Divided Attention Selectively Impairs Value-Directed Encoding

The current study used a VDR recognition memory paradigm to examine the subjective states of awareness underpinning value-directed encoding processes. Moreover, we also examined the extent to which executive resources are required for the effect of value on recognition memory using various divided attention manipulations at encoding. In Experiment 1, we administered a value-based remember – know recognition memory test. Given previous literature, we expected the effect that value had on recognition memory (measured with hit rates) to be localized to remember responses, with no difference in know responses. Increased memory for higher-valued items could be due to increased rehearsal, or to the use of executive resources.

Therefore, in Experiment 2 we administered the same value-based remember – know recognition memory test. However, participants in this experiment completed either an articulatory suppression or random number generation divided attention task during encoding. If the effect of value on memory is ascribed to increased rehearsal of the higher-valued information, one would expect articulatory suppression to impair the effect of value. If executive resources are required for value-based encoding, then only the random number generation condition should impair the effect of value on recognition memory performance.

Because both our articulatory suppression and random number generation tasks used numbers as the spoken distractions, we replicated Experiment 2 using non-verbal divided attention tasks at encoding (sequential and random tone detection; Mangels, Picton, & Craik, 2001). Additionally, using this divided attention manipulation allowed us to more directly vary the executive demands of the secondary task while holding all other features constant. All together, these experiments were designed to understand the effect of value on recognition memory processes and how these processes are affected by dividing attention at encoded.

Forty participants were recruited from the Arizona State University research participation pool and took part in a study on recognition memory. All participants were compensated with course credit upon completion of the experiment. One participant failed to make any “remember” responses, and was excluded from the “remember” – “know” analyses.

Stimuli consisted of 200 nouns from the Toronto noun pool (Friendly, Franklin, Hoffman & Rubin, 1982). Participants completed multiple study-test blocks. The study phases consisted of 40 words each randomly assigned either a high (7, 9) or low (1, 3) point value (i.e., 20 high value and 20 low value, 10 of each specific value). The test phases consisted of 80 words (40 from the most recent list and 40 new nouns, randomly intermixed) presented one at a time without point values. Participants classified these old and new items at test along with making judgments on subjective states of recollection (i.e., “remember”) and familiarity (i.e., “know”). Each study- test phase was considered 1 block and 5 blocks were completed by each participant.

All experiments followed a study protocol that was approved by Arizona State University’s Institutional Review Board. Written informed consent was obtained from each participant before beginning the study. Participants were instructed that they would be completing multiple study-test blocks in a recognition memory task. The participants were told that their objective was to remember as many words as possible, with the goal of maximizing their score on each recognition memory test. They studied lists of 40 randomly selected nouns presented to them for a duration of 2s at a time, with a randomly jittered interstimulus interval from 300 to 500ms in 17 ms increments. The nouns were randomly paired with an integer indicating that it was either a low (1, 3) or high (7, 9) value item with a total of 20 low and 20 high values in each study block, ten of each value. Participants were told they would earn the point value previously paired with the word if correctly recognized, regardless of confidence rating, and that they would lose 1 point for incorrectly identifying a new word as old.

Each word appeared in the center of the computer screen below a central fixation cross, with the paired value appearing simultaneously above the central fixation cross in the center of the screen. After studying each list of 40 words, a test phase followed with 80 words (with no point values presented) that included all 40 of the words from the previous list and 40 new words randomly intermixed. Test words were presented one at a time and participants judged whether they thought the word was old or new. Participants responded using a standard computer keyboard with the following response option assignments: Z “Definitely New”—left pinky, X “Maybe New”—left ring finger, C “Maybe Know”—left middle finger, V “Definitely Know”—left index finger, and M “Remember”—right index finger. The participants were told to respond with their first instinct, but the test was untimed.

