Mulligan, 2006; Troyer et al., 1999), or when the materials in the secondary task match those of the memory test (e.g., both tasks involve words) rather than mismatch them (e.g., one task involves words and the other involves numbers; Fernandes & Moscovitch, 2000, 2002, 2003; Skinner & Fernandes, 2008). Similar materials in the retrieval and secondary tasks are thought to create competition in material-specific representational systems, leading to interference and reduced retrieval accuracy.

These previous studies have all focused on the memory impairing effects of DA at retrieval, however in this investigation we focus on the potential memory enhancing effects of retrieval-phase DA. Recent evidence suggests that DA can, in fact, improve memory when DA occurs at encoding. In the attentional boost effect (ABE), target detection in a secondary target-monitoring task enhances memory for concurrently presented material, relative to distractor rejection (Lin et al., 2010; Swallow & Jiang, 2010). For example, participants might encode a list of photographic scenes each of which is presented with a small square in the centre of the image. Participants are asked to encode the scenes and monitor the colour of the squares. When a pre-specified target appears (e.g., white square), memory for the corresponding scene is enhanced – often to the level achieved under FA at encoding and sometimes exceeding that level – compared to when a distractor appears (e.g., black square). This effect occurs for a variety of to-be-remembered stimuli such as scenes, words, faces, and colour arrays, and for a variety of tests including recognition, free recall, and cued recall (Makovski et al., 2011; Mulligan et al., 2014; Prull, 2019; Spataro et al., 2013, 2015; Spataro et al., 2017; Swallow & Jiang, 2010, 2011, 2014; for a review, see Swallow & Jiang, 2013). This effect does not depend on the distinctiveness or rarity of targets because the ABE occurs when targets are as frequent as distractors (e.g., Swallow & Jiang, 2012) and the effect is not dependent on an overt motor response to targets because silent counting of targets produces the ABE as well (Swallow & Jiang, 2012). The effect is temporally precise, however, as memory is not enhanced for items preceding or succeeding the target by 100 ms. In addition, the effect is diminished when stimulus exposure times are relatively long (e.g., 4 s) because controlled or elaborative encoding processes that accrue over this time dilute the beneficial effect of target processing, which occurs almost immediately after stimulus onset (Mulligan & Spataro, 2015).

According to Swallow and colleagues (Swallow et al., 2022; Swallow & Jiang, 2013), an attentional process known as temporal selective attention is engaged when a response to a target is made, briefly enhancing the perceptual processing of the target and any co-presented material. This enhanced processing leads to improved encoding and the ABE on a later memory test. Ordinarily, dual-task situations create more interference than enhancement, leading to the usual decrement in memory following DA encoding, but under specific conditions the enhancing effect of temporal selection dominates over the impairing effect of interference, leading to the ABE. Those specific conditions include simultaneous presentation of target and to-be-remembered material (Swallow & Jiang, 2011), and relatively minimal target processing demands (Swallow & Jiang, 2010).

How would the target monitoring task affect memory if the task occurred during the retrieval phase? One prediction derives from a set of theories and principles that highlight the similarity between encoding and retrieval processes. This set includes functional accounts of memory (e.g., Craik, 1983; Kolers, 1973) and the principles of transfer-appropriate processing and encoding specificity (Morris et al., 1977; Tulving, 1983), all of which stress the idea that retrieval processes essentially recapitulate encoding processes. From this perspective, target detection at retrieval should elicit the same salutary effects on memory as target detection at encoding. For example, retrieval-phase target detection could enhance access to perceptual representations of stimuli and their contexts (Swallow & Jiang, 2010, 2013). Presumably, retrieval-phase target detection would selectively enhance memory for previously studied items, as memorial information that would be recovered through target processing would exist only for studied items, rather than new items for which there are no episodic representations to reinstate. Although the goal of this study was not to test the aforementioned theoretical frameworks directly, those frameworks nevertheless suggest a retrieval-enhancement hypothesis that predicts a selective enhancement of the hit rate following retrieval-phase target detection. If confirmed, that result would imply that temporal selective attention is a process that extends beyond encoding and can benefit retrieval as well.

A different prediction can be made from theoretical perspectives that emphasise different types of attention at encoding and retrieval. This theoretical view is perhaps best represented by the perceptual/reflective attention/memory (PRAM) framework, which makes a distinction between perceptual attention that is directed toward objects and features in the environment during encoding and reflective attention that is directed toward internal representations during retrieval (Chun & Johnson, 2011; Long et al., 2018). According to this perspective, target detection at retrieval would not necessarily produce the same effect on memory as target detection at encoding because a different form of attention – reflective attention – is operating at retrieval. This viewpoint would therefore not predict a selective effect of target detection on hit rates as the retrieval-enhancement hypothesis suggests.

We are aware of two studies that have investigated the effect of retrieval-phase target detection on memory. Makovski et al. (2011, Experiment 2a) presented colour patch arrays that were followed 1,500 ms later by a test array that was either identical to the preceding array or differed by one colour. The test array also contained a letter, which could either be a target (the letter T) or a distractor (other letters). Participants first responded to the letter, pressing a button if it was a target and withholding their response if it was a distractor, then made a same/different judgment on the test array. Targets had no significant effect on memory performance relative to distractors. In another study, Huang and Meng (2020) tested memory for words while participants simultaneously monitored the screen for the appearance of a target (a plus sign), pressing a button when the target appeared and withholding their response when a distractor appeared (a minus sign). In a second experiment, participants silently counted targets rather than responding with a keypress. In both experiments, participants were more likely to judge a test word as studied if the word appeared with a target than if the test word appeared with a distractor (i.e., both hits and false alarms increased in the presence of targets). Neither experiment produced a target-related improvement to recognition sensitivity.

Taken together, these studies suggest that retrieval-phase target detection during a working memory task has no significant effect on same/different judgments (Makovski et al., 2011), and that retrieval-phase target detection in a long-term recognition memory task increases hit and false alarm rates (Huang & Meng, 2020). Both outcomes appear to be inconsistent with the retrieval-enhancement hypothesis and theories that emphasise similarities between encoding and retrieval processes. The study by Makovski et al. had few participants (n = 10) and investigated working memory for nonverbal stimuli, which differs from the present experiments that involve long-term memory for verbal materials. Therefore, the Makovski et al. study is not considered further. The experiments by Huang and Meng (2020) closely resemble those of the present experiments and suggest that target detection increases positive recognition responses to both studied and new items. However, those experiments did not include a full attention condition in which participants completed the memory task in the absence of target monitoring. Without that condition, it is unclear whether target detection processes are necessary to increase hits and false alarms or whether the mere presence of targets and distractors that are ignored in a full attention condition is sufficient to produce these results. Therefore, further investigation is warranted.

