Psychological Bulletin Copyright 2006 by the American Psychological Association

2006, Vol. 132, No. 3, 354 –380 0033-2909/06/$12.00 DOI: 10.1037/0033-2909.132.3.354

Distributed Practice in Verbal Recall Tasks:

A Review and Quantitative Synthesis

Nicholas J. Cepeda Harold Pashler, Edward Vul, and John T. Wixted

University of California, San Diego and University of California, San Diego
University of Colorado at Boulder

Doug Rohrer
University of South Florida
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

The authors performed a meta-analysis of the distributed practice effect to illuminate the effects of
temporal variables that have been neglected in previous reviews. This review found 839 assessments of
distributed practice in 317 experiments located in 184 articles. Effects of spacing (consecutive massed
presentations vs. spaced learning episodes) and lag (less spaced vs. more spaced learning episodes) were
examined, as were expanding interstudy interval (ISI) effects. Analyses suggest that ISI and retention
interval operate jointly to affect final-test retention; specifically, the ISI producing maximal retention
increased as retention interval increased. Areas needing future research and theoretical implications are
discussed.

Keywords: spacing effect, distributed practice, meta-analysis, interstudy interval, retention interval

In the late 1800s, researchers began to demonstrate benefits Terminology

from distributed practice (Ebbinghaus, 1885/1964; Jost, 1897;
Thorndike, 1912). Since then, the topic of temporal distribution of The distributed practice effect refers to an effect of interstudy
practice has become one of the mainstays of learning and memory interval (ISI) upon learning, as measured on subsequent tests. ISI
research. Recent reviews have suggested that a benefit from dis- is the interval separating different study episodes of the same
tributed practice is often found both for verbal memory tasks, such materials. In the most typical spacing study, there are two study
as list recall, paired associates, and paragraph recall (Janiszewski, episodes separated by an ISI and some retention interval separating
Noel, & Sawyer, 2003), and for skill learning, such as mirror the final study episode and a later test. Generally, the retention
tracing or video game acquisition (Donovan & Radosevich, 1999). interval is fixed, and performance is compared for several different
The size of the distributed practice effect is often large. In spite of values of the ISI. In studies with more than two study episodes,
abundant evidence for distributed practice benefits, a number of retention interval still refers to the interval between the last of
empirical studies (e.g., Toppino & Gracen, 1985; Underwood, these study episodes and the final test.
1961; Underwood & Ekstrand, 1967) and a recent review of the When the study time devoted to any given item is not subject to
literature (Donovan & Radosevich, 1999) concluded that longer any interruptions of intervening items or intervening time, learning
spacing and/or lag intervals sometimes failed to benefit retention. is said to be massed (i.e., item A stays on the screen for twice as
The present review explores the effects of distribution of practice long as it would for a spaced presentation, without disappearing
upon retention of verbal information and seeks to elucidate the between presentations or disappearing for less than 1 s, such as the
conditions under which distributed practice does and does not length of time it takes a slide projector to change slides). In
benefit retention. contrast, learning is spaced or distributed when a measurable time
lag (1 s or longer) separates study episodes for a given item—that
is, either (a) item A appears, item A disappears for some amount
of time, and then item A reappears or (b) item A appears, item A
Nicholas J. Cepeda, Department of Psychology, University of Califor- disappears, item B (item C, etc.) appears and disappears, and then
nia, San Diego, and Department of Psychology, University of Colorado at item A reappears. For example, if a list of 20 items is presented
Boulder; Harold Pashler, Edward Vul, and John T. Wixted, Department of twice, and there are no delays between each consecutive presen-
Psychology, University of California, San Diego; Doug Rohrer, Depart- tation of the list, learning episodes for any given item are spaced
ment of Psychology, University of South Florida. (on average) by 20 items, and this would be described as spaced
This work was supported by the Institute of Education Sciences (U.S. learning. Learning is considered to be massed only when presen-
Department of Education Grants R305H020061 and R305H040108). We
tations of a given item in a list are separated by 0 items and a time
thank Jean Trinh for obtaining articles. We also thank Kelly Braun, Jane
Childers, Michael Kahana, and Phil Pavlik for providing raw data. Finally,
lag of less than 1 s. During massed learning, the participant sees a
we thank Derek Briggs for comments on the article and statistical advice. single presentation of the item for twice the presentation time of a
Correspondence concerning this article should be addressed to Nicholas comparable spaced item. The term spacing effect refers to en-
J. Cepeda, University of Colorado at Boulder, Department of Psychology, hanced learning during spaced as compared with massed study
345 UCB, Boulder, CO 80309-0345. E-mail: ncepeda@psy.ucsd.edu episodes for a given item. In contrast, the term lag effect refers to

354
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 355

comparisons of different levels of spacing, either differing num- than or greater than 1 day). The largest effect sizes were seen in
bers of items (e.g., Thios & D’Agostino, 1976) or differing low rigor studies with low complexity tasks (e.g., rotary pursuit,
amounts of time (e.g., Tzeng, 1973). We use the generic term typing, and peg reversal), and retention interval failed to influence
distributed practice to encompass both spacing and lag effects, effect size. The only interaction Donovan and Radosevich exam-
without distinguishing between them. ined was the interaction of ISI and task domain. It is important to
As noted above, studies of distributed practice must include at note that task domain moderated the distributed practice effect;
least two, but may include more than two, learning episodes. When depending on task domain and lag, an increase in ISI either
three or more learning episodes are presented, the ISIs may be increased or decreased effect size. Overall, Donovan and Rados-
equal (fixed), progressively longer (expanding), or progressively evich found that increasingly distributed practice resulted in larger
shorter (contracting). effect sizes for verbal tasks like free recall, foreign language, and
verbal discrimination, but these tasks also showed an inverse-U
Past Quantitative Reviews function, such that very long lags produced smaller effect sizes. In
contrast, increased lags produced smaller effect sizes for skill tasks
The literature on distributed practice is vast, and the topic has like typing, gymnastics, and music performance. Thus, the current
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

been qualitatively reviewed in a number of books and articles (e.g., article is the first review article to suggest that distributed practice
Crowder, 1976; Dempster, 1989; Greene, 1992; McGeoch & Irion, intervals can become too long, regardless of task domain. Their
1952; Ruch, 1928). Quantitative reviews are fewer in number: analysis omitted many articles that met their inclusion criteria (by
Four major quantitative reviews of distributed practice appear to our count, at least 55 articles that were published before 1999), and
exist (Donovan & Radosevich, 1999; Janiszewski, Noel, & Saw- only about 10% of their sample used verbal memory tasks.
yer, 2003; T. D. Lee & Genovese, 1988; Moss, 1996). The authors Janiszewski et al. (2003) performed the most extensive exami-
of these articles all concluded that distributed practice produces an nation of distributed practice moderators to date; they focused on
overall increase in retention, and they argued that the effect is 97 articles from the verbal memory task literature. Five factors
moderated by several important variables. This section summa- failed to influence effect size: verbal versus pictorial stimuli, novel
rizes each of these reviews and highlights some of the questions versus familiar stimuli, unimodal versus bimodal stimulus presen-
that remain unanswered. tation (e.g., auditory vs. auditory plus visual), structural versus
Moss (1996) reviewed 120 articles on the distributed practice semantic cue relationships, and isolated versus context-embedded
effect, across a wide range of tasks. She partitioned data by age of stimuli. Five factors influenced effect size magnitude: lag (longer
participant and type of material (verbal information, intellectual ISIs increased effect size), stimulus meaningfulness (meaningful
skills, or motor learning). For each study, Moss determined the stimuli showed a larger effect size than nonmeaningful stimuli),
direction of effect, if any. She concluded that longer ISIs facilitate stimulus complexity (semantically complex stimuli showed a
learning of verbal information (e.g., spelling) and motor skills larger effect size than structurally complex or simple stimuli),
(e.g., mirror tracing); in each case, over 80% of studies showed a learning type (intentional learning produced a larger effect size
distributed practice benefit. In contrast, only one third of intellec- than incidental learning), and complexity of intervening material
tual skill (e.g., math computation) studies showed a benefit from (intervening material that was semantically complex led to a larger
distributed practice, and half showed no effect from distributed effect size than intervening material that was structurally complex
practice. or simple). Unfortunately, Janiszewski et al. did not examine
T. D. Lee and Genovese (1988) reviewed 47 articles on distrib- retention interval effects. Even though they focused on verbal
uted practice in motor skill learning. Distributed practice improved memory tasks, there is only partial overlap between the articles
both acquisition and retention of motor skills. (Acquisition refers to used in Janiszewski et al.’s meta-analysis and those used in the
performance on the final learning trial, and retention refers to present meta-analysis (47 articles were used in both). Partial over-
performance after a retention interval.) T. D. Lee and Genovese’s lap occurred in part because Janiszewski et al. chose to include
findings dispute those of a prior review by Adams (1987; see also studies that used reaction time, frequency judgments, and recog-
Doré & Hilgard, 1938; Irion, 1966). Adams’s review concluded nition memory as final-test learning measures, whereas we did not.
that distributed practice has little or no effect on acquisition of
motor skills. In the 1960s, Hull’s (1943) learning theory was Summary of Past Quantitative Reviews
shown to poorly account for existing data. Adams suggests that
this discovery caused most researchers to stop studying the effects In summary, quantitative syntheses of the temporal distribution
of distributed practice on motor learning. In contrast to Adams’s of practice literature have suggested that a benefit from longer ISIs
claims and the 1960s negation of Hull’ theory, both T. D. Lee and is a fairly robust effect. Beyond that, however, few firm conclu-
Genovese’s (1988) review and Hull’s theory suggested that dis- sions seem warranted. For example, Donovan and Radosevich’s
tributed practice should improve motor learning. (1999) review suggested that increasingly distributed practice im-
In their meta-analysis of the distributed practice literature, Don- pairs learning, seemingly counter to Janiszewski et al.’s (2003)
ovan and Radosevich (1999) inspected 63 articles that used a wide review, which concluded that increasingly distributed practice
range of tasks. They examined the effects of several moderators: improved retention. Upon closer observation of Donovan and
methodological rigor (on a 3-point scale), mental requirements Radosevich’s findings, skill acquisition studies showed decreased
(low or high, based on whether “mental or cognitive skills” [p. final-test learning with longer ISIs, and verbal memory tasks
798] were required for task performance), overall complexity (low, showed nonmonotonic effects of ISI on final-test learning (final-
average, or high, based on the “number of distinct behaviors” [p. test performance improved as ISI increased from a few minutes to
798] required to perform the task), ISI (less than 1 min, 1–10 min, an hour and decreased as ISI reached 1 day or longer). Donovan
10 min–1 hr, and greater than 1 day), and retention interval (less and Radosevich’s review suggested that retention interval has no
 356 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

