Episodic memories predict adaptive value-based decision-
making
Vishnu Murty
a,*
, Oriel FeldmanHall
a,*
, Lindsay E. Hunter
a
, Elizabeth A Phelps
a,b,c
, and Lila
Davachi
a,b
a
Department of Psychology, New York University, New York NY 10003
b
Center for Neural Science, New York University, New York NY 10003
c
Nathan Kline Institute, Orangeburg, NY 10962
Abstract
Prior research illustrates that memory can guide value-based decision-making. For example,
previous work has implicated both working memory and procedural memory (i.e., reinforcement
learning) in guiding choice. However, other types of memories, such as episodic memory, may
also influence decision-making. Here we test the role for episodic memory—specifically item
versus associative memory—in supporting value-based choice. Participants completed a task
where they first learned the value associated with trial unique lotteries. After a short delay, they
completed a decision-making task where they could choose to re-engage with previously
encountered lotteries, or new never before seen lotteries. Finally, participants completed a surprise
memory test for the lotteries and their associated values. Results indicate that participants chose to
re-engage more often with lotteries that resulted in high versus low rewards. Critically, participants
not only formed detailed, associative memories for the reward values coupled with individual
lotteries, but also exhibited adaptive decision-making only when they had intact associative
memory. We further found that the relationship between adaptive choice and associative memory
generalized to more complex, ecologically valid choice behavior, such as social decision-making.
However, individuals more strongly encode experiences of social violations—such as being treated
unfairly, suggesting a bias for how individuals form associative memories within social contexts.
Together, these findings provide an important integration of episodic memory and decision-
making literatures to better understand key mechanisms supporting adaptive behavior.
Keywords
episodic memory; value; decision-making; dictator game; associative memory
Corresponding Author Information:
Lila Davachi, 6 Washington Place, New York University, New York NY 10003,
lila.davachi@nyu.edu, 212-998-7896.
*Denotes shared first authorship
HHS Public Access
Author manuscript
J Exp Psychol Gen
. Author manuscript; available in PMC 2017 May 01.
Published in final edited form as:
J Exp Psychol Gen
. 2016 May ; 145(5): 548–558. doi:10.1037/xge0000158.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
INTRODUCTION
Successfully navigating through the world requires individuals to integrate details from
previous experiences with current environmental demands. The seemingly ubiquitous need
for memory in decision-making would suggest that these two fields intimately collaborate to
decompose the complex relationship between the representation of past experiences and the
subsequent effects on choice. While previous research has demonstrated that memory can
influence decisions (Daw & Doya, 2006; Fischhoff, 1975; Hawkins & Hastie, 1990; E. J.
Johnson, Häubl, & Keinan, 2007; Madan, Ludvig, & Spetch, 2014; Montague & Berns,
2002; Tversky & Kahneman, 1973; E. U. Weber et al., 2007), empirical research directly
integrating detailed assays of episodic memory and value-based decision-making has
remained largely elusive, leaving critical questions unanswered. How does episodic memory
influence value-based choice? How do individuals encode value into episodic memory and
how are these memories then accessed during choice? And finally, is value encoding and
retrieval domain general such that the same mechanism is deployed across various
situational contexts? To address these questions, here we directly explore the interface
between episodic memory processes and value-based decision-making.
Research on episodic memory and decision-making has predominantly been studied
independently (Elke U. Weber, Goldstein, & Barlas, 1995). Behavioral and neural
investigations of episodic memory have revealed that representations of the past can be
supported by at least two distinct processes, which has successfully fractionated episodic
memory into discrete components (Brown & Aggleton, 2001; Davachi, 2006; Yonelinas &
Jacoby, 2012). For instance, individuals can have memories for individual features of an
experience without any retrieving any details of the context, which is known as item
memory. Alternatively, individuals can recall the associative relationships between discrete
features of an experience, resulting in memory for an item
and
the context in which it
occurred. Unlike item memory, these associative memories can involve the recovery of
dynamic multidimensional information reflecting specific, detailed experiences. Decades of
rigorous trans-species research has elucidated the process by which these memory
components are stored, represented, and retrieved (Davachi, 2006; Eichenbaum, Sauvage,
Fortin, Komorowski, & Lipton, 2012). Yet, how item and associative memory guide
decision-making has not been formally tested or described.
Historically, classic judgment and decision-making research has only loosely linked memory
(broadly construed) with value-based choice. For example, some theorists have loosely
drawn on working memory research to explain strategic decision-making. Early work
demonstrating that working memory is limited (Miller, 1956) has been used as evidence that
individuals do not act as pure ‘rationalists’, as decisions are often made under conditions
where there is finite information (Simon, 1956) (for a more comprehensive review see
Weber et al 1995). However, the transient nature of working memory representations may
not be an ideal memory system to support decision-making in isolation, both because of its
capacity limitations and fast decay rate.
