Value-Directed Retrieval: The Effects of Divided Attention at Encoding and Retrieval on Memory Selectivity and Retrieval Dynamics

Abstract

Value-directed remembering refers to the tendency to best remember important information at the expense of less valuable information, and this ability may draw on strategic attentional processes. In six experiments, we investigated the role of attention in value-directed remembering by examining memory for important information under conditions of divided attention during encoding and retrieval. We presented participants with lists of words of varying objective or subjective value and compared participants completing the study phase under full or divided attention, in addition to participants completing the testing phase under full or divided attention. Results revealed that certain forms of selectivity were impaired when attention was divided during encoding but not when attention was divided during retrieval. Participants initiated recall (i.e., probability of first recall [PFR]) with high-value words as well as with words they subjectively deemed important; these value-mediated PFR retrieval dynamics resisted influence from reduced attentional resources during encoding and retrieval. Thus, while value-directed remembering involves both strategic encoding and retrieval operations, attentional resources during encoding appear crucial for subsequent recollection of valuable and important information; however, attentional resources during retrieval may be less influential in strategic selective memory.

Full text

Journal of Experimental Psychology: Learning,
Memory, and Cognition
Value-Directed Retrieval: The Effects of Divided Attention at Encoding and
Retrieval on Memory Selectivity and Retrieval Dynamics
Dillon H. Murphy, Shawn T. Schwartz, and Alan D. Castel
Online First Publication, June 15, 2023. https://dx.doi.org/10.1037/xlm0001264
CITATION
Murphy, D. H., Schwartz, S. T., & Castel, A. D. (2023, June 15). Value-Directed Retrieval: The Effects of Divided Attention
at Encoding and Retrieval on Memory Selectivity and Retrieval Dynamics. Journal of Experimental Psychology: Learning,
Memory, and Cognition. Advance online publication. https://dx.doi.org/10.1037/xlm0001264

Value-Directed Retrieval: The Effects of Divided Attention at Encoding
and Retrieval on Memory Selectivity and Retrieval Dynamics
Dillon H. Murphy, Shawn T. Schwartz, and Alan D. Castel
Department of Psychology, University of California, Los Angeles
Value-directed remembering refers to the tendency to best remember important information at the expense of
less valuable information, and this ability may draw on strategic attentional processes. In six experiments, we
investigated the role of attention in value-directed remembering by examining memory for important infor-
mation under conditions of divided attention during encoding and retrieval. We presented participants with
lists of words of varying objective or subjective value and compared participants completing the study phase
under full or divided attention, in addition to participants completing the testing phase under full or divided
attention. Results revealed that certain forms of selectivity were impaired when attention was divided during
encoding but not when attention was divided during retrieval. Participants initiated recall (i.e., probability of
first recall [PFR]) with high-value words as well as with words they subjectively deemed important; these
value-mediated PFR retrieval dynamics resisted influence from reduced attentional resources during encod-
ing and retrieval. Thus, while value-directed remembering involves both strategic encoding and retrieval
operations, attentional resources during encoding appear crucial for subsequent recollection of valuable
and important information; however, attentional resources during retrievalmay be less influential in strategic
selective memory.
Keywords: selectivity, attention, retrieval, lag-recency effects, value-directed remembering
People tend to focus on valuable information at the expense of less
important information when presented with an abundance of infor-
mation to remember (Ariel et al., 2009;Castel, 2008;Castel et al.,
2002,2007,2013;Elliott et al., 2019;McGillivray & Castel,
2017;Schwartz et al., 2020,2023;Siegel & Castel, 2018a,2018b;
Siegel et al., 2021;Soderstrom & McCabe, 2011; see also
Knowlton & Castel, 2022;Madan, 2017 for review). This ability
is captured by value-directed remembering paradigms whereby
learners employ both bottom-up/automatic and top-down/strategic
value-based selectivity processes. In these value-directed remember-
ing tasks, to-be-remembered information is paired with point values
which count toward a participant’s score if correctly recalled. While
selective memory for high-value information is often attributed to
strategic attention in the encoding phase (e.g., Ariel et al., 2009,
2015;Castel, 2008;McGillivray & Castel, 2011;Schwartz et al.,
2020,2023), recent work indicates that retrieval dynamics may
also play a critical role in memory selectivity (Murphy & Castel,
2022a;Stefanidi et al., 2018).
When participants retrieve items from long-term memory, many
systematic trends are observed in their recall (Rohrer & Wixted,
1994). For example, the probability of first recall (PFR) examines
the probability that items from each serial position will be recalled
first, typically revealing that participants are most likely to initiate
recall with either the first or last presented item (Howard &
Kahana, 1999;Kahana et al., 2002). In addition to elevated PFR
for primacy and recency items, participants are most likely to initiate
recall with the most valuable items when words are paired with point
values (Murphy & Castel, 2022a;Murphy et al., 2022;Stefanidi et
al., 2018). Thus, there are strategic retrieval operations that contrib-
ute to memory selectivity.
Following the initiation of retrieval, more systematic tendencies
manifest in participants’output, particularly their recall transitions.
For example, items studied in close temporal proximity also tend to
be recalled in close proximity and in the same order with which
they were studied. This property is captured by lag conditional-
response probabilities (lag-CRPs; Kahana, 1996;seeHintzman,
2015 for a critique, but see Healey et al., 2019 for a response),
which illustrate the lag-recency effect: The use of just-recalled items
to assist the recall of words presented close together in the study
phase via the utilization of shared contextual features (Sederberg
et al., 2010;Spillers & Unsworth, 2011). Prior research indicates
that participants are more likely to retrieve adjacent items compared
Dillon H. Murphy https://orcid.org/0000-0002-5604-3494
Shawn T. Schwartz https://orcid.org/0000-0001-6444-8451
We would liketo thank Karina Agadzhanyan and Delaney Falsken for their
assistance with managing data collection. Additionally, we thank Drew
Murphy for assistance coding the data.
This research was supported in part by the National Institutes of Health
(National Institute on Aging; Award Number R01 AG044335 to Alan
D. Castel).
The authors certify that they have no affiliations with or involvement in any
organization or entity with any financial or nonfinancial interest in the subject
matter or materials discussed in this manuscript.
The experiments reported in this article were not formally preregistered, but
the stimuli and data have been made available on the Open Science Framework
at https://osf.io/hu9xk/?view_only=21f9dff5c28a4406bdf0f369fbb6b7f2.
Correspondence concerning this article should be addressed to
Dillon H. Murphy, Department of Psychology, University of California,
6526 Pritzker Hall, 502 Portola Plaza, Los Angeles, CA 90095, United
States. Email: dmurphy8@ucla.edu
Journal of Experimental Psychology:
Learning, Memory, and Cognition
© 2023 American Psychological Association
ISSN: 0278-7393 https://doi.org/10.1037/xlm0001264
1
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