effect on the magnitude of the distributed practice effect. This learning episodes. We assessed joint effects of ISI and retention
conclusion is at variance with a number of individual experimental interval by examining ISI effects separately for a number of
findings (e.g., Balota, Duchek, & Paullin, 1989; Bray, Robbins, & different retention intervals. Final-test performances following ex-
Witcher, 1976; Glenberg, 1976; Glenberg & Lehmann, 1980; see panding versus fixed ISIs also were compared. In addition to
Crowder, 1976, for a useful discussion). Notably, Donovan and providing additional clarity on the temporal variables just de-
Radosevich failed to include in their meta-analysis many studies scribed, another goal of the present study was to pinpoint, for
that showed retention interval effects. Even though distributed future research, important areas in which present distributed prac-
practice benefits are robust, temporal moderators affect distributed tice knowledge is severely limited. Although the literature on
practice through a complex interplay of time and task. distributed practice is indeed very large, the present review dis-
Given the heterogeneity of studies included in prior syntheses, closes (in ways that previous reviews have not) how sorely lacking
the omission of relevant studies, and the disparate conclusions of it is in the very sorts of information that are most needed if serious
these syntheses, one might wonder whether they paint an accurate practical benefits are to be derived from this century-long research
composite picture of the literature as a whole. In addition, prior tradition.
syntheses have examined the joint impact of ISI and retention We restricted our analysis to verbal memory tasks, in the broad-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

interval in a cursory fashion. If there is a complex interplay est sense. These have been used in by far the greatest number of
between ISI and retention interval, as some of the experimental studies of distributed practice (Moss, 1996). This restriction was
studies cited in the previous paragraph would suggest, then this is introduced because of the enormous heterogeneity of tasks and
likely to be of substantial import both for practical applications and performance measures used in the remainder of the distributed
for theoretical issues. The practical relevance is obvious: One can practice literature. It seemed unlikely that the literature would
hardly select an ISI that optimizes instruction unless one knows allow meaningful synthetic conclusions to be drawn from any
how learning depends upon ISI; if that function varies with reten- other single category of tasks or studies. Unlike previous review-
tion interval, this too must be considered in designing the most ers, we restricted our review to studies using recall as a perfor-
efficient procedures for pedagogy or training. Theories of the mance measure; we did not review studies that used performance
distributed practice effect are incomplete unless they can account measures like recognition or frequency judgments. To address
for joint effects of ISI, retention interval, and task. potential relearning confounds, we examined the effects of pro-
viding different numbers of learning trials during the second
Learning and Relearning Confounds session.

One potentially critical factor that has been overlooked in past Method
quantitative reviews of the distributed practice effect—potentially
undermining many of the conclusions drawn—is the highly vari- Literature Search
able choice of training procedures used in the second and subse-
Articles included in this analysis were selected by Nicholas J. Cepeda
quent learning sessions. In many studies, including some deserv-
using several sources. Lists of potential articles were given to Nicholas
edly well-cited research in this area (e.g., Bahrick, 1979; Bahrick J. Cepeda by Harold Pashler, Edward Vul, John T. Wixted, and Doug
& Phelps, 1987), participants were trained to a criterion of perfect Rohrer, on the basis of past literature searches for related studies.
performance on all items during the second and subsequent learn- PsycINFO (1872–2002) and/or ERIC (1966 –2002) were searched with a
ing sessions. With this procedure, an increase in ISI inevitably variety of keywords. A partial list of keyword searches includes “spacing
increases the amount of training provided during the second or effect,” “distributed practice,” “spac* mass* practice,” “spac* mass*
subsequent sessions. (This is because a longer ISI results in more learning,” “spac* mass* presentation,” “spac* mass* retention,” “mass*
forgetting between training sessions, thus necessitating a greater distrib* retention,” “spac* remem*,” “distrib* remem*,” “lag effect,”
number of relearning trials to reach criterion.) Thus, in designs that “distrib* lag,” “distrib* rehears*,” “meta-analysis spacing,” and “review
have this feature, distribution of practice is confounded with the spacing.” Portions of article titles were entered as keywords into searches
in these databases, and the resulting article lists were examined for poten-
amount of practice time during the second (and subsequent) ses-
tial articles. Primary authors were entered into PsycINFO searches, and
sions. This makes it impossible to know whether differences in their other articles were examined for relevance. Reference lists of all
final-test performance reflect distributed practice effects per se. To potential articles were examined for references to other potential studies.
avoid this confound, the number of relearning trials must be fixed. Reference lists from previous quantitative reviews (Donovan & Radosev-
(Either training to a criterion of perfect performance during the ich, 1999; Janiszewski et al., 2003; Moss, 1996) were examined. Internet
first learning session or providing a fixed number of learning trials searches were carried out (through http://www.google.com/) with the key-
during the first learning session and then presenting items, with words “spacing effect” and “distributed practice.” Current and older un-
feedback, a fixed number of times during the second and subse- published data were requested from researchers who (in our opinion) might
quent learning sessions seems to us a reasonable way to equalize be conducting distributed practice research or who might have older
initial learning without introducing a relearning confound.) unpublished data.

Inclusion Criteria
Current Meta-Analysis
Studies had to meet several criteria to be included. The material must
Our goal in the present article is to perform a quantitative have been learned during a verbal memory task (most commonly, paired-
integrative review of the distributed practice literature, tailored to associates/cued recall, list recall, fact recall, or paragraph recall; also, text
shed light on the critical temporal and procedural variables dis- recall, object recall, sentence recall, spelling, face naming, picture naming,
cussed above. To examine ISI effects, we examined the degree of and category recall). A recall test must have assessed performance at the
benefit produced by shorter and longer temporal gaps between time of final test. The experiment must have provided two or more learning
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 357

opportunities for each item (or one learning opportunity of the same Computation of ISI and Retention Interval Joint Effects
temporal length and separated by a lag less than 1 s, for massed items).
Experiments using children and older adults were included (with some To examine the joint effects of ISI and retention interval, we performed
caveats noted below). Studies using clinical populations were excluded. three separate lag analyses. The first lag analysis was designed to mirror
Out of 427 reviewed articles, a total of 317 experiments in 184 articles met the lag analysis performed by Donovan and Radosevich (1999) and
these criteria, providing 958 accuracy values, 839 assessments of distrib- Janiszewski et al. (2003). This analysis does not allow claims about relative
uted practice, and 169 effect sizes. benefits of specific ISIs, for reasons that are described below. The second
lag analysis does allow us to make claims about what specific ISI is
Data Coding optimal at each specific retention interval. The third (qualitative) lag
analysis was designed to dispel concerns about a potential confound
Time intervals were coded in days (e.g., 1 min ⫽ 0.000694 days, and 1 present in the first two lag analyses. In reading the following descriptions
week ⫽ 7 days). ISI and retention interval were computed on the basis of of absolute and difference lag analyses, the reader is referred to Figure 1.
authors’ reports of either the number of items and/or the amount of time Difference lag analyses. The first lag analysis was concerned with the
between learning episodes for a given item. When authors described lags differences in ISI and accuracy that are obtained when adjacent pairwise
in terms of the actual (or in some cases, typical) number of items inter- within-study experimental conditions are compared. For example, Figure 1
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

vening between learning episodes involving a given item, an estimate of shows data from two hypothetical studies. Each study used ISIs of 1 min,
This document is copyrighted by the American Psychological Association or one of its allied publishers.