Here we report four experiments in which we presented recognition test items with targets and distractors to determine how target detection impacts memory performance. In each experiment, participants first encoded a list of words without distraction then completed a yes/no recognition test in which each test item was presented simultaneously with a secondary task stimulus (a single-digit number). In the DA condition, participants responded to targets (odd numbers) and rejected distractors (even numbers) while simultaneously making recognition judgments. In the FA condition, participants ignored the secondary task stimuli and made only recognition judgments. Our interest focused primarily on the DA condition. An increase in hits but not false alarms for target-paired items in the DA condition would be consistent with the retrieval-enhancement hypothesis and theoretical views that emphasise the similarity of encoding and retrieval operations. Null effects of target detection or impairments in memory associated with target detection would be consistent with the PRAM framework and other such frameworks that emphasise different forms of attention operating at encoding and retrieval.

Experiment 1

Method

Design and participants

The design was a 2 (attention: FA, DA) X 2 (test status: studied, new) X 2 (trial type: target, distractor) mixed factorial with the first factor manipulated between groups and the remaining factors manipulated within groups. We determined the sample size a priori by using PANGEA, an online power analysis calculator for general ANOVA designs (Westfall, 2015). According to that analysis, 54 participants were needed in this design to achieve power of .95 to detect a medium-sized effect (Cohen’s d = .50) of trial type. We tested 56 participants in all (28 participants per attention group) to achieve complete counterbalancing of materials to conditions. All participants were undergraduate students at Whitman College who received course credit or pay and all experimental procedures were approved by the Institutional Review Board at Whitman College.

Materials

We selected 80 words from the MRC Database (Coltheart, 1981) to use as critical items. Critical items were three to ten letters in length (M = 5.50) and were mostly medium frequency nouns (Kuçera-Francis M frequency = 90, range 10-686). These words were divided into four lists of 20 words and each list was assigned to one of four conditions: target-paired studied words, target-paired new words, distractor-paired studied words, and distractor-paired new words. The assignment of lists to conditions was rotated across participants so that each list occurred in each condition an equal number of times. The four lists did not differ in measures of concreteness, familiarity, imageability, frequency, or word length, all Fs < 1. All 80 critical stimuli appeared at test together with 240 filler words for a total of 320 recognition test words. The filler words were also predominantly medium-frequency words (M = 101, range 10-897) that were 3–10 letters in length (M = 5.85). Filler words did not differ significantly from critical words on measures of concreteness, familiarity, imageability, frequency, or word length, all Fs≤ 2.13, all ps≥.15.

Procedure

We tested each participant individually in a single session that consisted of a study phase, a distraction phase, and a test phase (see Figure 1). In the study phase, participants encoded 44 words of which two were primacy buffers, two were recency buffers, 20 were critical words that would be paired with targets at test, and 20 were critical words that would be paired with distractors at test. Primacy and recency buffers were constant across all participants and the arrangement of critical words in the study list was random. Participants were instructed to pronounce each word aloud and to try to remember the words for a later memory test. Words were presented individually in black 48-point Arial font against a white background, each for 2 s in the centre of the computer screen followed by a 500 ms blank screen interstimulus interval.

Figure 1. Depiction of study and test phases. See text for details.

Participants advanced to the distraction phase immediately after the study phase. They received a sheet with 99 word stems (e.g., Tur___) and were given two minutes to complete as many stems as possible with country names, writing their responses on the lines provided.

The test phase followed immediately after the distraction phase. Participants made yes/no recognition decisions on 320 words, of which 80 were critical words and 240 were additional unstudied words that were used as filler items. The purpose of adding filler items to the test phase was to mirror the traditional design used in previous studies of the ABE in which many filler items intervene between critical items during encoding (Mulligan et al., 2014; Prull, 2019; Swallow & Jiang, 2010). These filler items remained constant for all participants. Half of the critical words (20 studied, 20 new) were paired with targets, the other half (20 studied, 20 new) were paired with distractors. All filler words were paired with distractors. A target was specified as a single-digit odd number (1, 3, 5, 7, or 9) that appeared below the test word; a distractor was specified as a single-digit even number (2, 4, 6, or 8). Note that target monitoring tasks with multiple target options have been used in previous research and produce an ABE when those tasks occur at encoding (Turker & Swallow, 2019).

We prepared test lists by creating 80 blocks of four words each. Each block consisted of three fillers and one critical word. Across blocks, critical words appeared equally often in the first, second, third, or fourth position, then the blocks were joined together in a single random order with the stipulation that a pair of critical words had at least one intervening filler word separating them. In this way, each block contained one test word from each of the four conditions (target-paired studied words, target-paired new words, distractor-paired studied words, and distractor-paired new words). This arrangement was unknown to the participants, as the test list appeared as a continuous series of 320 words with a short break provided halfway through the test.

On each trial, a test word and number appeared in the centre of the screen with the number immediately below the word for 500 ms, after which only the word remained for an additional 1,300 ms. The word disappeared after a total presentation time of 1,800 ms and was followed by a 200 ms blank screen interstimulus interval. Participants who were randomly assigned to the FA condition made a yes/no recognition response to each test word while ignoring the co-presented numbers, pressing the slash key (/) on the keyboard for a yes response (a positive response indicating that they recognised the word from the study phase), or the z key for a no response (a negative response indicating that they did not recognise the word). Participants who were randomly assigned to the DA condition carried out the recognition task while simultaneously monitoring the numbers, saying “hit” aloud each time an odd number target appeared and remaining silent if an even number distractor appeared. We instructed participants to complete both tasks to the best of their abilities. The experimenter said “miss” for any missed targets to keep participants on task. On target trials, participants were free to respond to either the word first or the number first.¹ A practice phase of 20 trials (two targets) preceded the 320-item test in both FA and DA conditions.

Results

Encoding and secondary task performance

In all analyses, α = .05, two-tailed, unless otherwise indicated. At encoding, participants successfully pronounced the words at near-perfect levels and average performance did not differ significantly across groups (FA = 99.5%, DA = 99.9%), t < 1. Within the DA condition, participants correctly identified odd numbers as targets 95.0% of the time (SD = 5.05) and they erroneously identified even numbers as targets less than 1% of the time (M = 0.45%, SD = 0.98).

Memory performance

We analysed the recognition data in two ways, conditionalized on valid trials and unconditionalized. Conditionalized analyses involved removing all invalid test trials and computing proportions from the remaining valid trials. An invalid test trial was one in which any of the following events occurred: (a) the word was improperly encoded, in which the participant misidentified the word or gave no response in the allotted time during the study phase, (b) a response timeout occurred in the test phase, in which the participant did not make a recognition response within the allotted time, (c) in the DA condition, the participant did not give a “hit” response to a target during the test phase, or (d) in the DA condition, the participant gave a “hit” response incorrectly to a distractor during the test phase. This process led to the removal of 1.12% and 5.58% of the critical trials in the FA and DA conditions, respectively. Unconditionalized data analyses did not exclude any recognition trials. Analyses of conditionalized data led to identical conclusions to those drawn from unconditionalized data except in minor cases where noted. Statistical results and table values represent the unconditionalized data in all experiments unless specified otherwise.

The results from Experiment 1 are shown at the top of Table 1. Values represent the proportion of yes responses to studied (hits) and unstudied words (false alarm rates; FARs). Under FA retrieval, targets did not significantly change hits or FARs compared to distractors. However, under DA retrieval, targets increased hits and FARs compared to distractors.

Table 1. Recognition performance, experiments 1–4 (SE in parentheses).

	FA		DA-Numbers		DA-Words
Experiment & Trial Type	Hits	FAR	Hits	FAR	Hits	FAR
Experiment 1
Distractors	.52 (.03)	.16 (.03)	.44 (.04)	.16 (.03)
Targets	.50 (.03)	.13 (.02)	.52 (.03)	.29 (.04)
Experiment 2
Distractors	.51 (.04)	.11 (.02)	.36 (.04)	.13 (.03)	.41 (.03)	.15 (.03)
Targets	.48 (.04)	.12 (.02)	.48 (.04)	.28 (.05)	.46 (.04)	.27 (.04)
Experiment 3
Distractors	.70 (.03)	.14 (.02)	.47 (.03)	.15 (.02)
Targets	.70 (.03)	.13 (.02)	.68 (.03)	.32 (.02)
Experiment 4
Distractors			.44 (.03)	.14 (.02)
Targets			.67 (.03)	.34 (.04)

Note. FA = full attention, DA = divided attention, FAR = false alarm rate.

Those observations were supported by a 2 (attention: FA, DA) X 2 (test status: studied, new) X 2 (trial type: target, distractor) mixed analysis of variance (ANOVA) with repeated measures on the latter two factors applied to the proportions of yes responses in the recognition test. That analysis revealed the following effects. First, a main effect of test status indicated that hits were significantly higher than FARs, F(1,54) = 260.78, p < .001, η_p²= .83, however, that effect was qualified by a significant interaction between test status and attention, suggesting that DA at retrieval negatively impacted recognition, F(1,54) = 7.52, p = .008, η_p²= .12. Follow-up analyses of the interaction indicated that DA had no significant effect on hit rates compared to FA (M = .48 vs. .51, respectively), F < 1, but DA significantly elevated false alarm rates (M = .23 vs. .15), F(1,54) = 4.53, p = .04, η_p²= .08. Thus, DA impaired recognition performance overall by increasing FARs.

Continuing with the main analysis, an effect of trial type that was nonsignificant in the unconditionalized data but significant in the conditionalized data (p = .03) indicated that participants responded yes to target-paired words more frequently than distractor-paired words overall, F(1,54) = 3.33, p = .07, η_p²= .06. Most importantly, a significant attention X trial type interaction emerged, F(1,54) = 8.59, p = .005, η_p²= .14. A follow-up analysis of the attention X trial type interaction revealed that targets and distractors had a nonsignificant effect on recognition in the FA condition (M = .32 and .34, respectively), F(1,27) = 2.04, p = .17, η_p²= .07, however, in the DA condition, participants responded yes to target-paired words significantly more than distractor-paired words (M = .41 and .30, respectively), F(1,27) = 6.66, p = .02, η_p²= .20. This finding suggests a retrieval-phase effect of target processing under DA but not under FA. No additional effects or interactions were significant in the main analysis, including the three-way interaction, all Fs≤1.89, all ps≥.17, all η_p²s≤.03.

Signal detection analysis

Figure 2 shows the unequal-variance signal detection measure of sensitivity (d_a) and Figure 3 shows the corresponding measure of bias (c_a). Increasingly positive values of d_a indicate better discrimination between old and new items and increasingly positive values of c_a indicate increasingly conservative response thresholds (i.e., a bias to respond no, indicating that the word is new). We calculated these values after applying the log-linear transformation to all hits and false alarms, which enabled us to calculate d_a and c_a when hit rates were 1.0 or when FARs were zero (Snodgrass & Corwin, 1988). We assumed σ_new/σ_old= .80 in our calculations of these measures (Mickes et al., 2007).

Figure 2. Mean discrimination scores (d_a) as a function of attention condition (FA = full attention, DA = divided attention) and trial type, Experiments 1–4 (error bars indicate ± 1 SE).

Figure 3. Mean bias scores (c_a) as a function of attention condition (FA = full attention, DA = divided attention) and trial type, Experiments 1–4 (error bars indicate ± 1 SE).

Figure 2 suggests that DA reduced measures of sensitivity (d_a), and that observation was confirmed in a 2 (attention: FA, DA) X 2 (trial type: target, distractor) mixed ANOVA. Only the main effect of attention was significant in that analysis, in which DA reduced d_a values compared to FA (M = 0.68 vs. 1.04), F(1,54) = 7.40, p < .01, η_p²= .12. All other effects were nonsignificant, all Fs≤2.55, all ps≥.11, all ��_p²s≤.05.

Figure 3 suggests that targets lowered bias scores in the DA condition, suggesting that participants adopted a more liberal criterion to target-paired words than to distractor-paired words. A similar 2 X 2 ANOVA on bias scores (c_a) supported that conclusion by returning a nonsignificant effect of trial type, F(1,54) = 3.48, p = .07, η_p²= .06,² no significant main effect of attention (F < 1), and a significant interaction between attention and trial type, F(1,54) = 10.02, p = .003, η_p²= .16. Follow-up analyses of the interaction revealed that bias scores did not differ significantly between target and distractor trial types in the FA condition (M = 0.58 vs. 0.50, respectively), F(1,27) = 2.47, p = .13, η_p²= .08, but in the DA condition, target-paired words had significantly lower c_a values compared to distractor-paired words (M = 0.27 vs. .60, respectively), F(1,27) = 7.63, p = .01, η_p²= .22.