More recently, the burgeoning field of value based decision-making has examined the
intersection of choice and memory within the context of many, repeated experiences (Doll,
Murty et al. Page 2
J Exp Psychol Gen
. Author manuscript; available in PMC 2017 May 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Shohamy, & Daw, 2015; Palombo, Keane, & Verfaellie, 2015). Results reveal that across
hundreds of trials, individuals can form simple stimulus-response associations that can guide
subsequent choice (Daw & Doya, 2006). The processes underlying stimulus response
learning are more characteristic of procedural memory (e.g., skill and condition memory, see
Schacter & Tulving, 1994). Although this work has robustly identified a role for types of
procedural memory in supporting stimulus-response learning, it does not preclude the
possibility that associative memory may also be influencing choice. Problematically,
however, exposing individuals to many, repeated experiences limits the ability to directly
disambiguate incremental procedural memory from episodic memory, making it difficult to
extract exactly how—and at what stage—episodic memory might influence choice. Despite
these limitations, empirical work has demonstrated that the retrieval of prior experiences
during choice shape decision-making (Gonzalez & Dutt, 2011; E. J. Johnson et al., 2007;
Ludvig, Madan, & Spetch, 2015; Madan et al., 2014; Elke U. Weber et al., 1995). However,
many of these prior studies do not characterize the nature of these retrieved memories,
making it difficult to extrapolate the exact role of episodic memory in decision-making.
While these research fields indicate that memory and choice are intimately linked (Johnson
et al., 2007; Elke U. Weber et al., 1995; Ludvig, Madan, & Spetch, 2015; Madan et al.,
2014), little is known about the quality of these memories, or the detailed nature of how they
shape value based choice. Given the challenges of testing episodic memory in the context of
repeated exposures, it is more efficacious to investigate episodic memory processes during
decision-making by exploring choices based on a single prior experience. This approach
harks back to more classic models of decision-making where single shot games were
considered the archetypal mode of understanding behavior (Camerer, 2003). Indeed, in many
everyday situations, individuals do not have the luxury of using many past experiences to
guide choice. Instead, individuals routinely make decisions with limited prior information,
where they have only experienced one previous relevant episode.
Based on previous work from the fields of episodic memory, decision-making, and learning,
there are two candidate mechanisms that may underpin value-based choice. One possibility
is that value-based choice does not rely on associative memory. In this case, only a simple
representation of the previous experience is retrieved without any specific details (Schacter
& Tulving, 1994). In line with this, a large literature on impression formation suggests that
individuals infer the value of an individual or object without retrieving any specific details
(Lee & Harris, 2013, Uleman et al, 2005). Indeed, patients with amnesia who have deficits
in associative memory, still have intact impression formation (M. K. Johnson, Kim, & Risse,
1985). An alternative possibility is that at the time of a decision, individuals may retrieve
enriched associative memories that consist of flexible relationships between specific features
of the past experience (Davachi, 2006; Schacter & Tulving, 1994). In this case, retrieved
memories are contextualized, containing complex details that are explicitly remembered
(i.e., value, place, person, etc.). Theoretically, these types of associative memories would
provide a highly informative representation of the previous experience that could optimally
guide future choice. For example, when deciding whether or not to trust someone, it may be
most efficient and beneficial to retrieve the specific actions and consequences of the past
experience with that individual.
Murty et al. Page 3
J Exp Psychol Gen
. Author manuscript; available in PMC 2017 May 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Decomposing whether value-based choice relies on associative-rich memories—as opposed
to less detailed forms of item memory or even no episodic memory—will help characterize
the dynamic processes and interplay between memory and choice. Our aim here was three-
fold: first, to identify and characterize whether individuals encode value representations into
episodic memory. Second, to explore whether these episodic memories subserve value based
decisions. And third, to test if this mechanism is deployed across contexts, including during
simple, non-social choices and during more complex, social choices. In the current study,
participants performed a decision-making task where they first encountered trial-unique
lotteries that varied in their reward outcomes (‘Reward Task’, Figure 1). After a short delay,
participants completed a decision-making task in which they decided between selecting a
previously encountered lottery or a new, never before seen lottery. Adaptive decisions are
indicated by selecting lotteries that were previously associated with high rewards more often
than those associated with low rewards. To probe episodic memory representations for these
lotteries, participants completed a surprise memory test for the lottery and their associated
value. We predicted that individuals would make adaptive choices during the decision-
making phase (i.e., choose high value and avoid low value), and that this would be related to
associative memory for the lotteries and their specific outcomes. We further test whether
these predictions generalize to a more complex, ecologically valid social context where
participants dyadically interact with other participants.