Before the experiment began, participants were briefed on the difference between remembering and knowing with the following instructions (adapted from Herzmann & Curran, 2011):

The effect of value on memory performance was analyzed using 3 paired-sample t-tests. Importantly, words were assigned value during encoding only and that means there is a common false alarm rate for both high and low value words (M = .23, SE = .02; analyses using a false-alarm correction procedure are reported in the supplemental material.). Memory performance was calculated as the mean proportion of old words successfully recognized regardless of dual- process judgment (total hits) and for remember and know (pooling over low and high confidence) judgments separately.

Memory performance as a function of value and judgment type is summarized in Table 1. 87.5% of participants in the sample (35/40) exhibited the typical VDR effect of better memory for higher- compared with lower-valued words in their overall hit rates. Accordingly, higher recognition accuracy was found for high- compared with low-value words in overall hit rates (t₍₃₉₎ = 5.163, p < 0.001, Cohen’s d = 0.90). When hit rates were conditionalized on the subjective state of awareness supporting those decisions, the value-driven gain in memory performance was due to an increase in “remember” responses (t₍₃₈₎ = 6.447, p < 0.001, Cohen’s d = 1.03). Importantly, no effect of value was observed for “know” responses (t₍₃₈₎ = –1.886, p = 0.067, Figure 1A).

These data show that value affects encoding processes leading to enhanced “remember” responses, supporting the executive encoding view of VDR effects. Therefore, in Experiment 2 we aimed to divide attention using either an executive-demanding secondary task (random number generation) or an articulatory-demanding task (articulatory suppression). We reasoned that dividing attention at encoding could be used to further dissociate encoding mechanisms that may be differentially affected by value in support of the executive view. Specifically, if executive processes are necessary for value-directed encoding then simultaneously completing the random number generation should attenuate the VDR effect, and this attenuation should be most prevalent in the “remember” responses.

To examine the underlying encoding mechanisms that value may affect to achieve greater rates of remember responses at test, in Experiment 2 we asked participants to perform either an articulatory suppression or random number generation task while studying the to-be remembered stimuli. Given articulatory suppression has been shown to disrupt articulatory rehearsal, while random number generation also taxes executive resources, we can use these manipulations to determine cognitive processes affected by value at encoding. If value at encoding leads to increased articulatory rehearsal, we would expect that articulatory suppression at encoding would reduce or eliminate the effect of value on later memory. To the extent that value at encoding affects executive resources, we would expect random number generation at encoding to reduce or eliminate the effect of value on memory. Besides the divided attention manipulations at encoding, this experiment was identical to Experiment 1.

Forty participants were recruited from the Arizona State University research participation pool and they took part in a study on recognition memory. All participants were compensated with course credit upon completion of the experiment.

The materials were identical to that of Experiment 1 with two exceptions. First, participants in Experiment 2 completed 4 study-test blocks instead of 5 study-test blocks. Second, during the encoding period participants completed either an articulatory suppression or random number generation divided attention task in each study-test block creating a 2 (Value: High versus Low) × 2 (Divided Attention: Random Number Generation versus Articulatory Suppression) fully within-subjects design.

Upon arrival, participants were instructed that they were to complete a memory task, but while studying the to-be remembered stimuli, they would also be completing one of two other secondary tasks. In the articulatory suppression condition, participants repeated the number “five” aloud on beat to a metronome at a pace of one response per second. In the random number generation condition, participants were asked to generate a random integer from one to nine aloud on beat to a metronome at a pace of one response per second. The participants were instructed that they should not use any patterns, and that their responses should be like “picking numbers out of a hat”. Participants were given one minute to practice each divided attention task before beginning the study. Otherwise, the memory task was identical to Experiment 1, except that four study-test blocks were completed (two with articulatory suppression at encoding, two with random number generation at encoding). The order in which participants completed each divided attention task at encoding was selected randomly.