Discussion

Experiment 1 produced several notable effects. In the FA condition, targets and distractors that were ignored at the time of retrieval had little effect on recognition performance. That finding suggests that the mere presence of targets and distractors does not influence recognition performance and addresses the limitation in Huang and Meng (2020) that we described in the introduction. However, in the DA condition, target detection was associated with increased hits and FARs for the accompanying words compared to distractor rejection. Target processing did not selectively increase hits, as the three-way interaction that would support such a conclusion was not significant. Rather, the evidence is most consistent with the conclusion that targets enhance positive recognition responses, regardless of whether the test item is studied or new. This enhancement was reflected as a change in bias or criterion whereby participants adopted a more liberal response threshold for target-paired words than for distractor-paired words in the DA condition. Finally, we note that retrieval-phase DA impaired recognition overall, as shown by an increase in FARs and reduced sensitivity values for DA in the signal-detection analysis.

Our results suggest that retrieval-phase target processing increases the likelihood of responding positively to recognition test items. Our goals in Experiment 2 were to replicate this result and to learn whether the nature of the secondary task influences the magnitude of this effect. As discussed in the introduction, Fernandes and Moscovitch (2000, 2002, 2003; Skinner & Fernandes, 2008) reported deleterious effects of retrieval-phase DA on memory that were more pronounced when secondary task materials matched those of the memory task (e.g., word-word) than when they mismatched (e.g., number-word, as was used in Experiment 1). Retrieval-phase DA effects may therefore depend on how much the secondary task materials interfere with access to material-specific representations in memory; when materials match, interference is thought to be high and DA effects on memory are pronounced. When materials mismatch, interference is relatively low and DA effects are minimised.

It is currently unknown whether material-specific effects modulate the target-related increase in positive recognition responses. Therefore, in Experiment 2 we used a word-based secondary task to explore this question and to determine whether the positive response shift in recognition generalises to different secondary tasks. We also retained the DA task involving numbers to replicate our results and to provide a comparison for the word-based secondary task.

Experiment 2

Method

Design and participants

The design was a 3 (attention: FA, DA-numbers, DA-words) X 2 (test status: studied, new) X 2 (trial type: target, distractor) mixed factorial with the first factor manipulated between groups and the remaining factors manipulated within groups. We increased the number of participants to 84 because this experiment included a third attention condition (28 participants per attention group). All participants were undergraduate students from Whitman College in Experiment 2, each of whom received course credit or pay.

Materials and procedure

All aspects of this experiment were identical to Experiment 1 except for the following changes. First, we included an additional attention condition, DA-words, in which participants made recognition judgments on words while simultaneously monitoring other words that were co-presented with the recognition test items. Specifically, participants responded “hit” when a word representing a manufactured object appeared below the test item (targets) and made no response when a word representing a naturally occurring object appeared (distractors). The DA-numbers condition was identical to the DA condition of Experiment 1, in which participants made recognition judgments while simultaneously monitoring for odd numbers. To create a parallel structure for the DA-words task, the materials for the word monitoring task involved nine three-letter words, of which five represented manufactured objects (cup, hat, ink, pen, pie) and four represented naturally occurring objects (bee, cat, log, sun). One recognition test item and one number (DA-numbers) or one recognition test item and one word (DA-words) was presented on each trial. The stimuli in the secondary tasks appeared immediately below the recognition test item, as in Experiment 1. The second change was that the secondary task stimuli always appeared in blue to differentiate them perceptually from the recognition test items, which appeared in black.

Results

Encoding and secondary task performance

Word identification was virtually perfect at encoding across FA, DA-numbers, and DA-words conditions (M = 100%, 99.9%, and 100%, respectively). In the two DA conditions, participants correctly responded to targets significantly more in the DA-numbers condition (M = 92.1%, SD = 7.00) than in the DA-words condition (M = 82.3%, SD = 12.07), t(54) = 3.72, p < .001. In addition, participants erroneously responded to distractors significantly less often in the DA-numbers condition (M = 0.27%, SD = 0.79) than in the DA-words condition (M = 1.07%, SD = 1.73), t(54) = 2.24, p = .03. Therefore, target monitoring was more successful in the DA-numbers condition than in the DA-words condition. We return to this issue later in this section.

Memory performance

Trials were identified as invalid according to the definitions described in Experiment 1 and were removed for conditionalized analyses. Invalid trials occurred 0.71%, 6.38%, and 11.65% of the time for critical trials in the FA, DA-numbers, and DA-words conditions, respectively. However, as in Experiment 1, the conclusions did not differ between conditionalized and unconditionalized data sets except in one case where noted, therefore the results from the unconditionalized data are reported.

The results in Table 1 suggest that (a) DA reduced memory performance overall relative to FA, (b) targets and distractors did not significantly affect hits or FARs under FA encoding, however, (c) under both DA conditions, target detection elevated hits and FARs compared to distractor rejection.

These observations were supported by a 3 (attention: FA, DA-numbers, DA-words) X 2 (test status: studied, new) X 2 (trial type: target, distractor) mixed ANOVA with repeated measures on the latter two factors applied to the proportions of yes responses in the recognition test. This primary analysis revealed the following effects. A main effect of test status indicated that hits were significantly higher than FARs overall, F(1,81) = 206.69, p < .001, η_p²= .72, but that effect was qualified by a test status X attention interaction, F(2,81) = 7.55, p = .001, η_p²= .16. A follow-up analysis of that interaction indicated that the attention manipulation did not significantly influence hits (M = .50, .42, and .44 for FA, DA-numbers, and DA-words conditions, respectively), F(2,81) = 1.20, p = .31, η_p²= .03, but it did influence FARs, although that effect just missed the standard criterion for significance (M = .12, .20, and .21 for FA, DA-numbers, and DA-words conditions, respectively), F(2,81) = 3.03, p = .054, η_p²= .07.³ In a second follow-up analysis, the test status X attention interaction remained significant when only the FA and DA-numbers conditions were analysed in a 2 X 2 ANOVA (p = .001) and when only the FA and DA-words conditions were analysed (p = .001). However, the interaction was not significant when the two DA conditions were directly compared (p = .85). Thus far, these results suggest that the two DA conditions negatively affected recognition performance, but they did not differ in the magnitude or manner by which they did so.

Returning to the primary analysis, the main effect of trial type was statistically significant, F(1,81) = 25.27, p < .001, η_p²= .24, suggesting that recognition differed for target-paired and distractor-paired words. However, most importantly, attention interacted with trial type, F(2,81) = 9.55, p < .001, η_p²= .19. Follow-up analyses indicated that targets and distractors had no significant influence on FA performance, F < 1. In contrast, participants responded yes to target-paired words significantly more than distractor-paired words in the DA-numbers condition (M = .38 vs. .25, respectively), F(1,27) = 27.44, p < .001, η_p²= .50, and in the DA-words condition (M = .37 vs. .28, respectively), F(1,27) = 13.40, p = .001, η_p²= .33. Additional follow-up analyses revealed that the trial type X attention interaction remained significant when the FA and DA-numbers conditions were directly compared in a 2 X 2 ANOVA (p < .001) and when the FA and DA-words conditions were directly compared (p = .004), but not when the two DA conditions were directly compared (p = .19). Thus, in the two DA conditions, targets increased hits and FARs relative to distractors and the magnitude of this increase did not differ across the conditions. In the FA condition, targets had nonsignificant effects on recognition.

A final result from the primary analysis was that the interaction between test status and trial type was significant, F(1,81) = 7.22, p = .009, η_p²= .08. A follow-up analysis indicated that, averaged across attention conditions, targets elevated FARs significantly more than they elevated hits. All remaining effects in the primary analysis were not significant, all Fs < 1.

Signal detection analysis

We calculated measures of sensitivity (d_a) and bias (c_a) in a manner identical to Experiment 1. The sensitivity results, depicted in Figure 2, suggest that the two DA conditions reduced d_a scores, reflecting a significant decrement to recognition performance under divided attention. That observation was confirmed in a 3 (attention: FA, DA-numbers, DA-words) X 2 (trial type: target, distractor) mixed ANOVA in which the main effect of attention was significant, F(2,81) = 7.35, p = .001, η_p²= .15. A follow-up one-way ANOVA using Tukey’s HSD test indicated that the FA condition had significantly higher d_a scores (M = 1.10) than either DA-numbers or DA-words conditions (M = 0.62 and 0.63, respectively), but that the two DA conditions did not differ significantly from each other. To complete the ANOVA results, the effect of trial type was significant, F(1,81) = 6.69, p = .01, η_p²= .08, indicating that sensitivity for target-paired words was worse compared to distractor-paired words overall (M = 0.70 vs. 0.86, respectively). The interaction between attention and trial type was not significant, F < 1.

A similar 2 X 3 ANOVA on the bias scores depicted in Figure 3 revealed a significant main effect of trial type, F(1,81) = 33.95, p < .001, η_p²= .30, and a significant interaction between trial type and attention, F(2,81) = 11.29, p < .001, η_p²= .22. The main effect of attention was not significant, F < 1. Following up on the interaction, bias scores for target-paired and distractor-paired conditions did not differ in the FA condition, F < 1. However, in both the DA-numbers and DA-words conditions, target-paired test items had significantly lower bias scores, indicating a more liberal response bias, compared to distractor-paired test items, F(1,27) = 37.63, p < .001, η_p²= .58 for DA-numbers and F(1,27) = 18.67, p < .001, η_p²= .41 for DA-words. Further 2 X 2 ANOVAs revealed a significant trial type X attention interaction when the analysis included only the FA and DA-numbers conditions (p < .001), or only the FA and DA-words conditions (p = .003). However, the interaction was nonsignificant when the two DA conditions were compared (p = .10). These results suggest that the target-related increase in hits and FARs in the two DA conditions reflects a more liberal response bias for target-paired words and that the type of material used in DA conditions (numbers vs. words) did not affect the magnitude of that liberal shift in bias.

Recognition performance after matching on DA performance

Although the differences in recognition performance were minimal across the two DA groups, the DA-words task was significantly more difficult than the DA-numbers task, a fact that complicates the interpretation of the recognition data. Therefore, we repeated the primary analyses of memory and signal detection measures after matching the two DA groups on target detection (N = 18 in each DA group, target detection = 88% in both groups, t < 1). The relevant tables are provided in the Appendix. The conclusions were unchanged from the analysis of the full samples. Specifically, in the recognition data, the interaction between trial type and attention that signalled a shift in positive responses in the two DA conditions but not the FA condition remained significant, F(2,61) = 6.52, p = .003, η_p²= .18, with no higher-order interaction, F < 1. In the signal detection measures, the main effect of attention on d_a scores remained significant, indicating worse performance in the DA conditions, F(2,61) = 5.18, p = .008, η_p²= .15, and the interaction between attention and trial type on c_a scores also remained significant, indicating that participants adopted a more liberal criterion for target-paired words than for distractor-paired words in the DA condition, F(2,61) = 6.87, p = .002, η_p²= .18.

Discussion

Experiment 2 replicated the results of Experiment 1 by showing that target processing at retrieval enhances hits and FARs and that this enhancement can be understood as a change in bias rather than sensitivity. The novel aspect of Experiment 2 is that the target-related enhancement generalised to a second DA condition in which targets and distractors consisted of verbal material, the same type of material used in the recognition test. No significant differences emerged between the two DA conditions in terms of this enhancement. The evidence is consistent with the conclusion that material type manipulations in the secondary task do not modulate the target-related enhancement of positive responses.⁴

Experiment 3

In Experiments 1 and 2, participants completed a recognition test in which targets were more likely to accompany studied items than new items and distractors were more likely to accompany new items than studied items. As described earlier, this arrangement of infrequent targets among frequent distractors is common in ABE studies when target and distractors are presented at encoding (Mulligan et al., 2014; Prull, 2019; Swallow & Jiang, 2010). Further work has determined that the encoding-phase ABE is not dependent on the relative rarity of targets and occurs when the ratio of targets and distractors is 1:1 (Makovski et al., 2011; Swallow & Jiang, 2012; but see Au & Cheung, 2020). Nevertheless, rare targets at retrieval could influence recognition responses. Specifically, participants in the DA condition could develop a strategy of responding positively to targets and negatively to distractors. At the extreme, participants could adopt this strategy regardless of the studied/new status of the recognition test item. This explanation could account for our results.

Therefore, in Experiment 3 we eliminated all filler trials, which consisted of unstudied items always paired with distractors, so that participants were tested only on studied items and an equal number of new items. Eliminating the fillers equates the number of studied and new test items and creates a 1:1 ratio of targets and distractors within each item type. If the target-related increase in hits and FARs occurs because participants adopt a strategy of responding based on the frequency and rarity of the targets and distractors, then no such increase should be seen in Experiment 3. However, if the increase seen in Experiments 1 and 2 reflects other mechanisms, then the same pattern of results that we observed in the previous experiments should occur in Experiment 3.

Method

Design and participants

The design of Experiment 3 was identical to that of Experiment 1. Fifty-six undergraduate students from Whitman College participated in Experiment 3 (28 participants in each attention condition) in exchange for course credit or pay and their data were included in the analyses that follow. Data from seven additional participants were collected but were excluded from the analyses reported here because those participants did not meet our required minimum level of responding to at least 70% of the recognition test items with either a yes or no response.