to more distant items and that CRPs are greater in the forward direc-
tion compared with the backward direction (Kahana, 1996).
Previous work using value-directed remembering paradigms sug-
gests that CRPs are similar in learners completing a value-directed
remembering procedure compared with controls where words
are not paired with point values (Stefanidi et al., 2018;seealso
Murphy & Castel, 2022a;Murphy et al., 2022). However, CRPs
may be related to selective memory such that processes that disrupt
the lag-recency effect may subsequently impact the recall of valuable
information. For example, the temporal organization of memory is
often related to strategic encoding and retrieval operations (e.g.,
Unsworth, 2016;Unsworth et al., 2019;Unsworth & Spillers,
2010), potentially implicating the lag-recency effect as a contributor
to selective memory. Moreover, the reinstatement of temporal–con-
textual information during retrieval may also reinstate metacognitive
or strategic processes that contribute to selective memory (see
Murphy et al., 2021). Thus, combined with evidence that learners ini-
tiate retrieval with valuable information (Murphy & Castel, 2022a;
Murphy et al., 2022;Stefanidi et al., 2018), there may be important
processes occurring during recall that contribute to selective memory,
and it is possible that a reduction in CRPs corresponds to a reduction
in selective memory.
In addition to items presented closer together in time being recalled
in close proximity, items that are semantically related tend to be
recalled together (Bousfield et al., 1954;Healey & Kahana, 2014;
Howard & Kahana, 2002;Romney et al., 1993). Schematic support
(as described by Craik & Bosman, 1992) occurs when prior knowl-
edge in a domain can facilitate memory for other information in that
domain. When presented with semantically related information, learn-
ers tend to focus on goal-relevant, important information (e.g.,
Murphy & Castel, 2021a,2021b,2022b;Murphy et al., 2023 but it
remains unclear how participants organize retrieval of semantically
related words according to subjective value.
Selective memory processes are driven by directing attentional
resources toward valuable information during encoding, subsequently
maximizing the likelihood that thisto-be-learned information—which
is paired with temporally proximate, associated value information—
will be later retrieved (Ariel et al., 2009;Castel et al., 2012).
Despite the apparent role that selective attention mechanisms play
in the successful encoding and retrieval of high-value information,
prior work has demonstrated that under certain circumstances, partic-
ipants can still be selective for valuable information even when atten-
tion at encoding is divided (e.g., Middlebrooks et al., 2017;Siegel &
Castel, 2018b). However, other work using various divided attention
tasks (e.g., articulatory suppression, random number generation, tone
detection) has demonstrated that some divided attention tasks (random
number generation, a difficult tone detection task) impair selective
memory while selectivity is preserved in other divided attention
tasks (articulatory suppression, an easy tone detection task; Elliott
& Brewer, 2019;seealsoMurphy & Castel, 2022c for instances
where divided attention at encoding impaired memory selectivity).
Thus, the effects of divided attention at encoding onmemory selectiv-
ity may depend on the degree to which attention is divided.
Memory selectivity can also be attenuated when concurrent pri-
mary and secondary tasks rely on overlapping processing resources
during encoding (e.g., a visual–spatial primary memory task paired
with a concurrent visual–spatial secondary pattern discrimination
task). For example, Siegel et al. (2021) investigated memory selectiv-
ity for valuable visuospatial information arbitrarily assigned along a
gradient of least to most important. Selectivity was preserved when
spatial and nonspatial auditory secondary distractor tasks, in addition
to a nonspatial visual secondary distractor task, were present during
encoding for the primary visual–spatial task. However, when the sec-
ondary distractor task was of the same domain (visual) and modality
(spatial), participants were no longer selective toward the more valu-
able visual–spatial information, thus diminishing selectivity effects.
Despite these findings, the current state of this intra- versus intermo-
dality research is not necessarily generalizable to other domains (e.g.,
auditory, verbal) due to a lackof empirical investigation outside of the
visuospatial domain. Furthermore, it is unclear what underlying
mechanism drives selectivity differences when concurrent memory
tasks rely on both encoding and retrieving competing information
streams within domains but across modalities.
Similar to when attention is divided at encoding, prior work has
shown that there are greater costs of divided attention at retrieval if
the primary and secondary tasks overlap (Fernandes & Moscovitch,
2000,2002,2003;Skinner & Fernandes, 2008). Yet, in terms of the
number of items able to be recalled, early work showed larger effects
of divided attention at encoding than at retrieval (Craik et al.,
1996;Naveh-Benjamin, Craik, Gavrilescu, & Anderson, 2000;
Naveh-Benjamin, Craik, Perretta, & Tonev, 2000;Naveh-Benjamin
et al., 1998;seealsoRohrer & Pashler, 2003). However, despite not
necessarily causing overall recall deficits, divided attention during
retrieval may impair the selective retrieval of more important informa-
tion. For example, the tendency to initiate recall with valuable or
important items and selectivity for this information may be reduced
with fewer available resources at retrieval. Thus, a full allotment of
attentional resources during both encoding and retrieval may be a crit-
ical component of engaging in the effortful processing of important
information (see Murphy & Castel, 2022c) such that people struggle
to selectively encode high-value items or engage in strategic retrieval
operations while completing a secondary task.
A learner’s goals and metacognitive strategies to achieve those
goals likely lead to focusing on important or valuable information,
and we were interested in how these processes are affected by
divided attention at encoding and retrieval. Specifically, participants
may employ strategic encoding operations, such as engaging in more
effective encoding strategies for high-value items (Hennessee et al.,
2019), and/or strategic retrieval operations like initiating recall with
high-value items or recalling important items before low-value items
to reduce potential output interference (Murphy & Castel, 2022a).
Thus, there may be important differences in how divided attention
at encoding and retrieval impacts selective memory.
The Current Study
In the current study, we used attentional manipulations to (a) elu-
cidate how divided attention at encoding alters previously described
effects of value on retrieval dynamics, (b) determine how divided
attention at retrieval impacts the dynamics of free recall and memory
for valuable information, and (c) examine how divided attention at
encoding and retrieval affects the tendency to remember subjectively
important information that could have consequences for forgetting
in a more applied context. In each experiment, we presented partic-
ipants with information of varying objective or subjective values to
remember for a later test. In Experiment 1, we compared participants
completing the study phase under full or divided attention while in
Experiment 2, we compared participants completing the testing
MURPHY, SCHWARTZ, AND CASTEL2
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

phase under full or divided attention. In Experiment 3, we directly
compared participants completing the study and retrieval phase
under divided attention using the same divided attention task.
While participants under divided attention at encoding may show
similar selectivity and retrieval tendencies (i.e., PFR, lag-CRPs) as
participants with full attention (as shown in Middlebrooks et al.,
2017;Murphy & Castel, 2022a), participants under divided attention
at retrieval may show reduced selectivity and PFR for highly valued
items. Alternatively, prior work suggests that dividing attention during
learning may hinder the attentional resources available for effortful
item-value encoding (see Yeung & Fernandes, 2021), which may be
more detrimental to subsequent value-directed memory performance
than dividing attention at retrieval. Furthermore, participants under
divided attentionat encoding could show detriments in memoryselec-
tivity not observed in participants under divided attention at retrieval
due to the secondary task demands detracting resources from the
effortful associative binding of memoranda and their relative value
while retrieval attempts under divided attention may be subjected to
more task-based competition and/or interference (see Fernandes &
Moscovitch, 2000,2002,2003;Skinner & Fernandes, 2008).
Potential differences in memory selectivity and retrieval tendencies
when attention is divided at encoding and retrieval may provide evi-
dence that value-directed remembering involves not simply the strate-
gic allocation of attention during encoding, but also strategic retrieval
operations which include recalling the most important or valuable
items before less important ones. Specifically, selectivity may depend
on strategic retrieval operations, such as initiating recall with high-
value or important items (as measured by PFR), leading to decreased
selectivity if attention is divided during recall. As such, dividing par-
ticipants’attention at encoding and retrieval may reveal potential
underlying behavioral mechanisms contributing to value-directed
remembering and may provide a more comprehensive and theoretical
approach to understanding how divided attention can influence
retrieval dynamics in a value-directed remembering context.
Lastly, the value of information can be objective (experimenter-
designated point values) or subjective (intrinsic importance), and
these different value assessments could differentially impact memory
processes. Specifically, objective point values provide a clear hierar-
chy of which information should be prioritized while determining
subjective value requires learners to consider the benefits of remem-
bering and/or the consequences of forgetting a given item. As such,
strategically remembering objectively and subjectively valuable infor-
mation may involve different memory processes that could be differ-
entially impacted by divided attention at encoding and retrieval.
Experiment 1a
In Experiment 1a, we presented participants with lists of words
paired with point values counting toward participants’scores if the
word was correctly recalled. Participants either completed the study
phase under full or divided attention; participants under divided atten-
tion simultaneously completed a digit detection task while studying
the words. All participants completed the test phase with full attention.
Method
Participants
In each experiment, participants were undergraduate students
recruited from the University of California, Los Angeles (UCLA)
Human Subjects Pool. Participants were tested online and received
course credit for their participation. Participants were excluded from
analysis if they admitted to cheating (e.g., writing down answers) in
a posttask questionnaire (they were told they would still receive credit
if they cheated). After one exclusion for cheating, our sample included
106 participants (M
age
=20.40, SD
age
=1.29). On the divided atten-
tion task (details below), participants correctly identified an average
of 2.41 out of eight sequences (SD =1.07)oneachlist.Therewas
an average of 1.69 incorrect detections (SD =1.30) on each list
whereby participants pressed the space bar to indicate that three odd
digits had been played when they had not. Ifa participant failed to iden-
tify at least one sequence (correctly or incorrectly) during a list, their
data for that list was excluded (as was the case in Middlebrooks et
al., 2017). This exclusion process resulted in 33 lists being excluded
from analysis (out of 330 total lists). A sensitivity analysis based on
the observed sample indicated that for a between-subjects analysis of
variance (ANOVA) with two groups (attention: full, divided)andsix
measurements (list: 1, 2, …,6), assuming α=.05, power =.80, and
an average correlation of r=.30 between repeated-measures (selectiv-
ity), the smallest effect the design could reliably detect is η
p
2
=.03.
Materials and Procedure
Participants were presented with a series of to-be-remembered
words with each word paired with a unique, randomly assigned
value between 1 and 20 indicating how much the word was
“worth.”Word–value pairs were separated by a colon with the
value presented to the right of the word (e.g., twig: 5). Both words
and point values were simultaneously displayed for 3 s each and in
the same font. Each point value was used only once within each list
and the order of the point values within lists was randomized. The
stimulus words were nouns that contained between four and seven let-
ters with an everyday occurrence rate of at least 30 times per million
(Thorndike & Lorge, 1944). Participants were told that their score
would be the sum of the associated values of the words that they
recalled and that they should try to maximize their score.
After the presentation of all 20 word–number pairs in each list,
participants were given an immediate, 1-min free recall test in
which they had to type as many words as they could from the list
(they did not need to recall the point values) into an on-screen text
box. To account for typographical errors in participants’responses,
we employed a real-time textual similarity algorithm where
responses with at least 75% similarity to the correct answer were
counted as accurate. Immediately following the recall period, partic-
ipants were informed of their total score for that list but were not
given feedback about the items they recalled (or failed to recall).
This was repeated for a total of six study-test cycles and participants
self-paced their breaks between lists.
In addition to their overall score, participants were scored for effi-
ciency via a selectivity index. For this metric, we calculated each
participant’s recall score relative to their chance and ideal score.
The ideal score was comprised of the sum of only the highest values
for the number of words recalled. For example, if a participant only
remembered three words, then ideally those words would be paired
with the three highest values (e.g., 18 + 19 + 20 =57). Chance
scores reflected no attention to valueand were calculated as the prod-
uct of the average point value and the number of recalled words. At
chance, the score in our example would result in 10.5 (the average
value of the points in the list) multiplied by the number of recalled
VALUE-DIRECTED RETRIEVAL 3
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