Experiment 1
METHODS
Participants—Participants were recruited from New York University and the surrounding
New York City community. Informed consent was obtained from each participant in a
manner approved by the University Committee on Activities Involving Human Subjects.
Participants were paid an initial $10 and an additional monetary bonus accrued during the
task. Experiment 1 included 30 participants (18 females; mean age 23.6±4.0 SD). Sample
size was determined by pilot data (not presented here) demonstrating that N=30 was
sufficient to procure significant results.
Stimuli—Stimuli for Experiment 1 consisted of 120 gray-scale images of houses (Krebs,
Boehler, De Belder, & Egner, 2015). We randomized the presentation of stimuli appearing in
the Reward, Decision, and Memory tasks across participants.
Task Design—In Experiment 1, participants completed 4 tasks in the following order:
Reward, Distracter, Decision, and Memory. During the Reward task, participants were
instructed that they would be playing 60 trials of a game involving 60 independent lotteries
which could pay up to $5 (see supplemental methods for details). They were told that each
lottery would be denoted by a particular trial-unique, house stimulus. Participants were
instructed to pay attention to the houses they encountered because they might encounter
them later in the experiment. Prior to starting the experiment, participants were required to
correctly answer questions about the nature of the task to ensure comprehension.
Murty et al. Page 4
J Exp Psychol Gen
. Author manuscript; available in PMC 2017 May 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Reward Task—On each trial of the Reward task, participants observed for 1.5 seconds a
screen that read, “Lottery is being generated” alongside a picture of a trial-unique house (Fig
1a). The text on the top of the screen was then replaced with the reward outcome associated
with that lottery for 3 seconds. Finally, the participants had up to 3.5 seconds to indicate how
they felt about the outcome of the lottery (1 = Good, 2 = Neutral, 3 = Bad). Following their
response, a fixation cross was shown with a jittered ITI varying between 1–3 seconds.
Lottery outcomes on each trial were randomly selected without replacement from a range of
values between $0.10-$1.50 (low value) and $3.60-$5.00 (high value), each in 10 cent
increments. Participants were told that one trial would be randomly selected to be paid out at
the end of the experiment.
Distractor Task—Following the Reward task, participants completed a Distracter task—a
10-minute task consisting of various math problems—to introduce a short delay before the
Decision and Memory tasks (Duncan, Tompary, & Davachi, 2014). On each trial of the
Distracter task, participants were asked to solve a math problem, the solutions to which were
all single digit responses. Subjects were given 9 seconds to solve each problem and were
instructed to use the number keys on the keyboard to indicate their response. Participants did
not receive feedback on their performance for this task. All participants had above chance
performance on the Distracter task.
Decision Task—After the Distracter task, participants completed the Decision task. In the
Decision task participants were instructed that they would see two house stimuli
representing two different lotteries (Fig 1a) and would have to decide which lottery they
would prefer to play. On each trial, participants were presented with a pair of images and
asked to choose between a trial-unique house and a schematic line drawing of a house (this
house schematic was the same on every trial). The trial unique house was either a previously
encountered house (from the Reward task) or novel, never before seen house. The schematic
line drawing of a house was indicative of a new lottery that would be selected at random. In
other words, on each trial, participants could select between a house stimulus (i.e. a
previously encountered stimulus or completely novel stimulus) and a schematic line drawing
of a house (see supplement methods for more details). Participants had 4 seconds to respond
and completed 90 trials. The trial-unique houses were selected randomly without
replacement from the 60 houses presented during the reward task and 30 never before seen
houses. During this task, participants only selected which lottery they preferred to play, and
feedback about the lottery value was never given. This allowed us to gather information on
individual preferences for lotteries.
Memory Task—Finally, participants completed a surprise Memory task, where we probed
item memory (i.e., recognition memory for houses) and associative memory (i.e., memory
for the value associated with individual houses). On each trial, participants were shown a
picture of a house, which was either previously shown during the Reward task (i.e., old) or
was never shown (i.e., new). To probe item memory, participants had to indicate whether
they had seen the house during the Reward task including their confidence (1 = high
confidence old, 2 = low confidence old, 3 = not sure, 4 = low confidence new, 5 = high
confidence new). To probe associative memory, if participants responded with a 1, 2, or 3,
Murty et al. Page 5
J Exp Psychol Gen
. Author manuscript; available in PMC 2017 May 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript