RESEARCH REPORT
When and Why a Failed Test Potentiates the Effectiveness of
Subsequent Study
Matthew Jensen Hays
University of Southern California
Nate Kornell
Williams College
Robert A. Bjork
University of California, Los Angeles
Teachers and trainers often try to prevent learners from making errors, but recent findings (e.g., Kornell,
Hays, & Bjork, 2009) have demonstrated that tests can potentiate subsequent learning even when the
correct answer is difficult or impossible to generate (e.g., “What is Nate Kornell’s middle name?”). In
3 experiments, we explored when and why a failed test enhances learning. We found that failed tests
followed by immediate feedback produced greater retention than did a presentation-only condition.
Failed tests followed by delayed feedback, by contrast, did not produce such a benefit— except when the
direction of the final test was reversed (i.e., the participants were provided with the target and had to
produce the original cue). Our findings suggest that generating an incorrect response to a cue both
activates the semantic network associated with the cue and suppresses the correct response. These
processes appear to have 2 consequences: If feedback is presented immediately, the semantic activation
enhances the mapping of the cue to the correct response; if feedback is presented at a delay, the prior
suppression boosts the learning of the suppressed response.
Keywords: spacing, testing, retrieval, forgetting, errorless learning
Tests are perhaps the most ubiquitous element of formal edu-
cation. For hundreds of years, almost every lecture, lab, and
seminar has concluded with a test. These criterion tests are in-
tended to diagnose the knowledge or ability possessed by the
student or trainee. In the past century, however, researchers have
discovered that tests do more than diagnose; they are potent
learning events (e.g., Gates, 1917; Spitzer, 1939). That is, tests
serve to create and/or strengthen associations between information
in memory. Indeed, successfully retrieving information during
learning often creates stronger memories than does re-studying
(e.g., Allen, Mahler, & Estes, 1969). This testing effect has been
demonstrated in a variety of domains, including children’s picture
naming (Wheeler & Roediger, 1992), high-school students’ com-
prehension of history lessons (Nungester & Duchastel, 1982),
college students’ comprehension of cognitive psychology course
material (Leeming, 2002), and undergraduate students’ recollec-
tion of idea units from prose passages (Roediger & Karpicke,
2006).
Nevertheless, tests are rarely used to improve learners’ compre-
hension or competence in mainstream education (e.g., Glover,
1989). The threat of a test is often used to encourage students to
study more (but with little effect on learning; see Haynie, 1997).
Educators may be reluctant to employ tests as learning events for
fear of the effects of incorrect responses. This fear is not without
substance. Butler and Peterson (1965) found that errors on tests
can be “stamped in,” meaning that producing an incorrect response
will cause that same incorrect response to be produced on later
tests (but see Metcalfe & Kornell, 2007). These findings reinforced
the emerging “errorless learning” movement (e.g., Terrace, 1963),
which posited that errors weakened instruction and created un-
wanted by-products of the training process. Indeed, errorless learn-
ing does appear to benefit clinical populations (e.g., Kern, Liber-
man, Kopelowicz, Mintz, & Green, 2002; Kessels & de Haan,
2003).
In normal classrooms and training settings, however, research
suggests that tests should still be used— even if students some-
times respond incorrectly (e.g., Marsh, Roediger, Bjork, & Bjork,
2007; Pashler, Zarow, & Triplett, 2003). Failed tests, it seems, do
not overwrite previously learned information or otherwise corrupt
cognition. Further, failed tests do not reduce the value of later
learning. On the contrary, Kornell et al. (2009) found that presen-
This article was published Online First May 14, 2012.
Matthew Jensen Hays, Institute for Creative Technologies, University of
Southern California; Nate Kornell, Department of Psychology, Williams
College; Robert A. Bjork, Department of Psychology, University of Cali-
fornia, Los Angeles.
Grant 29192G from the McDonnell Foundation supported this research.
We thank the members of CogFog for their suggestions and interpretations
of the findings reported in this article. We especially thank Barbara
Knowlton and John Nestojko for critical methodological suggestions
throughout our research.
Correspondence concerning this article should be addressed to Matthew
Jensen Hays, USC–ICT, 12015 Waterfront Drive, Los Angeles, CA 90094.
E-mail: matt@hayslab.com
Journal of Experimental Psychology: © 2012 American Psychological Association
Learning, Memory, and Cognition
2013, Vol. 39, No. 1, 290 –296
0278-7393/12/$12.00 DOI: 10.1037/a0028468
290