words. If a participant only recalled words paired with the highest
values, the resulting selectivity score would be 1 while a participant
who only recalled words paired with the lowest values would receive
a selectivity score of −1. Scores close to 0 indicate that a subject was
not selective (see Castel et al., 2002 for more details).
Participants were randomly assigned to either complete the
task with full attention (n=51) or divided attention (n=55).
Participants in the divided attention condition studied the
to-be-remembered items while completing a digit-detection task.
These participants were told that they would hear a series of digits
spoken aloud while they studied the words and that they should
press the spacebar on the keyboard every time they heard a sequence
of three odd digits in a row. One digit (numbers 1–9) was read per
second and the digits were randomly generated. For each participant
on each list, there were eight instances of three-odd-digit sequences
per list (when the spacebar should be pressed), and there was never a
sequence of four odd digits in a row.
Results
Analysis Plan
Sample size, mean, variance, skew, and kurtosis for performance
on the divided attention task, recall, and selectivity in each experi-
ment are shown in the Table A1. Linear regressions with average
divided attention task performance predicting average recall and
selectivity in each experiment are shown in Table 1. To examine
group differences in recall sensitivity for valuable information, we
conducted 2 (attention at encoding: full, divided)×6 (list: 1, 2, …,
6) mixed ANOVA on selectivity index scores. To examine recall per-
formance and further examine selectivity for valuable information,
we computed multilevel models (MLMs) where we treated the
data as hierarchical or clustered (i.e., multilevel) with items nested
within individual participants; observations were not nested within
items. Since recall at the item level was binary (correct or incorrect),
we conducted logistic MLMs. In these analyses, the regression coef-
ficients are given as logit units (i.e., the log odds of correct recall).
We report exponential betas (e
B
), and their 95% confidence intervals
(95% CI), which give the coefficient as an odds ratio (i.e., the odds of
correctly recalling a word divided by the odds of not recalling a
word). Thus, e
B
can be interpreted as the extent to which the odds
of recalling a word changed. Specifically, values greater than 1 rep-
resent an increased likelihood of recall while values less than 1 rep-
resent a decreased likelihood of recall. In each MLM, the only
random effect was the intercept which varied for each participant;
all fixed effects in each model are reported in the analyses. The anal-
ysis code for all models is available on OSF (Murphy, 2023).
Recall Performance and Selectivity
Selectivity as a function of attention at encoding and list is shown in
Figure 1a. To determine if participants under full and divided attention
were selective, we first conducted one-sample t-tests. Results revealed
that participants’selectivity scores both with full attention (M=.34,
SD =.26) and divided attention (M=.20, SD =.23) were different
from 0, full: t(50) =9.41, p,.001, d=1.32; divided: t(54) =
6.50, p,.001, d=.88. To examine group differences in selectivity,
a 2 (attention at encoding: full, divided)×6 (list: 1, 2, …,6) mixed
ANOVA did not reveal a main effect of list, F(5, 425) =2.07,
p=.068, η
p
2
=.02, but list interacted with attention at encoding,
F(5, 425) =2.90, p=.014, η
p
2
=.03, such that participants with full
attention became more selective with increased task experience.
Additionally, results revealed a main effect of attention at encoding,
F(1, 85) =4.09, p=.046, η
p
2
=.05, such that participants studying
the words with full attention were more selective than participants
studying under divided attention.
To examine recall and selectivity with value as a continuous pre-
dictor (see Figure 1b), a logistic MLM with item-level recall
Figure 1
Selectivity Index as a Function of Attention at Encoding and List (a)
and Probability of Recall as a Function of Attention at Encoding
and Word Value (b) in Experiment 1a
Note. Error bars reflect the standard error of the mean.
Table 1
Linear Regressions With Average Recall (Top) and Selectivity
(Bottom) Predicted by Average Divided Attention Task
Performance in Each Experiment
Experiment and measure BR
2
p
Experiment 1a—recall −.04 ,.01 .767
Experiment 1b—recall .29 .08 .061
Experiment 2a—recall −.01 ,.01 .950
Experiment 2b—recall .46 .21 ,.001
Experiment 3a—recall −.05 ,.01 .632
Experiment 3b—recall .03 ,.01 .762
Experiment 1a—selectivity .09 .01 .526
Experiment 1b—selectivity .20 .04 .195
Experiment 2a—selectivity .08 .01 .470
Experiment 2b—selectivity .02 ,.01 .861
Experiment 3a—selectivity .03 ,.01 .734
Experiment 3b—selectivity .17 .03 .053
MURPHY, SCHWARTZ, AND CASTEL
4
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

modeled as a function of value with attention at encoding (full,
divided) as a between-subjects factor revealed that value signifi-
cantly predicted recall, e
B
=1.08, 95% CI [1.08, 1.09], z=21.98,
p,.001, such that high-value words were better recalled than low-
value words. Additionally, attention significantly predicted recall,
e
B
=1.60, [1.27, 2.03], z=3.93, p,.001, such that participants
studying the words with full attention (M=.41, SD =.14) recalled
more words than participants studying the words under divided
attention (M=.30, SD =.12). Furthermore, value interacted with
attention, e
B
=1.04, [1.03, 1.06], z=5.72, p,.001, such that
value was a better predictor of recall for participants with full atten-
tion compared with participants under divided attention.
Retrieval Dynamics
To examine the dynamics of participants’recall, we first examined
the PFR as a function of word value (see Figure 2). Again, the PFR
captures how participants begin recall and in the present analysis,
refers to the proportion of the time the word with a given value
was the first word recalled. In our analyses of PFR, we do not report
the fixed effect for the different conditions as this effect is meaning-
less. A logistic MLM with PFR modeled as a function of value with
attention at encoding (full, divided) as a between-subjects factor
revealed that value significantly predicted PFR, e
B
=1.10, 95% CI
[1.08, 1.12], z=11.98, p,.001, such that participants tended to
begin recall with the highest valued words. However, value did
not interact with attention, e
B
=1.02, [.99, 1.05], z=1.23, p=.220.
In addition to the PFR, we calculated lag-CRPs to examine how a
word’s accompanying temporal and contextual information from the
study phase impacts recall. In this measure of how participants tran-
sition between responses during retrieval, lag is the ordinal distance
between successively recalled items (i.e., the lag between Items 1
and 6 would be 5), and the sign of the lag indicates the direction
of recall: Positive values indicate a forward transition and negative
values indicate a backward transition.The CRP for a recall transition
illustrates the likelihood that a word from serial position i+ lag is
recalled directly after a word from serial position i. The probability
of recalling an item from serial position xfollowed by the item
from position x+ lag is shown in Figure 3a.
1
To examine differences in the lag-recency effect as a function of
attention at encoding, we conducted a 5 (lag: 1–5; within-subjects
factor) ×2 (direction: forward vs. backward)×2 (attention
at encoding: full, divided) mixed ANOVA. Results revealed that
participants showed a forward preference for the direction of transi-
tions, F(1, 104) =104.23, p,.001, η
p
2
=.50, but this did not
differ as a function of attention, F(1, 104) =.13, p=.719,
η
p
2
,.01. Additionally, participants showed strong adjacency
effects, Mauchly’sW=.32, p,.001; Huynh–Feldt corrected
results: F(2.39, 248.66) =103.24, p,.001, η
p
2
=.50, but lag also
did not interact with attention, F(2.39, 248.66) =.27, p=.801,
η
p
2
,.01. There was an interaction between direction and lag,
Mauchly’sW=.27, p,.001; Huynh–Feldt corrected results:
F(2.33, 242.15) =40.68, p,.001, η
p
2
=.28, such that transitions
of lag 1 were more likely in the forward direction but there was
not a three-way interaction between direction, lag, and attention at
Figure 2
Probability of First Recall (PFR) as a Function of Attention at
Encoding and Word Value in Experiment 1a
Note. Error bars reflect the standard error of the mean.
Figure 3
Conditional-Response Probability (a) and Lag-Value Conditional-
Response Probability (b) Functions as a Function of Lag and
Attention at Encoding in Experiment 1a
Note. Error bars reflect the standard error of the mean.
1
When calculating the lag-CRPs, incorrect responses were included in
participants’output order. For example, if a participant recalled a correct
item, then an incorrect item, then another correct item, this last item’s output
position would be three.
VALUE-DIRECTED RETRIEVAL 5
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