Materials and procedure

All aspects of this experiment were identical to Experiment 1 except for the recognition memory test, in which all filler words were removed. The recognition test consisted of the 80 critical items, half of which were studied and half of which were unstudied. Half of the studied and half of the unstudied items appeared with odd numbers, the remaining items appeared with even numbers. The items were arranged so that no more than four studied items, four new items, four odd numbers, or four even numbers occurred consecutively. Participants practiced on eight items (four studied words drawn from the primacy and recency buffers and four new words, with half of each type of item presented with odd numbers and the remaining items presented with even numbers) before initiating the 80-item test.

Results

Encoding and secondary task performance

Participants identified words nearly perfectly at encoding and performance did not differ between FA and DA groups (99.7% and 100%, respectively), t(54) = 1.00, p = .32. Within the DA condition, participants correctly identified odd numbers as targets 92.8% of the time (SD = 7.53) and incorrectly identified even numbers as targets 4.0% of the time (SD = 3.36).

Memory performance

Invalid trials occurred 1.43% and 11.83% of the time in the FA and DA conditions, respectively, and were removed for conditionalized analyses. Those analyses led to identical conclusions to those derived from unconditionalized data. The analyses that follow are based on the unconditionalized data.

The results of Experiment 3 are presented in Table 1. A 2 (attention: FA, DA) X 2 (test status: studied, new) X 2 (trial type: target, distractor) mixed ANOVA on the proportion of positive responses to recognition items produced the following effects. First, hits were significantly higher than false alarms overall as indicated by a main effect of test status, F(1,54) = 503.05, p < .001, η_p²= .90. However, that main effect was qualified by an interaction involving attention, F(1,54) = 30.76, p < .001, η_p²= .36. Follow-up 2 X 2 ANOVAs indicated that DA reduced the hit rate relative to FA (M = .57 vs. .70, respectively), F(1,54) = 10.59, p = .002, η_p²= .16, and increased the FAR (M = .23 vs. .14, respectively), F(1,54) = 15.19, p < .001, η_p²= .22. Thus, compared to FA, DA impaired recognition by reducing hits and increasing FARs.

Returning to the main analysis, participants gave positive responses to target-paired words more frequently than distractor-paired words overall, as indicated by a significant main effect of trial type, F(1,54) = 36.33, p < .001, η_p²= .40, but most importantly, this effect interacted with attention, F(1,54) = 40.66, p < .001, η_p²= .43. All remaining effects in the main analysis were not significant, all Fs < 1. A follow-up analysis of the trial type X attention interaction indicated that participants did not significantly differ in their recognition responses to target-paired and distractor-paired words under FA (M = .42 in both conditions), F < 1, but participants responded positively to target-paired words significantly more than distractor-paired words under DA (M = .50 vs. .31, respectively), F(1,27) = 52.92, p < .001, η_p²= .66. Thus, targets enhanced positive recognition responses, a result similar to that of Experiments 1 and 2.

Signal detection analysis

Figure 2 suggests that DA reduced measures of sensitivity (d_a), and that observation was confirmed in a 2 (attention: FA, DA) X 2 (trial type: target, distractor) mixed ANOVA. Only the main effect of attention was significant, in which DA reduced d_a values compared to FA (M = 0.93 vs. 1.58), F(1,54) = 21.25, p < .001, η_p²= .28. Neither the effect of trial type nor the interaction between attention and trial type was significant in this analysis, both ps > .25.

A similar 2 X 2 ANOVA on bias scores (c_a) led to the same conclusion as previous experiments, namely, that participants in the DA condition adopted a more liberal criterion to target-paired words than to distractor-paired words (see Figure 3). Specifically, the ANOVA yielded significant effects of trial type, F(1,54) = 33.45, p < .001, η_p²= .38, no significant main effect of attention (F < 1), and a significant interaction between attention and trial type, F(1,54) = 39.68, p < .001, η_p²= .42. Follow-up analyses of the interaction revealed no significant difference between target and distractor trial types in the FA condition (M = 0.28 vs. 0.25, respectively), F < 1, but in the DA condition, target-paired words had significantly lower c_a values, indicating a more liberal response criterion compared to distractor-paired words (M = 0.00 vs. 0.57, respectively), F(1,27) = 53.22, p < .001, η_p²= .66.

Discussion

In the DA condition, participants once again made more positive recognition responses to test items paired with targets than distractors, regardless of whether the test items were studied or new. This result can be understood as a change in bias rather than a change in sensitivity. Recognition performance was not significantly affected by targets and distractors in the FA condition. The novel result of Experiment 3 is that this pattern of results occurred despite equating the number of studied and new test words and ensuring that the ratio of targets and distractors was equal within each type of test word. The results of Experiment 3 therefore argue against the explanation that the target-related increase in positive responses seen in Experiments 1 and 2 reflected a strategy of making positive responses to infrequent targets and making negative responses to frequent distractors.

Experiment 4

In the previous experiments, participants responded to targets by saying the word “hit” out loud. Doing so may have led participants to respond positively to recognition test items because the word “hit” implies the successful achievement or production of something (e.g., she hit upon a solution; they hit the big time). This positive connotation of the word “hit” may have driven participants to respond to recognition test items in the positive sense. To check on this possibility, we changed the response to targets in Experiment 4 to the word “odd,” which does not have this positive connotation. If the effects reported in the previous experiments are due solely to the positive connotation of the word “hit,” then the target-related increase in positive recognition responses should be eliminated when another response word is used. If the recognition response shift occurs regardless of the response word, then that shift should emerge again here.

In addition, we dropped the FA condition in Experiment 4. The previous experiments have thrice shown that the mere presence of targets and distractors in the FA condition does not significantly affect recognition responses, so including the FA condition again would not likely provide any new information.

Method

Design and participants

The design was a 2 (test status: studied, new) X 2 (trial type: target, distractor) factorial with both factors manipulated within groups. An a priori power analysis using PANGEA indicated that a sample size of 34 would have power of .80 to detect a medium-sized effect of trial type in this design. We tested 36 undergraduate students from Whitman College in Experiment 3 in exchange for course credit or pay. Data from six additional participants were collected but were excluded from the analyses reported here because those participants did not meet our required minimum of responding to at least 70% of the recognition test items with either a yes or no response.

Materials and procedure

All materials were identical to those used in Experiment 3. The procedure was identical to the DA condition of Experiment 3 in every way except the test phase, in which participants responded with the word “odd” to each odd number target instead of “hit.”