the time interval was derived. If this estimate could not be derived, usually 1 day, and 2 days. One study used a retention interval of 1 min, and the
either because presentation time for items was not given or because there other study used a retention interval of 7 days. In performing difference lag
was too much variability in the number of items between learning episodes, analyses, we computed between-condition accuracy differences by sub-
the data were excluded. When an experimental procedure employed a list tracting the accuracy for the next shorter ISI from the accuracy value for
presentation, retention interval varied with serial position; thus, retention the longer ISI: For each adjacent ISI pair from each study, accuracy
interval might be 10 s for one item and 1 min for another item. Because of difference ⫽ longer ISI accuracy ⫺ next short ISI accuracy. Likewise, the
this confound, we have reanalyzed the data, separating out list recall and ISI difference was computed in the same way: For each adjacent ISI pair
paired associates studies (see the Appendix). For most analyses, data were from each study, ISI difference ⫽ longer ISI ⫺ next shorter ISI.
separated into relatively small ranges of retention interval (e.g., less than 1 Following the example in Figure 1, the ISIs used in Study 1 were 1 min,
min, 1 min–less than 10 min, 10 min–less than 1 day, 1 day, 2–7 days, 1 day, and 2 days, resulting in two ISI differences. For ISIs of 2 days and
8 –30 days, 31 or more days. In some cases, the necessary temporal and/or 1 day, ISI difference ⫽ 2 days ⫺ 1 day ⫽ 1 day, and for ISIs of 1 day and
accuracy data were not available in the published article, but we were able 1 min, ISI difference ⫽ 1 day ⫺ 1 min ⫽ 1 day. Study 1 also yields two
to obtain these data directly from the study author. For these studies, the accuracy difference values. For ISIs of 2 days and 1 day, accuracy differ-
reader will not be able to calculate ISI, retention interval, and/or accuracy ence ⫽ 50 ⫺ 60 ⫽ ⫺10%, and for ISIs of 1 day and 1 min, accuracy
from the published article.) difference ⫽ 60 ⫺ 90 ⫽ ⫺30%.
As seen in Figure 1, the average accuracy difference value for a retention
Computation of Effect Size interval of 1 min–2 hr and an ISI of 1 day is the mean of these two Study
1 accuracy difference values: ⫺20%. The ISI difference and accuracy
Cohen’s d (Cohen, 1988) was selected as the measure of effect size, difference values for Study 2 are calculated and binned in a similar fashion.
because of its widespread use in the literature. To calculate d, the differ- ISI difference and accuracy difference values were calculated from all
ence in means was divided by the standard deviation. studies in the literature for which both difference values were calculable.
Choice of standard deviation is crucial, as it impacts observed effect size When plotting each data point, we binned that data point with other data
(Glass, McGaw, & Smith, 1981; Taylor & White, 1992). Statisticians differ points using similar or identical ISI and retention interval values. For
on the optimal type of standard deviation to use in computing effect size. example, data points using an ISI of 2 days were averaged with data points
Either control population standard deviation (Morris, 2000; Taylor & using an ISI of 7 days (when their retention intervals were from the same
White, 1992) or various other forms of standard deviation (cf. D’Amico, bin as well).
Neilands, & Zambarano, 2001; Gleser & Olkin, 1994; Johnson & Eagly, We computed effect sizes by dividing each accuracy difference value by
2000; Shadish & Haddock, 1994) are typically used. In this article, stan- the appropriate standard deviation. After this uncorrected effect size was
dard deviation was determined by use of the method advocated by obtained, the corrections described in the Computation of Effect Size
D’Amico et al. (2001), whereby standard deviation at each ISI was calcu- section were performed, when necessary. In many cases, standard deviation
lated, and a simple average was taken across conditions in that experiment. values were not available, and thus there are substantially fewer effect size
Studies that failed to report enough information to calculate this form of data points than there are accuracy difference data points. (By grouping
standard deviation were excluded from effect size analyses. data into ISI bins in this manner, we lost the ability to draw conclusions
In choosing to use this form of standard deviation, we implicitly as- about the relative benefits of specific ISIs. Instead, we were only able to
sumed that experimental conditions had equal variance (Becker, 1988; make claims about the expected accuracy differences that would result if
Cohen, 1988). In reality, variance between conditions is rarely numerically similar experimental manipulations of ISI had been used.)
equal. We feel that the present data adequately approximated this assump- Absolute lag analyses. Because we are interested in the relative ben-
tion, because rarely did variances at different ISIs differ by more than 10%. efits of specific ISIs, we also performed lag analyses on the basis of
As well, most of the data examined here exhibit neither ceiling nor floor absolute accuracy at specific ISIs and retention intervals. To compute
effects, a likely source of unequal variance. absolute lag effects, we first binned data into varying ranges of ISI and
For within-subject experiments, standard deviation was corrected for retention interval. We then averaged the accuracy values from every data
dependence between responses with the equation SDig ⫽ SDws [2(1 – ␳)]1/2 point within each ISI and retention interval bin. Referring again to the
from Morris and DeShon (2002; cf. Cortina & Nouri, 2000; Dunlap, hypothetical data in Figure 1, Study 1 used ISIs of 1 min, 1 day, and 2 days.
Cortina, Vaslow, & Burke, 1996; Gibbons, Hedeker, & Davis, 1993), One accuracy value (the accuracy at ISI ⫽ 1 day; 60% correct) would be
where SDig is the independent groups standard deviation, SDws is the placed into the ISI ⫽ 1 day, retention interval ⫽ 1 min–2 hr bin; another
within-subject standard deviation, and ␳ is the correlation between scores. accuracy value (the accuracy at ISI ⫽ 2 days) would be placed into the
In the current analysis, correction for dependence used the average of all ISI ⫽ 2–28 days, retention interval ⫽ 1 min–2 hr bin. Each study in
pairwise ISI correlations as input to the correction equation. When infor- Figure 1 yields three accuracy values that are grouped into ISI and
mation necessary for this correction was unavailable, these data were retention interval bins. (Note that each study in Figure 1 yielded one
excluded from effect size analyses. accuracy difference value for the difference lag analyses.)
 358 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

between-study comparisons. This was problematic, as overall level of

difficulty often differed substantially between studies. Because we did not
correct for these differences, the overall level of difficulty may not be
equivalent for every bin. Thus, both absolute and difference analyses were
confounded. This confound was present in prior meta-analyses as well.
Because of our concerns about this confound, we performed an additional
analysis, which uses within-study instead of between-studies methods to
determine how optimal ISI changes with retention interval. This third
analysis method does not include the just-described confound.
Within-study lag analyses. As a third method for determining if and
how optimal ISI changes as a function of retention interval, we qualita-
tively examined studies that included an optimal ISI. Studies with an
optimal ISI are those that included at least three different ISI conditions,
wherein one ISI condition had an accuracy value higher than the immedi-
ately shorter ISI and which was immediately followed by a longer ISI
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

condition with an equal or lower accuracy value. Thus, the optimal ISI can
This document is copyrighted by the American Psychological Association or one of its allied publishers.

be described as the shortest ISI that produced maximal retention. We

examined whether these optimal ISIs were longer for longer retention
intervals. (This analysis is subject to some caveats: First, it may be that the
highest accuracy in a study is a local maximum and that another ISI would
have produced higher accuracy had more ISIs been used in the study. The
smaller the range of absolute ISIs used, the greater is this potential
problem. Second, the actual observed optimal ISI varies, as not all ISIs
were tested within a given study. The degree to which the observed optimal
ISI might vary from the truly optimal ISI depends on the distance between
the immediately adjacent ISI values. Even with these caveats, we believe
that this analysis provides a good estimate of optimal ISI.)

Results and Discussion

Analyses examined the joint effects of ISI and retention interval
on final-test retention, as well as the effects of massed versus
spaced learning. We examined joint effects of ISI and retention
interval separately for paired associate and list recall tasks, and we
examined qualitative differences between studies—specifically,
the influence of experimental design, relearning method, and ex-
panding study intervals.

Spacing Effects: Massing Versus Spacing

The spacing effect hinges upon a comparison of massed and
spaced presentations of a to-be-learned item. (As noted above, if a
list of items was presented twice in immediate succession, this was
considered a spaced presentation, because the learning of any
given item took place on two different occasions in time. To
qualify as a massed presentation, there must have been either a
single uninterrupted presentation of the item during learning or a
lag shorter than 1 s.) Our analysis of massed versus spaced
learning compared massed learning with the shortest spaced learn-
ing interval provided within a given study. Studies that failed to
include a massed presentation were excluded, leaving 271 com-
parisons of retention accuracy and 23 effect sizes. Only accuracy
differences are reported, because of insufficient effect size data.
Independent samples t tests were used for analyses, as a conser-
Figure 1. Graphical representation of two hypothetical studies and the
vative measure, as some studies were between subjects and others
difference and absolute lag graphs that would result when lag analysis of
were within subject.
these studies is performed.
Spaced presentations led to markedly better final-test perfor-
mance, compared with massed presentations. For retention inter-
To determine the relative benefits of specific ISIs, we were interested in
vals less than 1 min, spaced presentations improved final-test
the changes in average accuracy across different ISI bins, for a given performance by 9%, compared with massed presentations (see
retention interval bin. However, different studies contribute data to each Table 1). This finding appears to run counter to what has some-
ISI bin, even within a given retention interval bin. Thus, our comparisons times been referred to as the “Peterson paradox,” wherein there is
of interest, for both difference and absolute lag analyses, involved purportedly a massing benefit at short retention intervals. Perhaps
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 359

Table 1
Percentage Correct on the Final Recall Test for Massed and Spaced Conditions, Number of Performance Differences and Studies,
Total Number of Participants Summing Across All Study/Condition Combinations, and Statistical Analyses, for Spaced Versus Massed
Presentations

% Correct No. of
performance No. of No. of
Retention interval Massed Spaced SE differences studies participants Statistical analysis

1–59 s 41.2 50.1 1.7 105 96 5,086 t(208) ⫽ 3.7, p ⬍ .001

1 min–less than 10 min 33.8 44.8 1.5 124 117 6,762 t(246) ⫽ 5.0, p ⬍ .001
10 min–less than 1 day 40.6 47.9 8.1 11 10 870 t(20) ⫽ 0.6, p ⫽ .535
1 day 32.9 43.0 6.0 15 15 1,123 t(28) ⫽ 1.2, p ⫽ .249
2–7 days 31.1 45.4 7.3 9 9 435 t(16) ⫽ 1.4, p ⫽ .190
8–30 days 32.8 62.2 8.8 6 6 492 t(10) ⫽ 2.3, p ⬍ .05
31 days or more 17.0 39.0 1 1 43
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

All retention intervals 36.7 47.3 1.1 271 254 14,811 t(540) ⫽ 6.6, p ⬍ .001
This document is copyrighted by the American Psychological Association or one of its allied publishers.