encoding, F(2.33, 242.15) =.48, p=.646, η
p
2
=.01. Moreover, there
was a main effect of attention, F(1, 104) =11.11, p=.001, η
p
2
=.10,
such that, overall, participants with full attention were more likely to
organize retrieval based on the temporal proximity in the study phase
compared with participants under divided attention.
The probability of recalling an item of value xfollowed by an item
of value x+ lag is shown in Figure 3b. To examine differences in this
lag-value effect (i.e., lag-v CRP, see Stefanidi et al., 2018) as a func-
tion of attention at encoding, we conducted a 5 (lag: 1–5; within-
subjects factor) ×2 (direction: increasing vs. decreasing)×2(atten-
tion at encoding: full, divided) mixed ANOVA. Results revealed that
participants showed a decreasing value preference for the direction of
transitions, F(1, 104) =15.13, p,.001, η
p
2
=.13, but thisdid not dif-
fer as a function of attention, F(1, 104) =3.89, p=.051, η
p
2
=.04.
Participants did not show lag-value effects, F(3.85, 400.82) =.39,
p=.810, η
p
2
,.01, and lag-value also did not interact with attention,
F(3.85, 400.82) =1.26, p=.288, η
p
2
=.01. Moreover, there was
not an interaction between direction and lag, F(3.49, 362.48) =
1.29, p=.277, η
p
2
=.01, and there was not a three-way interaction
between direction, lag, and attention at encoding, F(3.49,
362.48) =1.51, p=.204, η
p
2
=.01. Moreover, therewas a main effect
of attention, F(1, 104) =8.30, p=.005, η
p
2
=.07, such that, overall,
participants with full attention were more likely to organize retrieval
according to word value compared with participants under divided
attention.
Discussion
In Experiment 1a, participants studied words paired with point
values while under full or divided attention. Results revealed
that dividing participants’attention during the study phase reduced
participants’ability to remember the words, consistent with
prior research (see Castel & Craik, 2003;Craik et al., 1996;
Naveh-Benjamin, Craik, Perretta, & Tonev, 2000). Additionally,
selectivity was reduced in participants under divided attention, con-
sistent with some previous work suggesting that, under certain con-
ditions, selectivity can be impaired when attention is divided at
encoding (see Elliott & Brewer, 2019;Murphy & Castel, 2022c;
Siegel & Castel, 2018b). However, PFR was preserved when par-
ticipants’attention was divided during the study phase, although
participants with divided attention demonstrated reduced
lag-recency effects. Together, Experiment 1a suggests that the stra-
tegic remembering of valuable information may depend on strategic
encoding processes and these processes can be disrupted by a sec-
ondary task.
Experiment 1b
In Experiment 1b, rather than a list of unassociated words paired
with objective point values, we presented participants with lists of 20
to-be-remembered items along a theme and of varying subjective
value. After a recall test, participants ranked the to-be-remembered
items from most important to least important (similar to assigning
point values, see McGillivray & Castel, 2017). When words are
not paired with arbitrary point values but instead offer schematic
support, retrieval dynamics may reveal more about how participants
prioritize certain items, initiate recall, and transition between items
when considered in terms of their subjective importance or potential
consequences for forgetting.
Method
Participants
Participants were 93 undergraduate students (M
age
=20.46,
SD
age
=2.29); no participants were excluded for cheating. On the
divided attention task, participants correctly identified an average
of 3.08 out of eight sequences (SD =1.50) on each list and there
was an average of 1.61 incorrect detections (SD =1.15) on each
list. As in Experiment 1a, if a participant failed to identify at least
one sequence (correctly or incorrectly) during a list, their data for
that list was excluded. This exclusion process resulted in 37 lists
being excluded from analysis (out of 270 total lists). A sensitivity
analysis based on the observed sample indicated that for a between-
subjects ANOVA with two groups (attention: full, divided) and six
measurements (list: 1, 2, …,6), assuming α=.05, power =.80,
and an average correlation of r=.16 between repeated-measures
(selectivity), the smallest effect the design could reliably detect is
η
p
2
=.03.
Materials and Procedure
Participants were instructed that they would be presented with six
lists of 20 items, with each list containing items along a theme (going
camping, going on vacation, child’s party, going to class, making
lasagna, going on a picnic; stimuli available on OSF). Participants
were randomly assigned to either study the word lists with full atten-
tion (n=48) or divided attention (n=45; the same digit detection
task from Experiment 1a). Each item was presented one at a time,
for 3 s each, in a randomized order. After the presentation of all
20 items, participants were given a 1-min free recall test in which
they were asked to recall all the items from the just-presented list.
Participants were also instructed that after the recall test, they
would be presented with all 20 items from that list and asked to
rank the items from most important to least important. When ranking
the items after recall, participants clicked and dragged items to
change their rank order and were required to spend a minimum of
1 min on this portion of the task.
Results
Recall Performance and Selectivity
Although items in this study were not paired with point values
counting toward a task score, we still computed selectivity index
scores by reverse scoring participants’rankings (i.e., the item ranked
most important was given a “point value”of 20).
2
We then calculated
each participant’s recall “score”(sum of the values of recalled items)
relative to their chance and ideal score based on these reversed scored
rankings. The ideal score consisted of the sum of only the highest val-
ues, or in this case, the items ranked as most important by each partic-
ipant, for the number of items recalled. Chance scores reflected no
attention to rankings and were calculated asthe product of the average
ranking and the number of recalled items. At chance, the score in our
example would be 10.5 multiplied by the number of recalled items. If
2
This selectivity measure may not capture a pure form of participants’
feeling of importance since items were ranked after the memory test.
Specifically, items that are more easily accessed and remembered during
retrieval may subsequently be ranked as more important.
MURPHY, SCHWARTZ, AND CASTEL
6
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