Results

Encoding and secondary task performance

Participants identified 100% of the words at encoding. At retrieval, participants correctly identified odd numbers as targets 93.2% of the time (SD = 6.02) and incorrectly identified even numbers as targets 1.7% of the time (SD = 2.06).

Memory performance

Invalid trials occurred 10.38% of the time and were removed for conditionalized analyses. Those analyses led to identical conclusions as those derived from unconditionalized data. The following analyses that follow are based on the unconditionalized data.

The results from Experiment 4 are shown in Table 1. A 2 (test status: studied, new) X 2 (trial type: target, distractor) ANOVA on the proportion of positive responses in the recognition test yielded two significant main effects. Specifically, hits were significantly higher than FARs (M = 0.56 vs. 0.24, respectively), F(1,35) = 104.49, p < .001, η_p²= 0.75, and target-associated words were recognised significantly more than distractor-associated words (M = 0.50 vs. 0.29), F(1,35) = 45.00, p < .001, η_p²= 0.56. The interaction was not significant, F(1,35) = 1.50, p = .23, η_p²= 0.04.

Signal detection analyses

Figures 2 and 3 suggest that target processing had nonsignificant effects on discrimination but significantly affected bias. Specifically, d_a scores, representing discrimination, did not differ significantly across target and distractor conditions (M = 0.92 vs. 0.76, respectively), F(1,35) = 3.10, p = .09, η_p²= 0.08. In contrast, c_a scores, representing bias, were significantly lower for targets than for distractors (M = −0.02 vs. 0.60, respectively), indicating that target processing lowered the criterion for making positive recognition judgments, F(1,35) = 46.97, p < .001, η_p²= 0.57.

Discussion

The shift in positive recognition responses occurred when participants responded to targets using the word “odd” rather than “hit.” This conclusion converges with that of Huang and Meng (2020) who reported a target-related increase in hits and false alarms under DA when participants responded to targets using a keypress or by covertly counting targets. Taken together, the results of Huang and Meng (2020) and the current experiments suggest that the effect of retrieval-phase target detection on recognition performance does not depend on the specific type of response made to the target.

General discussion

In this study we asked whether retrieval-phase DA could enhance memory. At encoding, a growing body of research suggests that DA can do so in a phenomenon known as the attentional boost effect (Lin et al., 2010; Swallow & Jiang, 2010), which increases the ability to discriminate studied from new test items by increasing the hit rate for target-paired stimuli compared to distractor-paired stimuli. Here we found that target processing at retrieval did not enhance memory, but rather enhanced hits and false alarm rates to similar degrees, leading to no net difference in discrimination ability. This target-related enhancement of positive recognition responses occurred only under conditions of divided attention, not full attention (Experiments 1-3), but generalised across different distracting tasks (Experiment 2), different ratios of targets and distractors (Experiments 3 and 4), and different response types (Experiment 4). The effect can be understood as a change in bias in which participants adopt a more liberal criterion when making recognition decisions about target-paired words than when they make decisions about distractor-paired words. We discuss the theoretical implications of these results below.

In the introduction, we described a retrieval-enhancement hypothesis that was based on the idea that retrieval processes recapitulate encoding processes (Craik, 1983; Kolers, 1973; Morris et al., 1977; Tulving, 1983). That point of view implies that manipulations that enhance memory at the time of encoding will also do so when implemented at the time of retrieval. In previous studies of the ABE, target detection at encoding produces the standard ABE by increasing the hit rate for target-paired stimuli (e.g., Meng et al., 2019; Mulligan et al., 2014; Prull, 2019; Swallow & Jiang, 2010). The retrieval-enhancement hypothesis predicted that a similar outcome would occur following target detection at retrieval: a selective increase in the hit rate through an enhancement of retrieval processes in the DA condition. By this view, target detection would not affect responses to new words because, by definition, new words do not contain episodic information in memory whose retrieval can be enhanced. We did not find support for that hypothesis and our results are inconsistent with this theoretical perspective.

Our results appear to be more compatible with the idea that different forms of attention operate at encoding and retrieval and that a manipulation of attention that affects memory encoding does not necessarily produce the same result when the same manipulation occurs at the time of retrieval. The PRAM framework discussed in the introduction is representative of this perspective, in which perceptual/external attentional processes occur at encoding and reflective/internal attentional processes occur at retrieval. From this viewpoint, manipulations at encoding that affect later memory, such as divided attention, would not necessarily have similar effects when implemented at retrieval because attentional processes at retrieval differ from the attentional processing occurring at encoding. Therefore, one would expect to see asymmetries between encoding and retrieval with respect to divided attention. Those asymmetries exist in classic studies of the negative effects of DA on memory, with DA having a larger and more general effect when occurring at encoding and a smaller and more specific effect when occurring at retrieval (Baddeley et al., 1984; Craik et al., 1996; Naveh-Benjamin et al., 2000). The present experiments also hint at another asymmetry, this time with respect to the positive effects of target detection on memory. Specifically, whereas target detection during DA at encoding can enhance discrimination sensitivity between studied and new items, as is shown by the typical ABE, target detection during DA at retrieval lowers the response criterion, leading to an increase in hits and false alarms. Thus, the same manipulation of attention at encoding that enhances memory does not lead to a similar result when that manipulation occurs at retrieval.

Why would target detection increase hits and FARs in recognition judgments? Although the present experiments do not unequivocally suggest a mechanism, some clues may be found in studies of similar phenomena. For example, our results resemble the revelation effect, in which recognition test items are preceded by a problem to be solved, such as identifying a word as it is revealed letter-by-letter (e.g., _u_, _un, fun) or solving an anagram (e.g., ufn), before a recognition decision can be made. Hit and false alarm rates increase for test words that follow a problem-solving task compared to test words presented without such a task (e.g., Watkins & Peynircioğlu, 1990; Westerman & Greene, 1996; see Aßfalg et al., 2017 for a review). The revelation effect resembles a retrieval-phase DA experiment in some ways in that an additional task must be completed during the recognition test itself, however in revelation effect experiments the problem-solving task typically appears between recognition test items rather than occurring simultaneously with them. Therefore, attention may not be divided to the same degree as presenting secondary tasks simultaneously with recognition test items. Nevertheless, the impact of such problem-solving tasks on recognition performance resembles the results we report here.