this massing benefit occurs only with extremely short retention that prior meta-analyses’ use of ISI differences rather than absolute
intervals. For example, Peterson, Hillner, and Saltzman (1962) ISIs influenced their findings, as information is lost during differ-
found a massing benefit only when retention interval was 2 or 4 s ence computation. (Unfortunately, we do not have access to the
and not when retention interval was 8 or 16 s. Similarly, Peterson, actual data used in each review and thus cannot test these predic-
Saltzman, Hillner, and Land (1962) found a massing benefit at tions directly.)
retention intervals of 4 and 8 s, but Peterson, Wampler, Kirk- To examine how absolute ISI and ISI difference interacts with
patrick, and Saltzman (1963) failed to find a massing benefit at retention interval, we grouped the accuracy data into bins with
retention intervals of 8, 16, or 60 s. All these studies used very boundaries varying roughly by one log order of magnitude (limited
short ISIs, from 4 to 8 s. (The two tasks most predominantly used by the amount of data available). We would have preferred to use
by researchers—paired associate and list learning—were well rep- more precise log orders of magnitude to create our bins, but
resented across retention intervals.) Only 12 of 271 comparisons of combinations of ISI difference and retention interval are not
massed and spaced performance showed no effect or a negative evenly represented by the existing literature. Figure 2 plots each
effect from spacing, making the spacing effect quite robust. Most ISI difference and retention interval combination from every study
of these 12 comparisons used the same task type as studies that did included in our difference lag analyses. If this combination space
show a spacing benefit—paired associate learning. were evenly represented, Figure 2 would show a uniform “cloud”
We examined the interaction between magnitude of the spacing of data points. In addition to the irregular sampling of ISI differ-
effect and retention interval by calculating the difference in per- ence and retention interval combinations, large subsets of this
formance between massed and spaced presentations and collapsing combination space contain sparse amounts of data, or are missing
over each of seven retention interval ranges (see Table 1); there is data altogether. To best utilize the full range of data, we created
no hint that massed presentation was preferable to spaced, whether our own ISI and retention interval bins in a way that maximized
retention interval was very short (less than 1 min) or very long data usage while still attempting to capture log order of magnitude
(over 30 days). This suggests that there is always a large benefit changes.
when information is studied on two separate occasions instead of Accuracy difference and effect size lag analyses. The vast
only once. (Note that in every case examined here, the amounts of
majority of mean performance differences (80%) used a retention
study time for massed and spaced items were equivalent; thus, this
interval of less than 1 day, and only a few differences (4%) used
spacing benefit was not due to presentation time.)
a retention interval longer than 1 month (see Table 2). As men-
tioned earlier, Figure 2 shows this failure of the literature to fully
Lag Effects: Joint Effects of ISI and Retention Interval represent the space of ISI and retention interval combinations. This
Lag effects refer to changes in final-test memory performance as feature of the literature impacts our ability to analyze the qualita-
a function of change in ISI, when both ISIs and the differences tive findings from our difference lag analyses with inferential
between ISIs are greater than 0 s (in the current data set, at least statistics. (A recent case study critiquing meta-analysis technique
1 s). Prior reviews (Donovan & Radosevich, 1999; Janiszewski et suggests that statistical testing is not necessary to produce valid,
al., 2003) found different relationships between ISI and effect size; interpretable findings; Briggs, 2005).
Donovan and Radosevich (1999) reported nonmonotonic effects of For each study, we computed the accuracy difference that re-
ISI difference on effect size, whereas Janiszewski et al. (2003) sulted from each pairwise ISI difference, and we plotted the
found an increase in effect size as ISI difference increased. We average of these accuracy differences as a function of ISI differ-
have extended these previous reviews by including both ISI dif- ence and retention interval (see Figure 3). Only ISI difference by
ference and retention interval in our analysis. It is possible that retention interval bins that include three or more mean perfor-
Donovan and Radosevich and Janiszewski et al. found these dif- mance differences are shown. Several bins have fewer than three
ferent patterns because the optimal ISI difference changes as a mean accuracy differences, and accuracy difference values from
function of retention interval, and their reviews happened to in- bins with at least one data point are qualitatively consistent with
clude studies using different retention intervals. It is also possible the pattern of results shown in Figure 3. There is little, if any, ISI
 360 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 2. Scatter plot of interstudy interval (ISI) difference by retention interval, for all studies in the accuracy
difference lag analyses.

difference effect at retention intervals shorter than 1 day. In sharp pattern that optimal ISI difference increases as retention interval
contrast, for a 1-day retention interval, performance significantly increases is supported by quantitative analyses of the bin data (see
increased as ISI difference increased from 1–15 min to 1 day. Table 3). Furthermore, effect size data mirror these findings from
Qualitatively, one study suggested that performance should drop the accuracy data (see Figure 4).
when ISI difference increases beyond 1 day. The same pattern of Portions of our data are qualitatively similar to other meta-
results is seen with a 2- to 28-day retention interval: A 1-day ISI analysis findings. Like Donovan and Radosevich’s (1999) data,
difference produced a significant benefit over the 1- to 15-day ISI, our data show nonmonotonic effects of ISI difference. Like
and there was a marginally significant drop in performance as ISI Janiszewski et al.’s (2003) results, our data show generally im-
difference increased beyond 1 day. For retention intervals longer proved retention as ISI difference increases. Unfortunately, it is
than 1 month, we must rely on qualitative results, which suggest impossible to know whether we have confirmed these meta-
that the optimal ISI difference is longer than 1 day at retention analyses, because we do not know the retention interval values
intervals longer than 1 month. Overall, the results show a tendency used in each prior meta-analysis; however, our results provide a
for the greatest increases in final-test recall to be found at longer plausible mechanism by which these prior discrepant findings
ISI differences, the longer the retention interval. The qualitative might be reconciled.
For accuracy data, which are depicted in Figure 3, Table 4
shows the number of data points that use paired associate, list
Table 2
recall, or other types of tasks, and the overall number of data
Number of Performance Differences, Data Points, and Effect
Sizes, for Accuracy Difference, Absolute, and Effect Size Lag points, studies, and unique participants included in each bin. If the
Analyses, Respectively, by Retention Interval Range relative percentage of data points using each type of task changes
between bins, then changes in optimal ISI difference with change
No. of in retention interval could potentially be due to changes in the
Retention performance No. of No. of percentage of data points using each task type as opposed to
interval range differences data points effect sizes changes in retention interval. In the Appendix, Figures A1 and A2
2–59 s 174 301 14 (for paired associate and list recall tasks, respectively) illustrate
1 min–2 hr 259 452 53 that the joint effects of ISI difference and retention interval are due
1 day 27 52 16 to changes in retention interval and not to changes in task type.
2–28 days 56 108 31 Absolute ISI lag analyses. Although it is encouraging that
30 days or more 23 34 19
difference lag analyses show clear joint effects of ISI difference
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 361
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 3. For all studies in the accuracy difference lag analyses, accuracy difference between all adjacent pairs
of interstudy interval (ISI) values from each study, binned by difference in ISI and retention interval and
averaged across studies. When surrounded by ISI bins with lower accuracy values, the ISI bin showing the
highest accuracy value at each retention interval bin is indicated with an asterisk. Error bars represent one
standard error of the mean.

and retention interval, we are really interested in how absolute ISI difference ISI analyses but not in absolute lag analyses, and we
interacts with retention interval. On the basis of the absolute would expect an ISI change from 0 to 1 day to show a much larger
optimal ISI data, we can make concrete recommendations on how effect than an ISI change from 7 to 8 days.
large of a lag is optimal, given a particular retention interval. Mirroring accuracy difference data, most data points used a
Differences in performance between optimal and suboptimal ISI retention interval less than 1 day, and only a few data points
differences should be smaller and less meaningful as a measure of used a retention interval longer than 1 month (see Table 2). Just
ideal absolute ISI, compared with differences between optimal and as the literature failed to represent the full combination space of
suboptimal absolute ISIs. This is the case because ISI differences ISI differences and retention intervals for the difference lag
of 7– 8 days and ISI differences of 0 –1 day are combined in analyses, so too was the space of ISI and retention interval

Table 3
Shorter and Longer Interstudy Interval (ISI) Range, Retention Interval Range, Percentage
Correct at the Shorter and Longer ISI Range, and Statistical Analyses, for Accuracy Difference
Lag Analyses

% correct at ISI range

Shorter Longer ISI Retention
ISI range range interval range Shorter Longer SEM Statistical analysis

1–10 s 11–29 s 4–59 s 1.6 3.9 0.9 t(147) ⫽ 1.8, p ⫽ .077

11–29 s 1–15 min 4–59 s 3.9 ⫺0.9 1.2 t(75) ⫽ 1.4, p ⫽ .156
30–59 s 1 day 1 min–2 hr 3.4 1.0 2.5 t(61) ⫽ 0.9, p ⫽ .397
1–15 min 1 day 1 day 6.4 17.5 2.9 t(16) ⫽ 2.7, p ⬍ .05
1–15 min 1 day 2–28 days 1.5 10.3 2.5 t(26) ⫽ 2.4, p ⬍ .05
1 day 2–28 days 2–28 days 10.3 3.5 2.8 t(37) ⫽ 1.7, p ⫽ .091
1 day 2–28 days 30–2,900 days 6.5 9.0 2.7 t(15) ⫽ 0.7, p ⫽ .476
2–28 days 29–84 days 30–2,900 days 9.0 ⫺0.6 2.6 t(17) ⫽ 3.0, p ⬍ .01
 362 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 4. For all studies in the effect size lag analyses, effect sizes for all adjacent pairs of interstudy interval
(ISI) values from each study, binned by difference in ISI and retention interval and averaged across studies.
When surrounded by ISI bins with smaller effect size values, the ISI bin showing the largest effect size at each
retention interval bin is indicated with an asterisk. Error bars represent one standard error of the mean.

combinations inadequately sampled for the absolute lag analy- the optimal ISI was longer than 1 day at a supramonth retention
ses (see Figure 5). interval; Gordon (1925) showed that subday ISIs are optimal at
The plot of absolute ISI bin by retention interval bin is similar subday retention intervals and that supraday ISIs are optimal at
to the plot of ISI difference bin by retention interval bin (compare supraday retention intervals; Glenberg and Lehmann (1980)
Figures 6 and 3). Although there are small differences in the ISI showed results that mirror those of Gordon. These three studies are
bin showing optimal performance, in both cases, the trend is for consistent with a number of other studies (e.g., Balota, Duchek, &
the optimal ISI bin to increase as retention interval increases. Paullin, 1989; Glenberg, 1976; Peterson, Wampler, Kirkpatrick, &
Quantitative analyses are shown in Table 5, and the number of data Saltzman, 1963) that show within-study support for the hypothesis
points that used each task type is shown in Table 6. In the that optimal ISI increases as retention interval increases. Table 7
Appendix, data are separated by task type, either paired associate shows results for individual studies that examined ISIs and reten-
or list recall. As in the ISI difference lag analysis, only absolute ISI tion intervals of 1 day or more.
by retention interval bins that include three or more data points are Lag analysis summary. In summary, synthetic analyses sup-
shown. port the robustness and generality of ISI and retention interval joint
Within-study lag analyses. One problem with our absolute and effects that a few oft-cited individual experiments have sometimes
difference lag analyses is that different studies contribute differ- observed. Whereas earlier quantitative syntheses had sought to
entially to each bin. That is, each bin does not represent the same uncover effects of ISI difference or retention interval per se, the
combination of studies. For this reason, one must be wary that task present review suggests that the literature as a whole reflects
difficulty or other study-related factors played a role in differences nonmonotonic effect of absolute ISI upon memory performance at
between bins. A better comparison of lag effects would come from a given retention interval, as well as the positive relationship
within-study comparisons, across a wide range of ISIs and reten- between retention interval and the optimal absolute ISI value for
tion intervals, as this eliminates the problem with task difficulty. that retention interval.
To date, this massive study, which would need to include dozens
of ISI and retention interval combinations, has not been conducted.
Nonetheless, individual studies that represent a wide range of ISIs,
Experimental Design Issues
both sub- and supraday, at a single retention interval, are support- As noted in the introductory section, in examining commonly
ive of our findings: Cepeda et al. (2005) presented data in which used experimental designs, we found that a number of frequently
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 363

Table 4
Number of Performance Differences and Studies, Unique Participants, and Performance
Differences Using Paired Associate, List Recall, or Other Task Types, for Accuracy Difference
Lag Analyses, by Retention Interval Range and Interstudy Interval (ISI) Range

No. using

No. of Paired
Retention performance No. of No. of unique associate List recall Other
interval range ISI range differences studies participants tasks tasks tasks

4–59 s 1–10 s 79 28 1,539 35 41 3

4–59 s 11–29 s 70 39 2,083 20 48 2
4–59 s 30–59 s 18 12 694 6 12 0
4–59 s 1–15 min 7 4 327 5 2 0
1 min–2 hr 1–10 s 43 25 1,384 10 21 12
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

1 min–2 hr 11–29 s 91 50 2,736 27 59 5

This document is copyrighted by the American Psychological Association or one of its allied publishers.

1 min–2 hr 30–59 s 50 41 2,478 13 27 10

1 min–2 hr 1–15 min 52 40 3,295 18 13 21
1 min–2 hr 1 day 10 7 180 2 5 3
1 min–2 hr 2–28 days 13 9 618 3 0 10
1 day 30–59 s 9 8 469 4 5 0
1 day 1–15 min 14 9 667 6 6 2
1 day 1 day 4 4 86 0 3 1
2–28 days 30–59 s 3 3 174 1 0 2
2–28 days 1–15 min 14 6 613 6 6 2
2–28 days 1 day 14 11 902 5 4 5
2–28 days 2–28 days 25 18 4,118 7 6 12
30–2,900 days 1 day 4 3 106 4 0 0
30–2,900 days 2–28 days 13 3 294 12 0 1
30–2,900 days 29–84 days 6 3 160 6 0 0

cited studies contained serious design confounds or failed to im- study trials (with feedback) during the second session (Cepeda et
plement the claimed experimental manipulation. Given their ob- al., 2005). These experiments equated, across conditions, the de-
vious practical importance, we specifically examined studies that gree of initial learning (learning during the first session) and
used ISIs and retention intervals of 1 or more days (i.e., the studies avoided any confound between subsequent learning (learning dur-
in Table 7), to assess the quality of each study. ing the second session) and ISI. A number of studies had fixed
Studies contained several different confounds. One group of (Ausubel, 1966; Childers & Tomasello, 2002; Edwards, 1917;
studies provided learning to perfect performance and then relearn- Glenberg & Lehmann, 1980) restudy time, without feedback. Even
ing, with feedback, to the criteria of perfect performance (Bahrick, though the amount of relearning that took place during the second
1979; Bahrick et al., 1993; Bahrick & Phelps, 1987). These studies session was not assessed, relearning was not confounded in these
confounded number of relearning trials with ISI; that is, there was studies.
more relearning at longer ISIs. Some studies administered recog- To provide some indication of the importance of these method-
nition tests without feedback during learning sessions (in some ological issues, we examined the effect of ISI at similar retention
cases combined with recall tests; Burtt & Dobell, 1925; Spitzer, intervals, comparing the studies we judged to be confounded with
1939; Welborn, 1933). Because these studies did not provide those we judged to be nonconfounded. There are seven experi-
feedback, it is likely that no relearning occurred on the second and ments in five articles that used nonconfounded designs with ISIs
subsequent sessions for any item that elicited an error (see Pashler, and retention intervals of 1 day or more (Ausubel, 1966; Cepeda et
Cepeda, Wixted, & Rohrer, 2005). Some studies (Simon, 1979; al. 2005; Childers & Tomasello, 2002; Edwards, 1917; Glenberg &
E. C. Strong, 1973; E. K. Strong, 1916) provided unlimited restudy Lehmann, 1980). The Bahrick studies (Bahrick, 1979; Bahrick et
time that did not include testing with feedback. For these studies, al., 1993; Bahrick & Phelps, 1987), which confounded amount of
it is unclear how much information was acquired during relearning relearning and ISI, showed similar patterns to Cepeda et al. (2005),
sessions, because testing was not performed, and it is possible that Experiments 2a and 2b, which are unconfounded. The ideal ISI
the amount of relearning and ISI were confounded. Some studies indicated in all these studies is 1 month or more, at retention
were conducted outside a laboratory setting. For example, the intervals of 6 months or more. The Bahrick studies used far longer
studies by Simon (1979) and E. C. Strong (1973) relied on par- retention intervals than the Cepeda et al. study, making this com-
ticipants reading unsolicited direct mail advertising. Regular ad- parison less than perfect. Burtt and Dobell (1925) and Spitzer
herence to the paradigm was unlikely, as the authors of these (1939), who failed to provide relearning during relearning sessions
studies acknowledged. for items that elicited errors, found that an ISI of 7–10 days was
In contrast to these confounded studies, other studies appear free usually preferable to an ISI of 1–3 days, at retention intervals from
of major confounds. Several experiments provided either learning 10 –17 days. This contrasts with the unconfounded studies by
to perfect performance on the first session or a fixed number of Ausubel (1966); Cepeda et al., Experiment 1; and Glenberg and
first-session learning trials, followed by a small, fixed number of Lehmann (1980), who used similar retention intervals of 6 –10
 364 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 5. Scatter plot of interstudy interval by retention interval, for all studies in the absolute lag analyses.

days and who found that the ideal ISI was closer to 1–3 days than performance. Only a few studies have empirically examined this
7–10 days. Welborn (1933), who failed to provide relearning issue, however, resulting in 22 comparisons of retention accuracy
during relearning sessions for items that elicited errors, found and 8 effect size comparisons. Independent samples t tests were
effects similar to Cepeda et al.: In both studies, retention decreased used for analyses, as a conservative measure, as some studies were
as ISI increased beyond 1 day. However, Welborn used a retention between subjects (n ⫽ 7) and others were within subject (n ⫽ 11).
interval of 28 days, whereas Cepeda et al. used a retention interval Overall, expanding ISIs led to better performance than fixed
of 10 days. Two studies that used unlimited restudy time (Simon, intervals (see Table 8). Fifteen out of 18 studies used a paired
1979; E. C. Strong, 1973) are in line with similar unconfounded associate learning task, and we did not detect any systematic
studies (i.e., Ausubel, 1966; Cepeda et al., 2005, Experiment 1; differences related to type of task. Unfortunately, large standard
Glenberg & Lehmann, 1980), but one study that used unlimited errors, indicative of large between-study variability, make conclu-
restudy time (E. K. Strong, 1916) is not. Even with some incon- sions drawn from expanding versus fixed interval data necessarily
sistencies between confounded and unconfounded experimental tentative. Large between-study differences can be seen more dra-
designs, we believe that our analyses of ISI and retention interval matically by examining the empirical data from three different
joint effects are not undermined by experimental design problems researchers, shown in Table 9. All three researchers used ISIs and
plaguing some of the experiments included in our analyses. In- retention intervals of at least 1 day. One researcher (Tsai, 1927)
deed, regardless of whether the confounded studies are excluded, found better performance with expanding study intervals, one
the same basic conclusion would be drawn: Optimal ISI increases
(Cull, 2000) found better performance with fixed study intervals,
as retention interval increases.
and one (Clark, 1928) found no difference between fixed and
expanding intervals. In all three sets of studies, the average
Expanding Versus Fixed ISIs between-presentation ISI was the same for expanding and fixed
It often has been suggested that when items are to be relearned ISIs, and retention intervals overlapped across studies; use of
on two or more occasions, memory can be maximized by relearn- different ISIs and retention intervals does not explain differences
ing information at increasingly spaced (expanding) ISIs, as op- between each set of studies. Any number of differences may
posed to relearning at a fixed ISI (Bahrick & Phelps, 1987; explain these conflicting findings. One variable that might explain
Hollingworth, 1913; Kitson, 1921; Landauer & Bjork, 1978; between-study differences is the presence of feedback. Expanding
Modigliani, 1967; Pyle, 1913). One intuitive version of this for- intervals might benefit performance when feedback is withheld,
mulation says memory is best promoted when a learner undergoes because expanding intervals minimize the chance of forgetting an
tests that are as difficult as possible, while maintaining errorless item. (In the absence of feedback, forgetting an item usually causes
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 365
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 6. For all studies in the absolute lag analyses, accuracy, binned by interstudy interval (ISI) and retention
interval and averaged across studies. When surrounded by ISI bins with lower accuracy values, the ISI bin
showing the highest accuracy value at each retention interval bin is indicated with an asterisk. Error bars
represent one standard error of the mean.

the item to be unrecoverable; see Pashler et al., 2005) This feed- study opportunities for a given item, when study opportunities for
back hypothesis is supported by a single study (Cull, Shaughnessy, both the shorter and longer lag conditions are separated by 1 s or
& Zechmeister, 1996). Unfortunately, the feedback hypothesis more). In the present review, this spacing versus lag distinction
cannot be tested adequately with current data, because all three of proved helpful in quantifying the relationship between level of
the studies using ISIs and retention intervals longer than 1 day retention, ISI, and retention interval. When participants learned
either provided testing with feedback (Cull, 2000) or provided a individual items at two different points in time (spaced; lag of 1 s
fixed amount of item restudy time (Clark, 1928; Cull, 2000; Tsai, or more), equating total study time for each item, they recalled a
1927), which was functionally equivalent to providing feedback greater percentage of items than when the same study time was
(because the entire to-be-learned item was present). With the nearly uninterrupted (massed; lag of less than 1 s). This improve-
exception of Cull et al. (1996) and Landauer and Bjork (1978), ment occurred regardless of whether the retention interval was less
expanding interval studies that used retention intervals of less than than 1 min or more than 1 month. In short, for the spacing effect
1 day (Cull, 1995; Foos & Smith, 1974; Hser & Wickens, 1989; proper, we failed to find any evidence that the effect is modulated
Siegel & Misselt, 1984) all provided either a fixed amount of by retention interval. At first blush, this conclusion might seem to
restudy time for each entire item or testing with feedback. We are suggest that students are wrong to believe that cramming imme-
left with inadequate evidence to support or refute the feedback
diately before an exam is an effective strategy to enhance perfor-
hypothesis.
mance on the exam. However, a few hours of cramming would
typically involve repeated noncontiguous study of individual bits
General Discussion of information, rather than literal massing as examined in the
Although the distributed practice effect has spawned a large studies noted. Furthermore, most advocates of cramming probably
literature, prior meta-analyses (Donovan & Radosevich, 1999; have in mind the comparison between studying immediately prior
Janiszewski et al., 2003; T. D. Lee & Genovese, 1988) failed to to the exam and studying days or weeks prior to the exam.
distinguish spacing effects (a single presentation, or a lag less than A different pattern of results was observed for increases in ISI
1 s, vs. multiple presentations, or a lag of 1 s or more, of a given beyond the massed condition (i.e., from a nonzero value to an even
item; equal total study time for that item, whether in the spaced or larger nonzero value). When ISI was increased, participants re-
massed condition) from lag effects (less vs. more time between tained more information. However, for long ISIs, in proportion to
 366 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

Table 5
Shorter and Longer Interstudy Interval (ISI) Range, Retention Interval Range, Percentage
Correct at the Shorter and Longer ISI Range, and Satistical Analyses, for Absolute Lag Analyses

% correct at ISI range

Shorter ISI Longer ISI Retention
range range interval range Shorter Longer SEM Statistical analysis

1–10 s 30–59 s 2–59 s 49.4 54.1 2.5 t(162) ⫽ 1.4, p ⫽ .167

30–59 s 1 min–3 hr 2–59 s 54.1 48.8 2.7 t(90) ⫽ 1.3, p ⫽ .198
1–10 s 1 min–3 hr 1 min–2 hr 42.3 54.0 1.7 t(248) ⫽ 4.8, p ⬍ .001
1 min–3 hr 2–28 days 1 min–2 hr 54.0 35.7 4.9 t(161) ⫽ 3.4, p ⬍ .005
30–59 s 1 day 1 day 36.0 62.5 7.8 t(16) ⫽ 2.2, p ⬍ .05
11–29 s 1 day 2–28 days 26.4 52.8 7.6 t(21) ⫽ 2.5, p ⬍ .05
1 day 2–28 days 2–28 days 52.8 45.5 4.2 t(58) ⫽ 1.1, p ⫽ .270
1 min–3 hr 29–168 days 30–2,900 days 27.0 50.3 11.5 t(12) ⫽ 1.4, p ⫽ .180
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

retention interval, further increases in ISI reduced accuracy. Thus, Some researchers have suggested, with little apparent empirical
for a given retention interval, there was a nonzero value of ISI that backing, that expanding ISIs improve long-term learning (Holling-
optimized accuracy. (This is known as a nonmonotonic lag effect.) worth, 1913; Kitson, 1921; Landauer & Bjork, 1978; Pyle, 1913);
Moreover, the optimal ISI increased as retention interval in- in contrast, some empirical studies (Cull, 1995, 2000; Foos &
creased. For instance, at retention intervals of less than 1 min, ISIs Smith, 1974) have found that expanding intervals are less effective
of less than 1 min maximized retention; at retention intervals of 6 than fixed spacing intervals. Our review of the evidence suggests
months or more, ISIs of at least 1 month maximized retention. that, in general, expanding intervals either benefit learning or
These results clearly show that a single ISI does not produce produce effects similar to studying with fixed spacing. The liter-
optimal retention across a wide range of retention intervals. The ature offers examples of impaired performance with expanding
nonmonotonic effect of ISI upon retention and the dependency of intervals (Cull, 2000; Foos & Smith, 1974) and examples of
optimal ISI upon retention interval both appear to characterize the expanding interval benefits (Cull et al., 1996; Hser & Wickens,
literature as a whole, as well as a few well-known specific studies 1989; Landauer & Bjork, 1978; Tsai, 1927). We found no obvious
(e.g., Glenberg & Lehmann, 1980). systematic differences between studies that do and do not show

Table 6
Number of Data Points and Studies, Unique Participants, and Data Points Using Paired
Associate, List Recall, or Other Task Types, for Absolute Lag Analyses, by Retention Interval
Range and Interstudy Interval (ISI) Range

No. of data points using

Retention No. of No. of No. of unique Paired List recall Other

interval range ISI range data points studies participants associate tasks tasks tasks

2–59 s 1–10 s 113 62 3,248 41 66 6

2–59 s 11–29 s 96 57 2,694 29 59 8
2–59 s 30–59 s 51 35 1,707 21 24 6
2–59 s 1 min–3 hr 41 20 1,152 14 27 0
1 min–2 hr 1–10 s 101 66 3,711 24 59 18
1 min–2 hr 11–29 s 84 76 4,773 25 53 6
1 min–2 hr 30–59 s 93 80 4,785 34 40 19
1 min–2 hr 1 min–3 hr 149 83 4,867 45 64 40
1 min–2 hr 1 day 11 8 222 2 5 4
1 min–2 hr 2–28 days 14 9 390 3 0 11
1 day 1–10 s 4 4 60 1 3 0
1 day 30–59 s 12 11 552 3 6 3
1 day 1 min–3 hr 30 19 1,100 12 16 2
1 day 1 day 6 6 83 0 4 2
2–28 days 11–29 s 5 5 190 4 0 1
2–28 days 30–59 s 8 5 267 0 4 4
2–28 days 1 min–3 hr 35 20 1,215 12 15 8
2–28 days 1 day 18 15 892 4 5 9
2–28 days 2–28 days 42 24 3,344 14 11 17
30–2,900 days 1 min–3 hr 4 3 53 4 0 0
30–2,900 days 1 day 5 3 54 4 0 1
30–2,900 days 2–28 days 15 4 175 13 0 2
30–2,900 days 29–168 days 10 3 84 10 0 0
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 367

Table 7
Final Test Performance for Long Interstudy Interval (ISI), Long Retention Interval Studies

Retention Final test performance

Study ISI (days) interval (days) (% correct)

Ausubel (1966) 1 6 43
7 6 40
Bahrick (1979), Experiment 2 1 30 86
30 30 95
Bahrick, Bahrick, Bahrick, & Bahrick (1993) 14 360 62
28 360 67
56 360 76
14 720 55
28 720 61
56 720 67
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

14 1,080 45
This document is copyrighted by the American Psychological Association or one of its allied publishers.

28 1,080 62
56 1,080 66
14 1,800 36
28 1,800 46
56 1,800 60
Bahrick & Phelps (1987) 1 2,900 8
30 2,900 15
Burtt & Dobell (1925), Experiment 2 3 10 22
10 10 48
3 16 16
10 16 15
Burtt & Dobell (1925), Experiment 3 3 10 30
10 10 55
3 17 21
10 17 25
Cepeda et al. (2005), Experiment 1 1 10 74
2 10 69
4 10 68
7 10 69
14 10 65
Cepeda et al. (2005), Experiment 2a 1 168 33
7 168 47
28 168 56
84 168 43
168 168 45
Cepeda et al. (2005), Experiment 2b 1 168 9
7 168 14
28 168 26
84 168 19
168 168 17
Childers & Tomasello (2002), Experiment 1 1 1 58
3 1 58
1 7 53
3 7 53
Edwards (1917) 1 3 38
2 3 19
1 4 37
5 4 32
Glenberg & Lehmann (1980), Experiment 2 1 7 32
7 7 25
Simon (1979) 7 7 62
28 7 43
7 35 30
28 35 31
Spitzer (1939) 1 14 36
7 14 39
14 14 39
E. C. Strong (1973) 7 7 16
14 7 11
28 7 11
E. K. Strong (1916) 1 28 13
7 28 16
Welborn (1933) 1 28 72
3 28 63
 368 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

Table 8
Percentage Correct on the Final Recall Test for Expanding and Fixed Conditions, Number of Performance Differences and Studies,
Total Number of Participants Summing Across All Study and Condition Combinations, and Statistical Analyses, for Expanding Versus
Fixed Study Intervals

% Correct

Expanding Fixed No. of performance No. of No. of

Retention interval conditions conditions SEM differences studies participants Statistical analysis

1–59 s 91.0 91.0 1 1 24

1 min–less than 10 min 49.8 48.9 5.6 10 8 580 t(18) ⫽ 0.1, p ⫽ .91
10 min–less than 1 day 77.8 70.0 11.5 4 3 614 t(6) ⫽ 0.5, p ⫽ .65
1 day
2–7 days 66.3 59.5 10.9 4 3 185 t(6) ⫽ 0.4, p ⫽ .68
8–30 days 66.3 64.0 11.2 3 3 115 t(4) ⫽ 0.1, p ⫽ .89
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

31 days or more
This document is copyrighted by the American Psychological Association or one of its allied publishers.