The effects of value and divided attention on memory performance were analyzed using 3 repeated measures Analyses of Variance (ANOVAs) with the following within-subject factors, Value (High versus Low) and Divided Attention (Articulatory Suppression versus Random Number Generation). These analyses were conducted on the following dependent measures: proportion correctly recognized (hit rate), proportion given a “remember” response, and proportion given a “know” response. Importantly, words were assigned value during encoding only and that means there is a common false alarm rate for both high and low value words (M = .30, SE = .02). Analyses using a false-alarm correction procedure are reported in the supplemental material.

Memory performance as a function of value and judgment type is summarized in Table 1. A 2 (Value: High versus Low) × 2 (Divided Attention: Random Number Generation versus Articulatory Suppression) repeated measures ANOVA on hit rates revealed a main effect of value, F(1,39) = 12.96, p = .001, η²_partial = .249, a main effect of divided attention condition, F(1,39) = 42.54, p < .001, η²_partial = .522, and an interaction, F(1,39) = 6.81, p < .05, η²_partial = .149. Thus, we rejected the null hypothesis that the effect of value was similar across divided attention conditions. Post-hoc t-tests revealed that this interaction was driven by a difference between high and low value in the articulatory suppression condition (t₍₃₉₎ = 4.211, p < 0.001, Cohen’s d = 0.67) with no difference in the random number generation condition (t₍₃₉₎ = 1.158, p = .254; Figure 1B).

When hit rates were conditionalized on the subjective state of awareness supporting those decisions, we find similar results for “remember” responses. A 2 (Value: High versus Low) × 2 (Divided Attention: Random Number Generation versus Articulatory Suppression) repeated measures ANOVA on “remember” responses revealed a main effect of value, F(1,39) = 5.25, p < .05, η²_partial = .119, a main effect of divided attention condition, F(1,39) = 33.05, p < .001, η²_partial = .459, and an interaction, F(1,39) = 6.50, p < .05, η²_partial = .143. Post-hoc t-tests revealed that this interaction was driven by a difference between high and low value in the articulatory suppression condition (t₍₃₉₎ = 2.794, p < 0.01, Cohen’s d = 0.45) with no difference in the random number generation condition (t₍₃₉₎ = .361, p = .720).

A 2 (Value: High versus Low) × 2 (Divided Attention: Random Number Generation versus Articulatory Suppression) repeated measures ANOVA on “know” responses revealed no main effect of value, F(1,39) = 2.06, p = .159, no main effect of divided attention condition, F(1,39) = .055, p = .816, and no interaction, F(1,39) = .054, p = .817. Together, these data again show that value affects encoding processes leading to enhanced states of subjective recollection. In addition, random number generation at encoding eliminated this effect, while articulatory suppression at encoding did not. This result supports the view that executive resources are necessary for value-directed encoding effects on memory. One alternative explanation with the current experiment is that dividing attention via verbally spoken integers could have impaired encoding the value of the stimuli due to the conceptual overlap of the integers used to represent value, rather than disrupting executive resources. To this end, Experiment 3 was conducted using a non-verbal tone detection divided attention manipulation that lacked the conceptual overlap of integer values.

In Experiment 3 we asked participants to perform either an easy or difficult secondary task while studying the to-be remembered stimuli. The logic here was to use divided attention tasks that placed relative demands on executive processes but no obvious demands on articulatory processes. To the extent that value-based encoding places demands on executive resources, we would expect the more difficult secondary task to reduce or eliminate the effect of value on recollection. Besides the change in divided attention tasks at encoding, this experiment was identical to Experiment 2.

Forty participants were recruited from the Arizona State University research participation pool and they took part in a study on recognition memory. All participants were compensated with course credit upon completion of the experiment.

The materials were identical to that of Experiment 2 except instead of articulatory suppression and random number generation during encoding, participants completed either an easy or difficult tone detection secondary task in each study-test block creating a 2 (Value: High versus Low) × 2 (Divided Attention: easy tone detection versus difficult tone detection) fully within-subjects design. The tone detection task was adapted from Mangels et al. (2001). The stimuli were low (250 Hz), medium (750 Hz) and high (2250 Hz) tones played over standard computer speakers. The 750 Hz was presented at 55 dB sound pressure level with the other tones being adjusted to be equally loud. All the tones were created using Audacity.