Brandt et al. (2020) recently accounted for several result patterns within the revelation effect literature using MINERVA2, a global matching model of memory that assumes that all episodes in memory contribute to the outcome of a retrieval process (e.g., Hintzman, 1988). According to that approach, the problem-solving task in a revelation trial creates a temporary loss of contextual information in memory for studied items. This loss of contextual information in turn increases the contributions of pre-existing memory representations to the familiarity signal (echo intensity in MINERVA2) when the test item appears. The net effect is a flattening of the familiarity distributions of old and new items so that the proportion of each distribution that exceeds the response criterion increases (see Brandt et al., 2020 for details). The behavioural result is an increase in hits and FARs following problem-solving tasks. With respect to the target-related increases in hits and FARs that we have reported, an avenue for future research would be to determine whether target detection during a recognition test trial creates a similar temporary loss of contextual features. Target detection requires more cognitive resources than distractor rejection (Duncan, 1980), and this increased effort required for targets may contribute to a momentary loss of contextual information from working memory. In turn, the influence of pre-experimental memory traces on the familiarity signal may increase.

Another potential explanation involves the idea that subjective experiences can be used as sources of information in decision-making. For example, emotions, bodily sensations, and metacognitive experiences can influence judgments of risk, truth, and memory (e.g., Schwarz, 2012). In this vein, Dougal and Schooler (2007) reported that the feeling of solution discovery in problem-solving elevates the likelihood of responding “old” to recognition test items that immediately follow, compared to test items that follow unsolved problems. The interpretation of this phenomenon is that the feeling of discovery (the moment of insight, or “aha”) is used in a heuristic-like fashion to inform recognition. Solution discovery is misattributed to a feeling of recognition, thus elevating hits and FARs (see also Laukkonen et al., 2020). From this perspective, encountering a target during a recognition test may plausibly elicit a state of discovery that is misattributed to a feeling of recognition. Recent studies of the standard ABE, in which targets and distractors are presented at encoding rather than at retrieval, suggest that inferential processes can operate in a heuristic-like fashion at the time of test and can create response biases (Mulligan et al., 2022). When targets and distractors are presented at retrieval, a similar inferential, heuristic-like process may have been at work and may have created the response biases that we have reported.

How do our results fit with the dual-task interaction model, which explains the ABE as the result of a temporal selective attention process that is activated when a target is detected at encoding (Swallow & Jiang, 2013)? We suggest that temporal selective attention is activated by target detection at retrieval but its activation is irrelevant for the task at hand. Temporal selective attention is believed to increase perceptual processing (reviewed in Swallow et al., 2022), but standard recognition tests are driven primarily by semantic processes (e.g., Parks, 2013; Roediger, 1990). Therefore, the contribution of a perceptual process to aid recognition may be minimal. Nevertheless, enhanced perceptual processing could conceivably aid recognition. A potential future direction would be to explore the effect of retrieval-phase target processing on memory when the retrieval task involves perceptual discriminations, for instance by discriminating between two similar pictures of a landscape only one of which was studied, or by using a graphemic recognition test in which participants recognise words that have a similar appearance to studied words without recalling the words themselves (e.g., study freckle, test item: did you study a word that looked like fickle?; Challis et al., 1996). If temporal selective attention can improve access to perceptual representations in memory, then one may see memory enhancement following target detection in tests for which perceptual information is important. On balance, Makovski et al. (2011) examined this possibility in a working memory task for colour arrays in which test items were perceptually identical to studied items or differed by one colour. Participants made same/different judgments to the test items after detecting targets or after rejecting distractors. Target detection did not enhance accuracy in same/different judgments, a result that argues against the idea that target detection can improve perceptual memory. However, the small sample size raises concerns about the power of that experiment to detect significant differences that may exist. Additional studies investigating this possibility would therefore be informative.

In summary, encoding-phase division of attention that involves target detection in a secondary task can enhance memory in a phenomenon known as the attentional boost effect. When target detection occurs at the time of retrieval, no such boost to memory occurs. Instead, participants are more likely to make positive responses to all test items overall. Our results suggest another asymmetry between encoding and retrieval processes and mesh with theoretical views that emphasise different forms of attention operating at encoding and retrieval.

Acknowledgements

Portions of this research were presented at the annual meeting of the Psychonomics Society (November 2019, Montreal, Quebec). Data sets that were generated and analysed in this study are available from the first author upon request.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We left the response order unconstrained for target trials because, in most studies of the ABE in which targets are presented during the study phase, instructions do not specify the order of tasks to complete on target trials (i.e., whether to detect a target first then encode the associated item, or vice versa). For that reason, and to maximise comparisons with traditional ABE studies, the instructions did not specify an order of responses on target trials. Output interference was minimised by having participants respond in different modalities on target trials (i.e., vocally to the target and keypress to the word) and by giving participants practice trials before they began the test phase.

2 The main effect of trial type was significant in the analysis of conditionalized data (p = .04).

3 The attention effect on FARs reached statistical significance in the conditionalized analysis, p = .046.

4 One aspect of these results worth noting is that the DA-word condition did not impair recognition more than the DA-numbers condition. The DA-word condition should have had a larger negative effect according to the material-specific interference view described earlier (Fernandes & Moscovitch, 2000, 2002, 2003). We suggest two possible reasons for this null result. First, past studies that have reported material-specific interference have used free recall tests whereas the present experiment used a recognition test. Recognition is generally less sensitive to the impairing effects of DA compared to recall (Craik et al., 1996), so it is possible that we saw no difference between DA conditions due to the relative hardiness of recognition against DA. Second, in the only study of which we are aware that manipulated DA material types during a recognition test, Skinner and Fernandes (2008) found that material-specific interference was limited to “know” responses in a remember/know procedure. “Remember” responses, which index recollection, were impaired equally across DA conditions no matter the type of secondary task material. It is therefore possible that the absence of a material-specific DA effect in Experiment 2 was due to a high level of recollection, a process that seems to show only material-general effects of DA. A high level of recollection during the test phase is plausible considering that the study list was not particularly long (44 words), and the delay between study and test phases was short (2 min).

Appendix

Table A1. Experiment 2 recognition performance with Matched DA Groups (N = 28, 18, and 18 in FA, DA-Numbers, and DA-Words Conditions, respectively).

FA			DA-Numbers		DA-Words
Trial Type	Hit	FAR	Hit	FAR	Hit	FAR
Distractor	.51	.11	.34	.09	.41	.14
Target	.48	.12	.43	.22	.51	.28

Note. FA = full attention, DA = divided attention, FAR = false alarm rate.

Table A2. Experiment 2 signal detection measures with Matched DA Groups (N = 28, 18, and 18 in FA, DA-Numbers, and DA-Words Conditions, respectively).

FA			DA-Numbers		DA-Words
Trial Type	da	ca	da	ca	da	ca
Distractor	1.16	0.61	0.73	0.92	0.78	0.69
Target	1.04	0.63	0.56	0.53	0.66	0.34

Note. FA = full attention, DA = divided attention.