a participant only recalled the items that they ranked the highest, then
the resulting selectivity score would be 1 while a participant who only
recalledthe items that they ranked the lowest would receive a selectiv-
ity score of −1. Scores close to 0 indicate that a participant’s recall
was not sensitive to their rankings.
Selectivity as a function of attention at encoding and list is shown
in Figure 4a. To determine if participants under full and divided
attention were selective, we first conducted one-sample t-tests.
Results revealed that both selectivity scores with full attention
(M=.19, SD =.21) and divided attention (M=.25, SD =.14)
were different from 0, full: t(47) =6.38, p,.001, d=.92; divided:
t(44) =12.08, p,.001, d=1.80. To examine group differences in
selectivity for items participants ranked as important, a 2 (attention
at encoding: full, divided)×6 (list: 1, 2, …,6) mixed ANOVA did
not reveal a main effect of attention, F(1, 71) =2.18, p=.144,
η
p
2
=.03, such that participants studying the items with full attention
were similarly selective as participants studying the items under
divided attention. Additionally, results did not reveal a main effect
of list, F(5, 355) =.97, p=.433, η
p
2
=.01, and list did not interact
with attention, F(5, 355) =2.07, p=.068, η
p
2
=.03.
To examine recall and selectivity with participants’reverse scored
rankings as a continuous variable (see Figure 4b), a logistic MLM
with item-level recall modeled as a functionof participants’rankings
with attention at encoding (full, divided) as a between-subjects fac-
tor revealed that rank significantly predicted recall, e
B
=1.06, 95%
CI [1.06, 1.07], z=16.92, p,.001, such that items ranked as more
important to remember were better remembered than items ranked as
less important to remember. Additionally, attention significantly
predicted recall, e
B
=1.77, [1.30, 2.42], z=3.61, p,.001, such
that participants studying the items with full attention (M=.52,
SD =.19) recalled more items than participants studying the items
under divided attention (M=.38, SD =.13). Furthermore, rank
interacted with attention, e
B
=.97, [.97, 1.00], z=−2.08,
p=.038, such that rankings were a stronger predictor of recall for
participants under divided attention compared with participants
with full attention.
Retrieval Dynamics
To examine the dynamics of participants’recall, we examined the
PFR as a function of each participant’s rankings (see Figure 5). A
logistic MLM with PFR modeled as a function of participants’rank-
ings with attention at encoding (full, divided) as a between-subjects
factor revealed that rankings significantly predicted PFR, e
B
=1.02,
95% CI [1.01, 1.04], z=2.74, p=.006, such that participants
tended to begin recall with the top-ranked items as well as the
lowest-ranked item. This enhanced PFR for the lowest-ranked item
may result from increased distinctiveness (see Neath, 2010)asa
result of being considered the least important item or potentially
being less consistent with the list theme. However, rank did not inter-
act with attention, e
B
=.99, [.96, 1.02], z=−.65, p=.514.
To examine recall transitions, a 5 (lag: 1–5; within-subjects factor)
×2 (direction: forward vs. backward)×2 (attention at encoding:
full, divided) mixed ANOVA revealed that participants showed a for-
ward preference for the direction of transitions, F(1, 91) =5.24,
p=.024, η
p
2
=.05, and this differed as a function of attention,
F(1, 91) =12.16, p,.001, η
p
2
=.12, such that participants with
full attention showed a stronger forward preference than participants
under divided attention. Additionally, participants showed strong
adjacency effects, Mauchly’sW=.65, p,.001; Huynh–Feldt cor-
rected results: F(3.28, 307.58) =33.23, p,.001, η
p
2
=.27, but lag
did not interact with attention, F(3.38, 307.58) =2.52, p=.051,
η
p
2
=.03. There was also an interaction between direction and lag,
Mauchly’sW=.83, p=.046; Huynh–Feldt corrected results:
F(3.86, 351.37) =3.81, p=.005, η
p
2
=.04, such that transitions of
Figure 4
Selectivity Index as a Function of Attention at Encoding and List (a)
and Probability of Recall as a Function of Attention at Encoding
and Participants’Reverse-Scored Rankings (b) in Experiment 1b
Note. Error bars reflect the standard error of the mean.
Figure 5
Probability of First Recall (PFR) as a Function of Attention at
Encoding and Participants’Reverse-Scored Rankings in
Experiment 1b
Note. Error bars reflect the standard error of the mean.
VALUE-DIRECTED RETRIEVAL 7
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

lag 1 were more likely in the forward direction but there was not a
three-way interaction between direction, lag, and attention, F(3.86,
351.37) =.55, p=.693, η
p
2
=.01. However, there was a main effect
of attention, F(1, 91) =12.84, p,.001, η
p
2
=.12, such that partici-
pants with full attention demonstrated stronger lag-recency effects
than participants under divided attention (see Figure 6a).
The probability of recalling an item of rank xfollowed by an item
of rank x+ lag is shown in Figure 6b. To examine differences in the
lag-rank effect as a function of attention at encoding, we conducted
a 5 (lag: 1–5; within-subjects factor) ×2 (direction: increasing
vs. decreasing)×2 (attention at encoding: full, divided) mixed
ANOVA. Results revealed that participants did not show a
preference for the direction of transitions, F(1, 91) =.44, p=.507,
η
p
2
=.01, and this did not differ as a function of attention,
F(1, 91) =1.37, p=.244, η
p
2
=.02. However, participants showed
lag-rank effects, Mauchly’sW=.83, p=.048; Huynh–Feldt
corrected results: F(3.85, 350.75) =17.09, p,.001, η
p
2
=.16, and
lag-rank interacted with attention, F(3.85, 350.75) =3.12,
p=.016, η
p
2
=.03, such that participants with full attention showed
stronger rank-adjacency effects than participants under divided
attention. Furthermore, there was an interaction between direction
and lag, Mauchly’sW=.69, p,.001; Huynh–Feldt corrected
results: F(3.67, 334.35) =4.09, p=.004, η
p
2
=.04, such that transi-
tions of the lag-rank −1 were more likely than +1, but there was not a
three-way interaction between direction, lag, and attention
at encoding, F(3.67, 334.35) =1.15, p=.333, η
p
2
=.01. Finally,
there was a main effect of attention, F(1, 91) =12.13, p,.001,
η
p
2
=.12, such that participants with full attention demonstrated
stronger lag-rank effects than participants under divided attention.
Discussion
Although the divided attention task successfully reduced partici-
pants’ability to remember the items, there was some evidence that
selectivity for items ranked as important to remember was impaired
(though this effect might be smaller than Experiment 1a).
Regardless, PFR was again preserved under divided attention, indi-
cating that reduced attentional resources during encoding may not
impact certain strategic retrieval operations. However, CRPs were
reduced under divided attention, similar to Experiment 1a.
Since the to-be-remembered words in Experiment 1b were consis-
tent with a semantic theme, participants may have benefitted from
schematic support whereby prior knowledge enhances recall (see
Castel, 2005;Craik, 2002;Craik & Bosman, 1992;McGillivray &
Castel, 2017). Specifically, reduced attentional resources during
encoding may hinder one’s ability to encode valuable or important
words but participants can still engage in value-directed retrieval
by harnessing the benefits of schematic support to recall important
items, though learners with full attention may be better able to
remember important items. Thus, in addition to strategic encoding
processes, there are likely strategic retrieval operations that contrib-
ute to value-directed remembering.
Experiment 2a
Experiment 1 indicated that sensitivity to the value or importance
of to-be-remembered words may be impaired under divided atten-
tion during encoding. In Experiment 2a, rather than reducing partic-
ipants’attentional resources during encoding, we were interested in
the effects of divided attention during recall on selective memory
and strategic retrieval operations. If the retrieval trends and dynamics
like selectivity, PFR, and CRPs are disrupted when under divided
attention at retrieval, this would provide evidence that value-directed
remembering requires not simply the strategic allocation of attention
during encoding, but also strategic retrieval operations.
Although we used a digit detection task to divide participants’
attention during encoding in Experiment 1, in Experiment 2, all par-
ticipants completed the study phase with full attention but simulta-
neously completed either no task, a tone detection task, or an
animacy task during retrieval. We included the animacy task during
retrieval as this task may have a greater impact on participants’abil-
ity to remember words (i.e., the animacy task taxes the same modal-
ity as learning and remembering word lists) but may also impair
selective memory; the tone detection task involves a similar discrim-
ination decision as the animacy task and served as a nonverbal com-
parison to the animacy task. Moreover, we did not use the animacy
task to divide participants’attention during encoding as the animacy
task may result in recall intrusions. For example, participants may
mistakenly recall words from the animacy task which could disrupt
retrieval processes. Thus, if we used the animacy task during
Figure 6
Conditional-Response Probability (a) and Lag-Rank
Conditional-Response Probability (b) Functions as a Function of
Lag and Attention at Encoding in Experiment 1b
Note. Error bars reflect the standard error of the mean.
MURPHY, SCHWARTZ, AND CASTEL
8
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