Upon arrival, participants were instructed that they were to complete a memory task, but while studying the to-be remembered stimuli, they would also be completing one of two other ongoing tasks. In both secondary tasks, participants were instructed that they would need to recognize low, medium, and high tones by pressing the corresponding key (1, 2, and 3 on a standard keyboard number pad). The tones were presented at a pace of one per second. In the easy tone detection condition, tones occurred in a fixed order of low, medium, high. In the difficult tone detection condition, tones occurred in a random order. Otherwise, the memory task was identical to Experiment 2. Participants were given one minute to practice each divided attention task before beginning the study. The order in which participants completed each divided attention task at encoding was selected randomly.

The effects of value and divided attention on memory performance were analyzed using 3 repeated measures Analyses of Variance (ANOVAs) with the following within-subject factors, Value (High versus Low) and Divided Attention (Easy versus Difficult). These analyses were conducted on the following dependent measures: proportion correctly recognized (hit rate), proportion given a remember response, and proportion given a know response. Importantly, words were assigned value during encoding only and that means there is a common false alarm rate for both high and low value words (M = .29, SE = .02).

Memory performance as a function of value and judgment type is summarized in Table 1. A 2 (Value: High versus Low) × 2 (Divided Attention: easy tone detection versus difficult tone detection) repeated measures ANOVA on hit rates revealed a main effect of value, F(1,39) = 24.05, p < .001, η²_partial = .381, a main effect of divided attention condition, F(1,39) = 19.00, p < .001, η²_partial = .328, and an interaction, F(1,39) = 9.64, p < .01, η²_partial = .198. Thus, we rejected the null hypothesis that the effect of value was similar across divided attention conditions. Post-hoc t-tests revealed that this interaction was driven by a difference between high and low value in the easy tone detection condition (t₍₃₉₎ = 5.136, p < 0.001, Cohen’s d = 0.85) with a markedly smaller difference in the difficult tone detection condition (t₍₃₉₎ = 2.188, p = .035, Cohen’s d = 0.36, Figure 1C).

When hit rates were conditionalized on the subjective state of awareness supporting those decisions, we find similar results for “remember” responses. A 2 (Value: High versus Low) × 2 (Divided Attention: easy tone detection versus difficult tone detection) repeated measures ANOVA on “remember” responses revealed a main effect of value, F(1,39) = 12.35, p = .001, η²_partial = .241, a main effect of divided attention condition, F(1,39) = 13.60, p = .001, η²_partial = 259, and an interaction, F(1,39) = 6.35, p < .05, η²_partial = .140. Post-hoc t-tests revealed that this interaction was driven by a difference between high and low value in the easy tone detection condition (t₍₃₉₎ = 3.583, p = 0.001, Cohen’s d = 0.63) with no difference in the difficult tone detection condition (t₍₃₉₎ = 1.764, p = .085).

A 2 (Value: High versus Low) × 2 (Divided Attention: easy tone detection versus difficult tone detection) repeated measures ANOVA on “know” responses revealed no main effect of value, F(1,39) = 2.22, p = .145, no main effect of divided attention condition, F(1,39) = .004, p = .949, and no interaction, F(1,39) = .464, p = .500. Together, these data again show that value affects encoding processes leading to enhanced states of subjective recollection. In addition, a difficult secondary task at encoding eliminated this effect, while an easy secondary task at encoding did not. This pattern of results replicates Experiments 1 and 2 while further supporting the view that executive resources are necessary for value-directed encoding effects on memory.

Across the current study, three primary findings emerged. First, the effects of value-based encoding consistently increased recognition memory performance for high compared with low- value information. Second, this improvement in recognition memory for high-value information was consistently localized to enhancement of subjective states of “remembering” (i.e., recollection) with no significant changes in subjective states of “knowing” (i.e., familiarity). Third, difficult divided attention tasks (random number generation and random tone detection) administered during value-based encoding mitigated these effects. However, easy divided attention tasks (articulatory suppression and sequential tone detection) had relatively little impact on the enhancement in remembering high-value information.

