Retrieval practice protects memory
against acute stress
Amy M. Smith,* Victoria A. Floerke, Ayanna K. Thomas
More than a decade of research has supported a robust consensus: Acute stress
impairs memory retrieval. We aimed to determine whether a highly effective learning
technique could strengthen memory against the negative effects of stress. To bolster
memory, we used retrieval practice, or the act of taking practice tests. Participants
first learned stimuli by either restudying or engaging in retrieval practice. Twenty-four hours
later, we induced stress in half of the participants and assessed subsequent memory
performance. Participants who learned by restudying demonstrated the typical
stress-related memory impairment, whereas those who learned by retrieval practice
were immune to the deleterious effects of stress. These results suggest that the effects
of stress on memory retrieval may be contingent on the strength of the memory
The effects of experimentally induced stress on memory have been studied for more than a decade (1–7). The results support a robust consensus that stress impairs mem- ory retrieval (8). These studies used a common method whereby participants learn words
or images and return 24 hours later for a memory test. Before testing, psychosocial stress is
induced. Critically, the memory test is administered ~25 min after stress introduction [for
exceptions, see (5, 6)], when the stress hormone
cortisol reaches peak poststress levels in the
blood. Researchers have primarily examined
memory after this delay because cortisol has
been shown to affect brain regions that are implicated in memory retrieval (9).
Previous research on this topic has not been
expressly concerned with the quality of encoding during initial learning. Before encoding,
participants were typically instructed to “
memorize” stimuli. However, the processes that take
place at encoding influence memory representation and accessibility (10). Without guidance
as to how to approach learning material, participants may choose ineffective encoding strategies, resulting in unstable memory representations.
Many participants in these studies likely chose
to learn by rereading, given that this method is
often reported as the most popular study strategy (11). Rereading is a poor learning strategy,
insofar as it creates relatively weak memory
representations (12). Thus, it is unclear whether
all memories are subject to the detrimental effects of stress, or whether only weakly encoded
representations are vulnerable.
In our experiment, we addressed this by
strengthening memory at encoding through the
use of retrieval practice, the act of taking practice
tests. Among a host of options for study techniques, we chose retrieval practice for two reasons.
First, retrieval practice has consistently yielded
long-term memory retention that is equal to or
better than restudying (13–15) and a plethora of
other learning strategies such as mental imagery
(12), concept mapping (16), and the keyword
mnemonic (17). Thus, we chose to use retrieval
practice as an encoding technique because it
had the most potential to create memories that
were resilient to stress. Second, retrieval prac-
tice is an easily implemented learning strategy
(12). We reasoned that if retrieval practice was
successful at creating stress-resistant memories,
our findings could be readily applied in real-
world scenarios (e.g., test anxiety).
A second limitation of previous research on
stress and memory concerns the timing of the
final memory test. Researchers typically assessed
memory 25 min after stress induction and found
detrimental effects. However, contesting the con-
sensus that stress generally impairs retrieval,
recent research showed that participants who
were tested immediately after stress induction
exhibited memory performance that was better
than or comparable to a no-stress control group
(5, 6). Thus, a secondary aim of our study was
to investigate the potentially facilitative effects
of the immediate stress response in the context
of a retrieval practice encoding manipulation.
In our experiment, 120 participants studied
either 30 concrete nouns or 30 images of nouns,
one item at a time. Half of the items in each list
were of negative valence and half were of neu-
tral valence. Whether words or images were
studied first was counterbalanced. Sixty par-
ticipants then engaged in study practice (SP),
in which they restudied the 30 items. The other
60 participants engaged in retrieval practice
(RP), in which they recalled as many items as
they could remember. RP participants were not
given feedback on the free recall test or on any
subsequent tests. This procedure (item presen-
tation followed by restudy or free recall) was
then repeated for the 30 items of the other
type. Afterward, SP participants restudied all
60 stimuli, whereas RP participants attempt-
ed to recall the words and images in any order.
After a short distractor task, SP participants
again restudied all 60 items, and RP participants
attempted to recall all items.
Twenty-four hours later, 30 SP and 30 RP
participants underwent stress induction, and
30 SP and 30 RP participants completed a time-matched nonstressful task. Our encoding and
stress manipulations were fully crossed, so there
were four between-subject groups: nonstressed
SP, stressed SP, nonstressed RP, and stressed RP.
During stress induction, participants gave extemporaneous speeches and solved math problems in front of two judges and three peers (18).
1046 25 NOVEMBER 2016 • VOL 354 ISSUE 6315 sciencemag.org SCIENCE
Tufts University, 490 Boston Avenue, Medford, MA 02155, USA.
*Corresponding author. Email: email@example.com
Fig. 1. Average number of items accurately recalled on tests 1 and 2. Test 1 was administered
immediately after the onset of stress. Test 2 followed after a 25 min delay. Retrieval practice (RP)
refers to the learning technique in which participants study stimuli and take three subsequent recall
tests. Study practice (SP) refers to the learning technique in which participants study stimuli four
times. Tests occurred on the day after learning. Error bars represent standard errors of the mean.
*P < 0.05; **P < 0.01; ***P < 0.001.