encoding, both encoding and retrieval processes may be affected
rather than just encoding processes.
Method
Participants
After 10 exclusions due to cheating, participants were 130 under-
graduate students (M
age
=20.24, SD
age
=1.61). Participants were
also excluded for failing to complete the divided attention tasks
with at least 50% accuracy (as seen in Siegel & Castel, 2018b;
Siegel et al., 2021). This exclusion process resulted in the exclusion
of 10 participants for poor tone detection performance and 15 partic-
ipants for poor animacy performance. On the tone discrimination
task, participants correctly identified 83.4% of the tones (SD =
15.7%) and on the animacy task, participants correctly identified
74.8% of the items (SD =14.3%). A sensitivity analysis based on
the observed sample indicated that for a between-subjects
ANOVA with three groups (attention: full, divided by tone task,
divided by animacy task) and six measurements (list: 1, 2, …,6),
assuming α=.05, power =.80, and an average correlation of
r=.35 between repeated-measures (selectivity), the smallest effect
the design could reliably detect is η
p
2
=.03.
Materials and Procedure
The task in Experiment 2a was similar to the task in Experiment
1a. All participants studied the words with full attention but either
completed the recall phase with full attention (n=52) or divided
attention. Participants under divided attention either completed a
tone identification task (n=44) or an animacy task (n=34) during
retrieval. In the divided attention conditions, the tone identification
and animacy tasks occurred while participants simultaneously
tried to recall the to-be-remembered words for 1 min.
Participants recalling the to-be-remembered items while complet-
ing a tone identification task were told that they would hear a series
of low-pitched (400 Hz) and high-pitched (900 Hz) tones during the
test phase. Each tone was played for 1 s with a 3-s interstimulus
interval between each tone. Tone sequences were randomly gener-
ated for each participant. Participants were instructed to indicate
(on the keyboard) whether each pitch they heard was low or high,
and the text “awaiting tone response”would appear on the screen
if participants did not respond to the tones. Participants completed
a short tone discrimination practice session before beginning the
task.
In the animacy divided attention task, while participants recalled
the to-be-remembered items, they were simultaneously read a list
of 15 items (one item was read every 3 s) and had to indicate
whether each item was an animal or a manmade object via key-
board clicks (adapted from Fernandes & Moscovitch, 2002).
Each recall phase contained a pseudorandomized sequence of 15
animal and manmade objects generated in accordance with the fol-
lowing three conditions: (a) animal and manmade objects appeared
at least four times each in the sequence, (b) the longest same-
category (i.e., animal or manmade) sequential occurrence did not
exceed three-in-a-row, and (c) both animal and manmade objects
changed (i.e., animal-to-manmade or manmade-to-animal) at
least three times throughout the sequence.
Results
Recall Performance and Selectivity
Selectivity as a function of attention at retrieval and list is shown
in Figure 7a. To determine if participants were selective, we first
conducted one-sample t-tests. Results revealed that participants’
selectivity scores with full attention (M=.31, SD =.24), divided
attention via the tone task (M=.35, SD =.28), and divided attention
via the animacy task (M=.31, SD =.27) were different from 0, full:
t(51) =9.26, p,.001, d=1.28; tones: t(43) =8.30, p,.001, d=
1.25; animacy: t(33) =6.80, p,.001, d=1.17. To examine group
differences in selectivity, a 3 (Attention at Retrieval: Full, Divided
by Tone Task, Divided by Animacy Task)×6 (List: 1, 2, …,6)
mixed ANOVA revealed a main effect of list, F(5, 615) =7.83,
p,.001, η
p
2
=.06, such that participants became more selective
with increased task experience but list did not interact with attention
at retrieval, F(10, 615) =.60, p=.819, η
p
2
=.01. Moreover, results
did not reveal a main effect of attention, F(2, 123) =.24, p=.787,
η
p
2
,.01, such that participants were similarly selective whether
recalling words under full or divided attention.
To examine recall and selectivity with value as a continuous pre-
dictor (see Figure 7b), we conducted a logistic MLM with item-level
recall modeled as a function of value with attention at retrieval (full,
divided by tone task, divided by animacy task) as a between-subjects
factor. In this model and all subsequent models, participants in the
full attention condition served as the reference group. Results
revealed that value significantly predicted recall, e
B
=1.10, 95%
Figure 7
Selectivity Index as a Function of Attention at Retrieval and List (a)
and Probability of Recall as a Function of Attention at Retrieval
and Word Value (b) in Experiment 2a
Note. Error bars reflect the standard error of the mean.
VALUE-DIRECTED RETRIEVAL 9
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

CI [1.10, 1.11], z=30.81, p,.001, such that high-value words
were better recalled than low-value words. Additionally, when com-
paring the divided attention conditions with participants with full
attention, participants completing the tone task during recall
(M=.38, SD =.10) recalled a similar proportion of words as partic-
ipants with full attention (M=.41,SD=.14), e
B
=.88, [.69, 1.11],
z=−1.07, p=.283. However, participants completing the animacy
task during recall (M=.35, SD =.13) recalled fewer words than par-
ticipants with full attention, e
B
=.77, [.59, .99], z=−2.04,
p=.042. Furthermore, value was a better predictor of recall for par-
ticipants completing the tone task during recall than participants
with full attention, e
B
=1.02, [1.00, 1.03], z=1.97, p=.049, but
participants completing the animacy task during recall and partici-
pants with full attention were similarly selective, e
B
=1.00, [.98,
1.01], z=−.55, p=.580.
Retrieval Dynamics
To examine the dynamics of participants’recall, we examined the
PFR as a function of word value (see Figure 8). A logistic MLM with
PFR modeled as a function of value with attention at retrieval (full,
divided by tone task, divided by animacy task) as a between-subjects
factor revealed that value significantly predicted PFR, e
B
=1.10,
95% CI [1.08, 1.11], z=13.11, p,.001, such that participants
tended to begin recall with the highest valued words. However,
there were no significant interactions with value between partici-
pants completing the tone task during recall and participants with
full attention or participants completing the animacy task during
recall and participants with full attention, both ps..438.
To examine differences in the lag-recency effect as a function of
attention at retrieval (see Figure 9a), we conducted a 5 (lag: 1–5;
within-subjects factor) ×2 (direction: forward vs. backward)×3
(attention at retrieval: full, divided by tone task, divided by animacy
task) mixed ANOVA. Results revealed that participants showed a
forward preference for the direction of transitions, F(1, 127) =
118.14, p,.001, η
p
2
=.48, but this did not differ as a function of
attention, F(2, 127) = .34, p=.713, η
p
2
=.01. Additionally, partici-
pants showed strong adjacency effects, Mauchly’sW=.44,
p,.001; Huynh–Feldt corrected results: F(2.81, 357.22) =164.25, p,.001, η
p
2
=.56, and lag interacted with attention,
F(5.63, 357.22) =4.06, p,.001, η
p
2
=.06, such that participants
whose attention was divided by tones demonstrated a reduced
lag-recency effect for lags of 1. There was also an interaction
between direction and lag, Mauchly’sW=.52, p,.001; Huynh–
Feldt corrected results: F(3.00, 380.94) =36.72, p,.001,
η
p
2
=.22, such that transitions of lag 1 were more likely in the for-
ward direction but there was not a three-way interaction between
direction, lag, and attention, F(6.00, 380.94) =1.08, p=.371,
η
p
2
=.02. Furthermore, there was not a main effect of attention,
F(2, 127) =1.02, p=.363, η
p
2
=.02.
The probability of recalling an item of valuexfollowed by an item
of value x+ lag is shown in Figure 9b. To examine differences in the
lag-value effect as a function of attention at retrieval, we conducted
a 5 (lag: 1–5; within-subjects factor) ×2 (direction: increasing vs.
decreasing)×3 (attention at retrieval: full, divided by tone task,
divided by animacy task) mixed ANOVA. Results revealed that
participants showed a decreasing preference for the direction of
transitions, F(1, 127) =38.61, p,.001, η
p
2
=.23, but this did not
differ as a function of attention, F(2, 127) =.48, p=.621,
η
p
2
=.01. Additionally, participants did not show lag-value effects,
Figure 8
Probability of First Recall (PFR) as a Function of Attention at
Retrieval and Word Value in Experiment 2a
Note. Error bars reflect the standard error of the mean.
Figure 9
Conditional-Response Probability (a) and Lag-Value
Conditional-Response Probability (b) Functions as a Function of
Lag and Attention at Encoding in Experiment 2a
Note. Error bars reflect the standard error of the mean.
MURPHY, SCHWARTZ, AND CASTEL
10
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

F(4, 508) =1.77, p=.134, η
p
2
=.01, and lag-value also did not
interact with attention, F(8, 508) =1.14, p=.332, η
p
2
=.02.
Furthermore, there was not an interaction between direction
and lag, F(4, 508) =1.34, p=.254, η
p
2
=.01, and there was not a
three-way interaction between direction, lag, and attention at
retrieval, F(8, 508) =1.04, p=.403, η
p
2
=.02. Moreover, there
was not a main effect of attention, F(2, 127) =.49, p=.613,
η
p
2
=.01, indicating that divided attention during recall did not influ-
ence the organization of recall according to value.
Discussion
In Experiment 2a, we again presented participants with words
paired with point values, but participants either recalled the words
under full or divided attention (tone discrimination or animacy
task). Results revealed that the animacy task, but not the tone dis-
crimination task, impaired overall recall performance, consistent
with prior work indicating that there are greater costs of divided
attention at retrieval if the tasks overlap (Craik et al., 1996;
Fernandes & Moscovitch, 2000,2002,2003;Naveh-Benjamin,
Craik, Gavrilescu, & Anderson, 2000;Naveh-Benjamin, Craik,
Perretta, & Tonev, 2000;Naveh-Benjamin et al., 1998;Siegel et
al., 2021;Skinner & Fernandes, 2008) as well as research suggesting
that a secondary task during recall often has minimal effects on
retrieval (e.g., Rohrer & Pashler, 2003). However, despite some
recall impairments, there were no differences in selectivity for valu-
able information as a function of attention at retrieval (although there
was some evidence that selectivity was enhanced for participants
completing the tone task). Moreover, there were no group differ-
ences in PFR, and the lag-recency effect was preserved when
under divided attention during retrieval. Thus, the ability to selec-
tively remember valuable information, and the retrieval operations
contributing to selective memory, appear to be preserved under
divided attention during retrieval.
Experiment 2b
In Experiment 2b, we presented participants with lists of
to-be-remembered words along a theme (similar to Experiment
1b) rather than unassociated words. Similar to Experiment 2a, all
participants completed the study phase with full attention but the
retrieval phase under either full or divided attention (via a tone detec-
tion or animacy task). Consistent with Experiment 2a, we expected
participants to demonstrate preserved selectivity and strategic
retrieval operations when under divided attention during retrieval.
Method
Participants
After exclusions, participants were 128 undergraduate students
(M
age
=20.02, SD
age
=1.42); no participants were excluded for
cheating. Participants were also excluded for failing to complete
the divided attention tasks with at least 50% accuracy. This exclusion
process resulted in the exclusion of seven participants for poor tone
detection performance and 21 participants for poor animacy perfor-
mance. On the tone discrimination task, participants correctly iden-
tified 88.3% of the tones (SD =12.3%) and on the animacy task,
participants correctly identified 79.6% of the items (SD =10.1%).
A sensitivity analysis based on the observed sample indicated that
for a between-subjects ANOVA with three groups (attention: full,
divided by tone task, divided by animacy task) and six measurements
(list: 1, 2, …,6), assuming α=.05, power =.80, and an average cor-
relation of r=.05 between repeated-measures (selectivity), the
smallest effect the design could reliably detect is η
p
2
=.02.
Materials and Procedure
The task in Experiment 2b was similar to the task in Experiment
1b. All participants studied the words with full attention but either
completed the recall phase with full attention (n=49) or divided
attention. Similar to Experiment 2a, participants under divided atten-
tion either completed a tone detection task (n=46) or an animacy
task (n=33) during retrieval.
Results
Recall Performance and Selectivity
Selectivity as a function of attention at retrieval and list is shown
in Figure 10a. To determine if participants under full and divided
attention were selective, we again scored participants for recall effi-
ciency using their reverse-scored rankings. One-sample t-tests
revealed that participants’selectivity scores with full attention
(M=.17, SD =.19), divided attention via the tone task (M=.17,
SD =.14), and divided attention via the animacy task (M=.16,
SD =.12) were different from 0, full: t(48) =6.03, p,.001,
d=.86; tones: t(45) =8.62, p,.001, d=1.27; animacy:
Figure 10
Selectivity Index as a Function of Attention at Retrieval and List (a)
and Probability of Recall as a Function of Attention at Retrieval
and Participants’Reverse-Scored Rankings (b) in Experiment 2b
Note. Error bars reflect the standard error of the mean.
VALUE-DIRECTED RETRIEVAL 11
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

t(32) = 7.71, p,.001, d=1.34. To examine group differences in
selectivity for items participants ranked as important, a 3 (attention
at retrieval: full, divided by tone task, divided by animacy task)×6
(list: 1, 2, …,6) mixed ANOVA revealed a main effect of list,
F(5, 570) =3.06, p=.010, η
p
2
=.03, such that selectivity decreased
with task experience but list did not interact with attention
at retrieval, F(10, 570) =1.02, p=.424, η
p
2
=.02. Additionally,
results did not reveal a main effect of attention, F(2, 114) =.17,
p=.846, η
p
2
,.01.
To examine recall and selectivity with participants’reverse scored
rankings as a continuous measure (see Figure 10b), a logistic MLM
with item-level recall modeled as a functionof participants’rankings
with attention at retrieval (full, divided by tone task, divided by ani-
macy task) as a between-subjects factor revealed that participants’
rankings (reverse scored) significantly predicted recall, e
B
=1.05,
95% CI [1.04, 1.05], z=15.18, p,.001, such that items ranked
as important to remember were better remembered than items ranked
as less important to remember. Additionally, when comparing the
divided attention conditions and participants with full attention, par-
ticipants completing the tone task during recall (M=.46, SD =.12)
recalled fewer items than participants with full attention (M=.55,
SD = .16), e
B
=.68, [.53, .89], z=−2.86, p=.004. Furthermore,
participants completing the animacy task during recall (M=.34,
SD =.13) recalled fewer items than participants with full attention,
e
B
=.39, [.29, .52], z=−6.45, p,.001. However, there were no
significant interactions with rankings between participants complet-
ing the tone task during recall and participants with full attention or
participants completing the animacy task during recall and partici-
pants with full attention, both ps..249.
Retrieval Dynamics
To examine the dynamics of participants’recall, we first examined
the PFR as a function of each participant’s reverse-scored rankings
(see Figure 11). A logistic MLM with PFR modeled as a function
of participants’rankings with attention at retrieval (full, divided by
tone task, divided by animacy task) as a between-subjects factor
revealed that rankings significantly predicted PFR, e
B
=1.04, 95%
CI [1.02, 1.04], z=5.34, p,.001, such that participants tended to
begin recall with the top-ranked items as well as the lowest-ranked
items. Similar to Experiment 1b, this increased tendency to initiate
recall with items ranked as least important to remember may be attrib-
utable to the potentially increased distinctiveness of these items.
However, there were no significant interactions with rankings between
participants completing the tone task during recall and participants
with full attention or participants completing the animacy task during
recall and participants with full attention, both ps..362.
CRPs as a function of direction, lag, and attention at retrieval are
shown in Figure 12a. A 5 (lag: 1–5; within-subjects factor) ×2
(direction: forward vs. backward)×3 (attention at retrieval: full,
divided by tone task, divided by animacy task) mixed ANOVA
revealed that participants showed a forward preference for the direc-
tion of transitions, F(1, 125) =28.33, p,.001, η
p
2
=.19, but this
did not differ as a function of attention, F(2, 125) =.54, p=.585,
η
p
2
=.01. However, participants showed strong adjacency effects,
Mauchly’sW=.62, p,.001; Huynh–Feldt corrected results:
F(3.32, 415.54) =76.97, p,.001, η
p
2
=.38, and lag interacted
with attention, F(6.65, 415.54) =1.50, p=.170, η
p
2
=.02, such
that participants with full attention showed the strongest adjacency
Figure 11
Probability of First Recall (PFR) as a Function of Attention at
Retrieval and Participants’Reverse-Scored Rankings in
Experiment 2b
Note. Error bars reflect the standard error of the mean.
Figure 12
Conditional-Response Probability (a) and Lag-Rank Conditional-
Response Probability (b) Functions as a Function of Lag and
Attention at Encoding in Experiment 2b
Note. Error bars reflect the standard error of the mean.
MURPHY, SCHWARTZ, AND CASTEL
12
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

effects for lag 1. There was also an interaction between direction and
lag, Mauchly’sW=.66, p,.001; Huynh–Feldt corrected results:
F(3.33, 415.61) =8.06, p,.001, η
p
2
=.06, such that participants
were most likely to transition forward one lag, but there was not a
three-way interaction between direction, lag, and attention, F(6.65,
415.61) =1.26, p=.269, η
p
2
=.02. Moreover, there was a main
effect of attention, F(2, 125) =6.45, p=.002, η
p
2
=.09, such that
participants with full attention demonstrated stronger lag-recency
effects than participants completing the animacy task during recall,
p
holm
=.004, d=.29, and participants completing the tone task dur-
ing recall, p
holm
=.014, d=.24; however, there were no differences
in the lag-recency effect between the divided attention conditions,
p
holm
=. 439, d=.07.
The probability of recalling an item of rank xfollowed by an item
of rank x+ lag is shown in Figure 12b. To examine differences in the
lag-rank effect as a function of attention at retrieval, we conducted
a 5 (lag: 1–5; within-subjects factor) ×2 (direction: increasing
vs. decreasing) 3 (attention at retrieval: full, divided by tone task,
divided by animacy task) mixed ANOVA. Results revealed that par-
ticipants did not show a preference for the direction of transitions,
F(1, 125) =.54, p=.465, η
p
2
,.01, and this did not differ as a func-
tion of attention, F(2, 125) =.22, p=.803, η
p
2
,.01. However, par-
ticipants showed lag-rank effects, Mauchly’sW=.83, p=.008;
Huynh–Feldt corrected results: F(3.75, 469.05) =32.04, p,.001,
η
p
2
=.20, but lag-rank did not interact with attention, F(7.51,
469.05) =1.35, p=.223, η
p
2
=.02. Moreover, there was an interac-
tion between direction and lag, F(4, 500) =2.40, p=.049, η
p
2
=.02,
such that participants were most likely to transition decreasing by
one rank, but there was not a three-way interaction between direc-
tion, lag, and attention at retrieval, F(8, 500) =1.96, p=.050,
η
p
2
=.03. Additionally, there was a main effect of attention,
F(2, 125) =5.17, p=.007, η
p
2
=.08, such that participants with full
attention demonstrated stronger lag-rank effects than participants
completing the animacy task during recall, p
holm
=.013, d=.25,
but not participants completing the tone task during recall, p
holm
=
.946, d=.01; moreover, participants completing the tone task during
recall demonstrated stronger lag-rank effects than participants com-
pleting the animacy task during recall, p
holm
=.013, d=.26.
Discussion
In Experiment 2b, both divided attention tasks impaired recall but
there were no group differences in selectivity for important informa-
tion or PFR. However, the lag-recency effect was impaired in partic-
ipants under divided attention during recall, indicating that reduced
attentional resources during retrieval can impair one’s ability to
recruit a recalled item’s accompanying temporal–contextual infor-
mation to recall additional words.
Experiment 3a
Since we used different divided attention tasks in Experiments 1
and 2, we cannot directly compare the effects of divided attention
at encoding and retrieval as any observed differences between
encoding and retrieval could potentially be explained by the divided
attention task used. As such, in Experiment 3, we directly compared
divided attention at encoding and retrieval using the same task (tone
discrimination in Experiment 3a and digit detection in Experiment
3b). Generally, we expected to replicate the effects observed in
Experiments 1 and 2 such that selectivity is impaired when attention
is divided at encoding but not when attention is divided at retrieval.
Method
Participants
After exclusions, participants were 155 undergraduate students
(M
age
=19.86, SD
age
=1.19); three participants were excluded for
cheating. Participants were also excluded for failing to complete the
divided attention tasks with at least 50% accuracy. This exclusion pro-
cess resulted in the exclusion of 11 participants for poor performance
during encoding and 19 participants for poor performance during
retrieval. When the tone discrimination task occurred during encod-
ing, participants correctly identified 89.0% of the tones (SD =
9.4%). When the tone discrimination task occurred during retrieval,
participants correctly identified 81.4% of the tones (SD =12.0%)—
this difference was significant, t(102) =3.59, p,.001, d=.71. A
sensitivity analysis based on the observed sample indicated that for
a between-subjects ANOVA with three groups (attention: full, divided
attention during encoding, divided attention during retrieval)andsix
measurements (list: 1, 2, …,6), assuming α=.05, power =.80, and
an average correlation of r=.33 between repeated-measures (selec-
tivity), the smallest effect the design could reliably detect is η
p
2
=.03.
Materials and Procedure
The task in Experiment 3a was similar to the task in Experiment
2a. One group completed both the study and test phases with full
Figure 13
Selectivity Index for Each Group as a Function of List (a) and
Probability of Recall for Each Group as a Function of Word
Value (b) in Experiment 3a
Note. Error bars reflect the standard error of the mean.
VALUE-DIRECTED RETRIEVAL 13
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

attention (n=51), one group completed the study phase under
divided attention but the test phase with full attention (n=55),
and one group completed the study phase with full attention but
the test phase under divided attention (n=49). The divided attention
task required participants to identify tones as either low- or high-
pitched (similar to the procedure used in Experiment 2).
Specifically, during the study or test phase (each 60 s long), partic-
ipants heard 20 tones played for 1 s each with 2 s between each tone.
Participants’task was to indicate whether the tone they heard was
low- or high-pitched.
Results
Recall Performance and Selectivity
Selectivity for each group as a function of list is shown in
Figure 13a. To determine if participants were selective, we first con-
ducted one-sample t-tests. Results revealed that participants’selec-
tivity scores with full attention (M=.31, SD =.26), divided
attention during encoding (M=.28, SD =.24), and divided atten-
tion during retrieval (M=.33, SD =.22) were different from 0,
full: t(50) =8.59, p,.001, d=1.20; divided attention during
encoding: t(54) =8.60, p,.001, d=1.16; divided attention during
retrieval: t(48) =10.54, p,.001, d=1.51. To examine group dif-
ferences in selectivity, a 3 (Attention: full, divided attention during
encoding, divided attention during retrieval)×6 (list: 1, 2, …,6)
mixed ANOVA revealed a main effect of list, F(5, 740) =6.61,
p,.001, η
p
2
=.04, such that participants became more selective
with increased task experience but list did not interact with atten-
tion, F(10, 740) =.97, p=.467, η
p
2
=.01. Moreover, results did
not reveal a main effect of attention, F(2, 148) =.56, p=.571,
η
p
2
=.01, such that participants were similarly selective whether
recalling words under full or divided attention.
To examine recall and selectivity with value as a continuous pre-
dictor (see Figure 13b), a logistic MLM with item-level recall mod-
eled as a function of value with attention (full, divided attention
during encoding, divided attention during retrieval) as a between-
subjects factor revealed that value significantly predicted recall,
e
B
=1.10, 95% CI [1.09, 1.10], z=31.99, p,.001, such that
high-value words were better recalled than low-value words.
Additionally, when comparing the divided attention conditions
with participants with full attention, participants under divided atten-
tion during encoding (M=.32, SD =.10) recalled a smaller propor-
tion of words than participants with full attention (M=.43,SD
=.16), e
B
=.58, [.45, .75], z=−4.21, p,.001. However, partici-
pants under divided attention during recall (M=.40, SD =.14)
recalled a similar proportion of words as participants with full atten-
tion, e
B
=.87, [.67, 1.12], z=−1.08, p=.281. Neither comparison
interacted with value, both ps..177.
Retrieval Dynamics
To examine the dynamics of participants’recall, we examined the
PFR as a function of word value (see Figure 14). A logistic MLM
with PFR modeled as a function of value with attention (full, divided
attention during encoding, divided attention during retrieval) as a
between-subjects factor revealed that value significantly predicted
PFR, e
B
=1.10, 95% CI [1.08, 1.11], z=14.42, p,.001, such
that participants tended to begin recall with the highest valued
words. However, there were no significant interactions with value
between participants completing the tone task during recall and par-
ticipants with full attention or participants completing the animacy
task during recall and participants with full attention, both
ps..142.
To examine differences in the lag-recency effect as a function of
attention at retrieval (see Figure 15a), we conducted a 5 (lag: 1–5;
within-subjects factor) ×2 (direction: forward vs. backward)×3
(attention: full, divided attention during encoding, divided attention
during retrieval) mixed ANOVA. Results revealed that participants
showed a forward preference for the direction of transitions,
F(1, 152) =163.91, p,.001, η
p
2
=.52, but this did not differ
as a function of attention, F(2, 152) =.16, p=.851, η
p
2
,.01.
Additionally, participants showed strong adjacency effects,
Mauchly’sW=.35, p,.001; Huynh–Feldt corrected results:
F(2.47, 375.50) =236.56, p,.001, η
p
2
=.61, but lag did not inter-
act with attention, F(4.94, 375.50) =.37, p=.869, η
p
2
=.01. There
was an interaction between direction and lag, Mauchly’sW=.46,
p,.001; Huynh–Feldt corrected results: F(2.78, 422.99) =58.03,
p,.001, η
p
2
=.28, such that transitions of lag 1 were more likely
in the forward direction, but there was not a three-way interaction
between direction, lag, and attention, F(5.57, 422.99) =.63,
p=.691, η
p
2
=.01. However, there was a main effect of attention,
F(2, 152) =4.20, p=.017, η
p
2
=.05, such that the lag recency effect
was greater with full attention than divided attention at encoding,
p
holm
=.015, d=.15, but there were no other significant compari-
sons, both ps..129.
The probability of recalling an item of valuexfollowed by an item
of value x+ lag is shown in Figure 15b. To examine differences in
the lag-value effect as a function of attention at retrieval, we con-
ducted a 5 (lag: 1–5; within-subjects factor) ×2 (direction: increas-
ing vs. decreasing)×3 (attention: full, divided attention during
encoding, divided attention during retrieval) mixed ANOVA.
Results revealed that participants showed an increasing pref-
erence for the direction of transitions, F(1, 152) =27.78, p,.001,
η
p
2
=.16, but this did not differ as a function of attention,
F(2, 152) =.46, p=.635, η
p
2
=.01. Additionally, participants did
not show lag-value effects, F(4, 608) =.48, p=.752, η
p
2
,.01,
and lag-value also did not interact with attention, F(8, 608) =.92,
p=.499, η
p
2
=.01. There was an interaction between direction and
Figure 14
Probability of First Recall (PFR) for Each Group as a Function of
Value in Experiment 3a
Note. Error bars reflect the standard error of the mean.
MURPHY, SCHWARTZ, AND CASTEL
14
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

lag, F(4, 608) =2.67, p=.031, η
p
2
=.02, such that transitions greater
than one lag value were more likely in the increasing direction, but
there was not a three-way interaction between direction, lag, and atten-
tion at retrieval, F(8, 608) =.13, p=.998, η
p
2