These VDR effects in recognition memory contribute to a growing body of literature examining how value-based encoding positively influences subsequent memory in a broad range of tasks including delayed free recall (Castel et al., 2002; Castel et al., 2008; Stefanidi et al., 2018), cued recall (Wolosin et al., 2012) recognition memory (Adcock et al., 2006; Hennessee et al., 2017; Cohen et al., 2017; Gruber & Otten, 2010; Gruber, Ritchey, Wang, Doss, & Ranganath, 2016), and source memory (Shigemune, Tsukiura, Kambara, & Kawashima, 2014). Across these studies, value-directed encoding leads to improved memory for high-value information. The current study aimed to evaluate two aspects of the nature of this improvement by evaluating subjective states of awareness accompanying recognition memory decisions and dividing attention while participants encoded high versus low-value information.

The current results are broadly consistent with previous research indicating that the effect of value at encoding selectively increases strong memories (whether they be recollective or simply highest memory strength). Replicating prior studies (Gruber & Otten, 2010; Gruber, et al., 2016; Hennessee et al., 2017; Cohen et al., 2017), value-directed encoding in the current study consistently led to more recollective experiences being reported at test with little influence on familiarity-based experiences. Recollection is generally influenced by depth of processing manipulations (deep > shallow), generation effects (generation > reading), pharmacological intervention (benzodiazepine > placebo), and attention at encoding (full > divided). VDR effects in recognition memory contribute to this growing list of manipulations that selectively enhance recollection (or remember responses in a remember-know paradigm).

However, it is still inconclusive whether two distinct processes contribute to recognition memory decisions, or a single-process based on memory strength. Given prior research showing VDR effects localized to strong/recollective memories, this data may prove useful for testing between contemporary models of recognition memory. Specifically, it is an interesting question how single process models of recognition memory can capture the effects found in prior research and the current research.

In Experiments 2 and 3, we aimed to evaluate the role of attention in encoding high-value information. In Experiment 2 we found that a divided attention task intended to block articulatory processes had relatively smaller effects on VDR than a divided attention task intended to block executive processes. Given prior work examining the role of executive processes in establishing more recollective experiences (Unsworth & Brewer, 2009), we predicted that random number generation would attenuate or even eliminate the VDR effect in recollection, which it did. In addition, we failed to find support in Experiment 2 that articulatory rehearsal processes were influenced by value-directed encoding. It is important to note, however, that the articulatory suppression condition in the current study was less difficult than the random number generation condition. Therefore, the interpretation that the effect of value on memory relies more on executive functions than articulatory process could be confounded by task difficulty. Future studies should address this concern.

In Experiment 3 we conceptually replicated Experiment 2 by examining how more difficult executive tasks attenuate value-based encoding effects in recollection relative to less difficult executive tasks. We found that a more difficult executive demanding task attenuated the effect of value on memory, while a less difficult task did not. Together, the data suggest that executive resources are necessary for the effect of value on memory. Executive functions are thought to arise from a network of regions in the brain that include the dorsolateral prefrontal cortex (dlPFC; Kane & Engle, 2002). Similarly, neuroimaging and neurostimulation studies have shown that the dlPFC is necessary for random number generation (Jahanshahi & Dirnberger, 1999; Jahanshahi, Dirnberger, Fuller, & Frith. 2000). Therefore, given evidence from the current study that executive resources are necessary for the effect of value on memory, the current study may provide indirect evidence for the role of the dlPFC in value-directed encoding.

Although this study revealed the importance of executive resources for value at encoding to improve “remember” responses, it does not provide direct evidence for the specific neural or cognitive processes utilized when encoding higher-valued information. Previous literature has revealed at least two possible systems that are activated during value-directed memory encoding, the mesolimbic reward system (via dopaminergic midbrain-hippocampal projections, Lisman & Grace, 2005; Shohamy & Adcock, 2010) and fronto-temporal strategy driven processes (Cohen, Rissman, Suthana, Castel, & Knowlton, 2014; Cohen, Rissman, Suthana, Castel, & Knowlton, 2016). In addition, higher-valued information has been shown to affect selective attention and working memory processes (Anderson, 2013; Ariel & Castel, 2014; Krawczyk, Gazzaley, & D’Esposito, 2007), which could also have an effect on later memory (Blumenfeld & Ranganath, 2007). Importantly, the reward system and selective strategy use at encoding typically lead to different effects on later states of awareness, with the former leading to enhanced states of recollection alone and the latter to enhanced states of recollection and familiarity (Cohen et al., 2017).

Our data clearly showed an effect of value on “remember” responses alone across all three experiments. According to previous literature, this result would be consistent with memory enhancement via the mesolimbic dopamine system. However, across Experiments 2 and 3 we found that disrupting executive resources mitigated the effect of value on recollection. One way to interpret this finding is that disrupting executive resources prevented the use of engaging deep semantic strategies, and that this lead to the modulation of value-driven enhancement of recollection. Under this interpretation one would expect that we would have found value related improvements to both remember and know responses, which we did not. Another interpretation is the more difficult divided conditions had an effect on executive resources that may be necessary for activation of phasic dopaminergic responses. There is evidence that the prefrontal cortex (an underlying structure crucial for executive resources) drives the midbrain to initiate motivated behavior (Ballard et al., 2011; Murty, Ballard, & Adcock, 2016). It is not possible from this study to dissociate if one or both these systems are responsible for the benefit of value at encoding to later memory, as disrupting executive resources could prevent selective attention, PFC driven midbrain activation, and later metacognitive strategy selection. Further, individual differences likely exist amongst the use of these systems. Future studies should address these concerns.

Contrary to our results, a study conducted by Middlebrooks, Kerr, and Castel (2017) investigating the effects of value on delayed free recall under various divided attention conditions did not find divided attention to influence the effect of value on memory performance. Although conceptually similar, the current study utilized a recognition memory paradigm, while Middlebrooks and colleagues utilized a free recall paradigm. The contrasting results may indicate different encoding processes or strategies used to selectively remember higher-valued stimuli in free recall versus recognition memory paradigms. Future studies should address this contradiction.

It is well observed that capacity limits of our memory systems necessitate the attenuation of most of the information we encounter daily. Given these capacity limits, important or highly- valued information must be selected and prioritized over less important information, although the underlying mechanisms remain unclear. Across three experiments we demonstrate that value at encoding enhances memory for higher-valued items. Specifically, this effect is localized to subjective states of recollection with no effect on familiarity. Additionally, disrupting executive resources by dividing attention at encoding eliminated the effect of value on recollection, while disrupting articulatory rehearsal processes at encoding had little effect. Our results suggest that executive resources at encoding are necessary for the prioritization and selective encoding of important information, expanding our knowledge of mechanisms that influence selection and prioritization of items for later remembering. These results dovetail with recent research aimed at clarifying the relation between motivational and cognitive processes at play when people study and recognize important information.

Data can be found on OSF at osf.io/3dh5w.

We would like to extend a special acknowledgement to Sam McClure and Justin Fine for their helpful feedback on previous drafts of this manuscript.

GB was supported by the Army Research Institute during the completion of this work (grant number W911NF-17-1-0175).

The authors have no competing interests to declare.

B.L. Elliott and G.A. Brewer developed the study concept and implemented the design. B.L. Elliott analyzed, interpreted the data, and drafted the manuscript. Both authors approved the final version of the manuscript for submission.

Portions of this research were presented in Blake Elliott’s Masters Thesis at Arizona State University.

The author(s) of this paper chose the Open Review option, and the peer review comments are available at: http://doi.org/10.1525/collabra.156.pr