Why Testing Improves Memory:Mediator Effectiveness HypothesisMary A. Pyc* and Katherine A. Rawson

Anintuitive but incorrect assumption is thatlearning only occurs during study and thattesting is useful only for evaluating the

state of memory (1). However, testing improvesmemory, as demonstrated with a wide variety ofmaterials and learners (2). Given the wealth ofresearch establishing the empirical benefits of test-ing, it is surprising that the mechanisms underly-ing testing effects are not well understood.

We propose the mediator effectiveness hy-pothesis, stating that testing improves memoryby supporting the use of more-effective media-tors during encoding (a mediator is a word,phrase, or concept that links a cue to a target).What makes a mediator effective? Two key fac-tors are mediator retrieval (i.e., mediator is re-callable when prompted with the cue) andmediator decoding (i.e., mediator elicits the targetfrom memory) (3). The mediator effectivenesshypothesis assumes that mediators generatedduring testing (versus restudy only) are morelikely to be subsequently retrieved and decoded,increasing recall of target responses.

To test this hypothesis, we presented118 participants with 48 Swahili-English translation pairs (e.g., wingu-cloud) for an initial study trial and thenthree blocks of practice trials. In the test-restudy group, each practice trial for anitem involved a cued recall test followedimmediately by restudy. In the restudygroup, each trial involved only restudy.On the initial study trial and each re-study trial, all participants generatedand reported keyword mediators (key-words are mediators that look or soundsimilar to the foreign language cue andare semantically related to the Englishtarget) (4). On the final test 1 weeklater, participants received only the cue(cue only, C group, as in prior re-search), the cue with their own medi-ator from practice (cue plus mediator,CMgroup), or the cuewith a prompt torecall their own mediator before recall-ing the target (cue plus mediator recall,CMR group).

The mediator effectiveness hypoth-esis makes two key predictions. Con-cerning mediator retrieval, retrieval ofmediators at the final test in the CMRgroup will be greater after test-restudyversus restudy practice. Concerningmediator decoding, retrieval of targets

at final test in the CM group will be greater aftertest-restudy versus restudy practice.

Replicating the robust testing effects from priorresearch, test-restudy practice produced almost athreefold increase in final test performance in theC group (Fig. 1A). According to the mediator ef-fectiveness hypothesis, this benefit is due to dif-ferential effectiveness of mediators, with testingimproving mediator retrieval and decoding.

Confirming the prediction of increased me-diator retrieval, recall of mediators at final testin the CMR group was greater after test-restudyversus restudy practice (51% versus 34%). Forconverging evidence, consider final test perform-ance in the C and CM groups. Providing me-diators at final test significantly improved recallafter restudy practice but not after test-restudypractice, suggesting that explicitly providingme-diators was largely redundant with participants’spontaneous retrieval of mediators in the test-restudy group.

Confirming the prediction of increased medi-ator decoding, final test performance in the CM

group was significantly greater after test-restudyversus restudy practice. Even when differences inmediator retrieval were circumvented by provid-ing participants with their mediators, mediatorswere more likely to elicit recall of targets aftertest-restudy versus restudy practice. As converg-ing evidence, the same pattern is apparent whenexamining recall as a function of mediator re-trieval in the CMR group (Fig. 1B). Althoughmediator retrieval benefited recall in both groups,the benefit was greater after test-restudy versusrestudy practice. Furthermore, in trials in whichmediators were retrieved, recall was greater aftertest-restudy versus restudy practice.

Results support the mediator effectivenesshypothesis and offer one theoretical explana-tion for why testing is beneficial for memory:Mediators generated during encoding are moreeffective (i.e., more likely to be retrieved anddecoded) with test-restudy versus restudy prac-tice. We are not claiming that mediator effective-ness is the only mechanism underlying testingeffects. However, mediator effectiveness may bean important contributor.

Why did testing yield more-effective medi-ators? Successfully retrieving mediators duringpractice may enhance their memory strength.Additionally, retrieval failures during encodingmay promote shifting from less- to more-effectivemediators (5), and retrieval failure occurs dur-ing testing but not restudy. Consistent with thisidea, shifting to new keywords was more likelyduring test-restudy versus restudy practice (25%versus 19% of trials). Importantly, the mediatoreffectiveness hypothesis defines two compo-nents of mediator effectiveness (i.e., mediatorretrieval and decoding), and results provide evi-dence for a contribution of each of these factorsto the testing effect.

References and Notes1. J. D. Karpicke, H. L. Roediger 3rd, Science 319,

966 (2008).2. H. L. Roediger, J. D. Karpicke, Perspect. Psychol. Sci.

1, 181 (2006).3. J. Dunlosky, C. Hertzog, A. Powell-Moman, Dev. Psychol.

41, 389 (2005).4. Materials, methods, and keyword example are available

as supporting material on Science Online.5. H. P. Bahrick, L. K. Hall, J. Mem. Lang. 52, 566

(2005).6. Research supported by a Collaborative Award from the

James S. McDonnell Foundation 21st Century ScienceInitiative in Bridging Brain, Mind and Behavior.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/330/6002/335/DC1Materials and MethodsReferences

26 April 2010; accepted 6 August 201010.1126/science.1191465

BREVIA

Department of Psychology, Kent State University, Kent, OH44242, USA.

*To whom correspondence should be addressed. E-mail:mpyc@kent.edu

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

mediator retrieved mediator not retrieved

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

C CM CMR

restudy onlytest-restudy

B

tseT laniF no dellace

R yltcerroC s

metI fo noitroporP

A

Fig. 1. (A) Final recall by group. (B) Final recall in CMR groupas a function of whether mediators were retrieved beforeattempting target recall. Error bars represent standard errors.

www.sciencemag.org SCIENCE VOL 330 15 OCTOBER 2010 335

on

Nov

embe

r 11

, 201

4w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from