All retention intervals 62.0 58.6 4.6 22 18 1518 t(42) ⫽ 0.5, p ⫽ .61

expanding interval benefits, although one difference that might (Massed items have associations only to the two immediately
account for interstudy variability is the presence or absence of adjacent items, whereas spaced items have associations to at least
feedback. Given the practical import of multisession study (almost three and usually four adjacent items. Spaced items have more
all learning takes place on more than two occasions), this topic associations because each spaced item is sandwiched between two
clearly deserves further research. items in the first session and sandwiched between two different
items in the second session.) Because these and other theories are
Implications for Theories of Distributed Practice able to make differential predictions for spaced versus massed
presentations, as well as for changes in lag, our theoretical discus-
Many theories purport to account for distributed practice effects, sion applies to both spacing and lag effects. In other words, our
and little consensus has been achieved about the validity of these theoretical discussion applies to distributed practice effects, where
accounts. Although a thorough theoretical analysis of the distrib- distributed practice includes both spacing and lag effects.
uted practice task is well beyond the scope of the present, rela- The first class of theoretical accounts that we discuss is deficient
tively focused, review (for reviews of distributed practice, see processing theory. Deficient processing theory is based on mech-
Glenberg, 1979; Hintzman, 1974), it is of interest to examine how anisms that alter the amount of focus received by particular items.
some of the principle conclusions reached in the present review An example of deficient processing theory is the inattention theory
might affect the credibility of some frequently discussed theories. (Hintzman, 1974). Inattention theory suggests that when the ISI is
We focus on four theories in detail, without in any way implying short, processing of the second presentation is reduced in quality
that other theories lack merit. and/or quantity: The learner pays less attention to something that
To date, theorists have failed to distinguish between spacing and is, by virtue of the short ISI, relatively more familiar. Deficient
lag effects. This makes it difficult to know how broadly theorists processing theory has struck many writers as offering an intu-
intended their theories to be applied. Theories often predict that itively reasonable account of why massed presentations would
spaced and massed items will be processed differently—for exam- produce inferior memory. The fact that massed presentations are
ple, the inattention theory predicts that spaced items will receive normally inferior even when retention interval is very short, as
greater attentional focus; the encoding variability theory predicts noted above, certainly seems consistent with this account. This
that spaced items will contain more interitem associations. account also enjoys support from a study that suggests it is the
trace of the second presentation, rather than the first, that is
reduced when ISI is shorter than optimal (Hintzman, Block, &
Table 9
Summers, 1973).
Percentage Correct on Final Test, for Fixed and Expanding
Can deficient processing theory handle one of our meta-
Study Intervals, for Studies with a Retention Interval of at Least
analysis’s primary findings, the joint effects of ISI and retention
1 Day
interval? Suppose Study 1 yields a single memory trace, which is
Retention Fixed study Expanding study then further strengthened as a consequence of Study 2, and further
ISI interval intervals intervals suppose this trace is characterized by two parameters: the strength
Study (days) (days) (% correct) (% correct) of the trace and its rate of decay. These two parameters are found
in a number of functions used to describe forgetting, including the
Clark (1928) 2 21 63 63
Cull (2000), Exp. 3 2 3 98 84 commonly preferred power law function described by Wixted and
Cull (2000), Exp. 4 2 8 89 82 Ebbesen (1997). If Study 2 strengthens the trace without affecting
Tsai (1927), Exp. 2 2 3 48 61 its decay parameter, then even if the degree of strengthening is
Tsai (1927), Exp. 2 2 7 36 46 assumed to vary in some arbitrary fashion with ISI, there will have
Tsai (1927), Exp. 3 2 3 56 74
Tsai (1927), Exp. 3 2 17 40 54
to be a single value of ISI that yields the strongest trace. This ISI
would produce optimal later recall, regardless of how long the final
Note. ISI ⫽ interstudy interval; Exp. ⫽ Experiment. test is delayed. Thus, this version of the deficient processing theory
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 369

is inconsistent with the effect of retention interval on optimal ISI, Recent simulations (see Cepeda et al., 2005) demonstrate that a
as seen in the present integrative review. simple contextual drift mechanism—in conjunction with certain
One could, of course, hypothesize that it is not just strength, but reasonable assumptions about the function relating similarity to
also decay rate, that is modified by Study 2 (making the account retrieval probability— can readily produce distributed practice ef-
closer to suggestions by Pavlik & Anderson, 2003; Reed, 1977; fects. Briefly, we created a simple model of encoding variability,
and Wickelgren, 1972, discussed below), but this assumption is at based solely on contextual drift over time. Both context and time
odds with classic findings in the forgetting literature. That is, vary on a single dimension. Over time, location in one-
variations in the degree of attention paid to a study item appear to dimensional contextual space changes and this change is either
affect either the quantity or the quality of processing, but not both. toward or away from the context at time x. Encouragingly, our
Direct manipulations of the quantity of processing are known to simulations reveal that this simple version of encoding variability
have a large effect on the degree of learning (a proxy for strength) theory predicts both nonmonotonic effects of ISI and that the
while having little or no effect on the rate of forgetting (Anderson, optimal ISI increases in a predictable fashion as retention interval
2000; Underwood & Keppel, 1963; Wixted, 2004). Similarly, increases (with the optimal ratio of ISI to retention interval de-
manipulating the quality of processing at encoding by manipulat- creasing as retention interval itself grows).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

ing depth of processing has a large effect on the degree of learning Encoding variability theory appears to encounter substantial
but a negligible effect on the rate of forgetting (McBride & problems when accounting for certain other findings (e.g.,
Dosher, 1997). ISI, in contrast, has a large effect on the rate of Bellezza, Winkler, & Andrasik, 1975; Dempster, 1987b). One
forgetting. Specifically, as ISI increases, the rate of decay de- potential problem for encoding variability theory comes from Ross
creases, which is to say that longer ISIs produce more gradual and Landauer (1978), who showed that greater spacing between
forgetting curves. Nevertheless, it is conceivable that variations in two instances of two different words presented at various list
attention affect the quality of processing in some other, as yet positions did not enhance the probability that the subject would
unspecified, way. If so, then the deficient processing theory may later recollect either the first- or the second-presented item. In most
yet be able to accommodate our findings. In light of the available versions of the encoding variability theory, one would expect such
evidence, however, the effect of ISI on the rate of forgetting seems an enhancement for precisely the redundancy-related reasons
not to be an indirect result of the effect of that manipulation on noted above (see Raaijmakers, 2003, for a model of encoding
attention. variability that, according to its author, can be reconciled with
Things become more complicated if one assumes that Study 1 Ross and Landauer’s results). A second potential problem with
and Study 2 produce two independent traces. One could, for encoding variability theory is when participants are deliberately
example, suppose that the stronger is the trace resulting from Study induced to encode items in a more variable fashion, this often fails
1 (call this Trace 1) at the time of Study 2, and the weaker is the to produce a later recall benefit or fails to modulate the distributed
trace formed from Study 2 (Trace 2). Once again, however, if it is practice effect (Dempster, 1987a; Hintzman & Stern, 1977; Maki
assumed that Trace 1 strength affects the strength but not the decay & Hasher, 1975; Maskarinec & Thompson, 1976; McDaniel &
rate of Trace 2, this independent-trace account also fails to explain Pressley, 1984; Postman & Knecht, 1983).
the dependence of optimal ISI upon retention interval. A third explanation for the distributed practice effect is termed
In summary, deficient processing theory appears to be threat- consolidation theory (Wickelgren, 1972). Upon the second presen-
ened by complex joint effects of ISI and retention interval that tation of a repeated item, consolidation theory proposes that a new
were revealed in the literature, as documented in the present (second) trace is formed that inherits the state of consolidation of
review. Although it would obviously be premature to say that all the first occurrence of that item. If the ISI is 1 week, more
versions of the deficient processing account are falsified, the consolidation into long-term memory will have occurred than if
challenges appear substantial. (The deficient processing account the ISI is 1 day, and the second trace will inherit this higher state
confronts a separate difficulty in the finding that providing rewards of consolidation. If the delay is too long, say 1 year, there will be
for remembering does not reduce distributed practice effects; no initial memory trace whose consolidation state can be inherited,
Hintzman, Summers, Eki, & Moore, 1975.) and thus retention of that item will be lowered. This theory, as well
A second widely discussed class of models is usually termed as related accounts proposed by Pavlik and Anderson (2003) and
encoding variability theory (Glenberg, 1979; Melton, 1970). In the Reed (1977), quite directly predicts that, for a given retention
simplest versions of this account, traces stored when an item is interval, ISI varies nonmonotonically; it may or may not also
studied represent the context in which the item is stored, as well as predict that optimal ISI increases monotonically with retention
the item itself. Over time, the prevailing context is assumed to interval.
undergo random drift. As a result, the average distance between One experimental result that appears to undercut consolidation
any prior context and the current context will increase with the theory is the finding of Hintzman et al. (1973), which suggests that
passing of time. The account assumes that the shorter the distance learning produced by Study 2, rather than learning produced by
between the context existing at retrieval and the context that Study 1, is decreased when the Study 2 presentation follows
existed at study, the greater the likelihood of retrieval success. closely after the Study 1 presentation (see Murray, 1983, for
Thus, as the ISI between Study 1 and Study 2 increases, the arguments that this finding may not be definitive). If Study 1
probability of later recall might grow, simply because it becomes processing were interrupted, as purported in consolidation theory,
more likely that the retrieval context will be similar to at least one then Study 1 and not Study 2 learning should be decreased.
of the study contexts. This can predict that the probability of later Study-phase retrieval theory (Braun & Rubin, 1998; Murray,
recall will grow as ISI increases, because it becomes more likely 1983; Thios & D’Agostino, 1976) provides a fourth explanation of
that the retrieval context will be similar to at least one of the study the distributed practice effect. In this theory, the second (restudy)
contexts. presentation serves as a cue to recall the memory trace of the first
 370 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

presentation. This is similar to consolidation theory, but unlike in increases as the duration over which information needs to be
consolidation theory, consolidation of the first-presentation mem- retained increases. For most practical purposes, this retention
ory trace is not interrupted. Study-phase retrieval is supported by interval will be months or years, so the optimal ISI will likely be
empirical evidence: A lag effect is found when retrieval of the first well in excess of 1 day. Obviously, there is a need for much more
presentation is required (Thios & D’Agostino, 1976); in contrast, detailed study on this point, despite the time-consuming nature of
no lag effect is found when retrieval is not required. Notably, such studies. One question of particular practical interest is
interrupting or otherwise diminishing study-phase retrieval can whether ISIs that are longer than the optimal ISI produce large
eliminate the distributed practice effect (Thios & D’Agostino, decrements in retention or only minor ones. If they produce only
1976). The mechanism(s) by which retrieval of the first- minor decrements in retention, then a simple principle “seek to
presentation trace helps later retrieval has been left open to inter- maximize lag wherever possible” may be workable. On the other
pretation: Sources of benefit may include increased contextual hand, if these decrements are substantial, then a serious consider-
associations or strengthened first-presentation traces. As in con- ation of the expected duration over which memory access will be
solidation theory, if the first-presentation memory trace cannot be needed may often be needed if one is to maximize the efficiency
retrieved, then later retrieval will be less likely; thus, study-phase of learning.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

retrieval theory predicts nonmonotonic lag effects. It is unclear

whether study-phase retrieval theory predicts that optimal ISI Analysis Limitations
increases monotonically with retention interval.
In summary, the findings gleaned in the present quantitative The present analysis is subject to many of the same limitations
synthesis appear to have a significant bearing on the four potential present in all meta-analyses (for discussion, see Hedges & Olkin,
theories of the distributed practice effect discussed here. At least 1985; Hunter & Schmidt, 1990). For example, there is no way to
on the basis of our preliminary analysis, study-phase retrieval, accurately calculate the number of studies with null findings (i.e.,
consolidation, and encoding variability theories survive as candi- a lack of distributed practice effect), because many studies never
date distributed practice theories, whereas deficient processing reach publication. This “file drawer problem” (Rosenthal, 1979)
theory does not readily survive. Notably, only encoding variability reflects the reluctance of journals to publish null findings. Hunter
theory has been shown, through mathematical modeling, to pro- and Schmidt (1990) point out that the file drawer problem tends to
duce increases in optimal ISI as retention interval increases. It be a nonissue when large effect sizes are identified, as in the
remains unclear whether consolidation and/or study-phase re- present analysis, because of the enormous ratio of unpublished to
trieval theory can produce this effect and whether these results can published data that would be needed to invalidate a large effect
be reconciled with the empirical challenges that have been arrayed size.
against them, as noted above. Further analytic work is needed to
explore in more detail the relationship between potential theories Limitations of Currently Available Data
of distributed practice and the finding that optimal ISI increases as
retention interval increases. As noted above, new studies are sorely needed to clarify the
effects of interstudy and retention intervals that are educationally
Educational Implications of Findings relevant, that is, on the order of weeks, months, or years. It is clear
from existing studies that the distribution of a given amount of
A primary goal of almost all education is to teach material so study time over multiday periods produces better long-term reten-
that it will be remembered for an extended period of time, on the tion than study over a few-minute period, but it is unclear how
order of at least months and, more often, years. The data described quickly retention drops off when intervals exceed the optimal ISI.
here reaffirm the view (expressed most forcefully by Bahrick, If the field of learning and memory is to inform educational
2005, and Dempster, 1988) that separating learning episodes by a practice, what is evidently needed is much less emphasis on
period of at least 1 day, rather than concentrating all learning into convenient single-session studies and much more research with
one session, is extremely useful for maximizing long-term reten- meaningful retention intervals (see Bahrick, 2005, for similar
tion. Every study examined here with a retention interval longer comments).
than 1 month (Bahrick, 1979; Bahrick, et al., 1993; Bahrick & The effects of nonconstant (i.e., expanding or contracting) learn-
Phelps, 1987; Cepeda et al., 2005) demonstrated a benefit from ing schedules on retention are still poorly understood. Expanding
distribution of learning across weeks or months, as opposed to study intervals rarely seem to produce much harm for recall after
learning across a 1-day interval; learning within a single day long delays, but there is insufficient data to say whether they help.
impaired learning, compared with a 1-day interval between study This has not stopped some software developers from assuming that
episodes; learning at one single point in time impaired learning, expanding study intervals work better than fixed intervals. For
compared with a several-minute interval between study episodes. example, Woźniak and Gorzelańczyk (1994; see also SuperMemo
The average observed benefit from distributed practice (over World, n.d.) offered a “universal formula” designed to space
massed practice) in these studies was 15%, and it appeared to hold repetitions at an interval that will produce 95% retention, based on
for children (Bloom & Shuell, 1981; Childers & Tomasello, 2002; Bahrick and Phelps’s (1987) proposal that the ideal spacing inter-
Edwards, 1917; Fishman, Keller, & Atkinson, 1968; Harzem, Lee, val is the longest ISI before items are forgotten.
& Miles, 1976) as well as adults. After more than a century of We sometimes found it necessary to focus on change in accu-
research on spacing, much of it motivated by the obvious practical racy as a measure, instead of the more traditional effect size
implications of the phenomenon, it is unfortunate that we cannot measure, because the variance data necessary to compute effect
say with certainty how long the ISI should be to optimize long- size were lacking in most published results in this area. It was very
term retention. The present results suggest that the optimal ISI encouraging to observe that results differed little depending upon
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 371

whether accuracy difference or effect size was examined. Future memory durability over the range of time to which educators
research in the area of distributed practice should report the sample typically aspire. We have little doubt that relatively expensive and
size, means, and standard deviations for each ISI data point, even time-consuming studies involving substantial retention intervals
in cases of no significant difference, so that effect size can be will need to be carried out if practical benefits are to be wrung
calculated in future meta-analyses (American Psychological Asso- from distributed practice research; it is hoped that the present
ciation, 2001). As well, it would be useful if researchers reported review will help researchers to pinpoint where that effort might be
pairwise correlations between ISIs, so that dependence between the most useful and illuminating.
responses can be corrected, whenever the design is within subjects.
Almost all distributed practice data in our analysis (85%) are
based on performance of young adults (see Table 10). Although
References
most studies using children show a distributed practice effect, there References marked with an asterisk indicate studies included in the
simply is insufficient data to make strong claims about the simi- meta-analysis.
larity between children’s and adults’ responses to distributed prac-
tice, when retention interval is 1 day or longer. Until empirical data *Actkinson, T. R. (1977). The effects of structural, presentation, and
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

response variables on storage and recall of simple prose materials

This document is copyrighted by the American Psychological Association or one of its allied publishers.

examining the distributed practice effect in children are collected,

(Doctoral dissertation, Texas Christian University, 1976). Dissertation
using retention intervals of months or years and ISIs of days or Abstracts International, 37, 5388.
months (no usable data meeting these criteria currently exist, to our Adams, J. A. (1987). Historical review and appraisal of research on the
knowledge), we cannot say for certain that children’s long-term learning, retention, and transfer of human motor skills. Psychological
memory will benefit from distributed practice. Bulletin, 101, 41–74.
American Psychological Association. (2001). Publication manual of the
Summary American Psychological Association (5th ed.). Washington, DC: Author.
Anderson, J. R. (2000). Learning and memory: An integrated approach
More than 100 years of distributed practice research have dem- (2nd ed.). New York: Wiley.
onstrated that learning is powerfully affected by the temporal *Austin, S. D. M. (1921). A study of logical memory. American Journal of
distribution of study time. More specifically, spaced (vs. massed) Psychology, 32, 370 – 403.
learning of items consistently shows benefits, regardless of reten- *Ausubel, D. P. (1966). Early versus delayed review in meaningful learn-
tion interval, and learning benefits increase with increased time ing. Psychology in the Schools, 3, 195–198.
*Azerrad, J., & Kennedy, W. A. (1969). Short-term recall of CVC tri-
lags between learning presentations. On the other hand, it seems
grams. Psychological Reports, 24, 698.
clear that once the interval between learning sessions reaches some *Bahrick, H. P. (1979). Maintenance of knowledge: Questions about
relatively long amount of time, further increases either have no memory we forgot to ask. Journal of Experimental Psychology: Gen-
effect upon or decrease memory as measured in a later test. The eral, 108, 296 –308.
magnitude of the observed distributed practice benefit depends on Bahrick, H. P. (2005). The long-term neglect of long-term memory: Rea-
the joint effects of ISI and retention interval; retention interval sons and remedies. In A. F. Healy (Ed.), Experimental cognitive psy-
influences the peak of this function. Distributing learning across chology and its applications: Decade of behavior (pp. 89 –100). Wash-
different days (instead of grouping learning episodes within a ington, DC: American Psychological Association.
single day) greatly improves the amount of material retained for *Bahrick, H. P., Bahrick, L. E., Bahrick, A. S., & Bahrick, P. E. (1993).
sizable periods of time; the literature clearly suggests that distrib- Maintenance of foreign language vocabulary and the spacing effect.
Psychological Science, 4, 316 –321.
uting practice in this way is likely to markedly improve students’
*Bahrick, H. P., & Phelps, E. (1987). Retention of Spanish vocabulary over
retention of course material. Results also show that despite the 8 years. Journal of Experimental Psychology: Learning, Memory, and
sheer volume of the distributed practice literature, some of the Cognition, 13, 344 –349.
most practically important questions remain open, including mag- *Balota, D. A., Duchek, J. M., & Paullin, R. (1989). Age-related differ-
nitude of the drop-off produced by use of a supraoptimal ISI, the ences in the impact of spacing, lag, and retention interval. Psychology
relative merits of expanding (as compared with uniformly spacing) and Aging, 4, 3–9.
learning sessions, and the range of ISI values needed to promote *Batchelder, W. H., & Riefer, D. M. (1980). Separation of storage and
retrieval factors in free recall of clusterable pairs. Psychological Review,
87, 375–397.
Table 10 Becker, B. J. (1988). Synthesizing standardized mean-change measures. Brit-
Number of Performance Differences and Effect Sizes, by Age ish Journal of Mathematical and Statistical Psychology, 41, 257–278.
Group *Bellezza, F. S., Winkler, H. B., & Andrasik, F., Jr. (1975). Encoding
processes and the spacing effect. Memory & Cognition, 3, 451– 457.
No. of performance No. of *Bellezza, F. S., & Young, D. R. (1989). Chunking of repeated events in
Age group differences effect sizes memory. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 15, 990 –997.
Preschool 29 13 *Bird, C. P., Nicholson, A. J., & Ringer, S. (1978). Resistance of the
Elementary school 26 3 spacing effect to variations in encoding. American Journal of Psychol-
Junior high school 29 3 ogy, 91, 713–721.
High school 14 6 *Bjork, R. A., & Allen, T. W. (1970). The spacing effect: Consolidation or
Young adult (18–35) 714 97
differential encoding? Journal of Verbal Learning and Verbal Behavior,
Middle-aged adult (36–60) 2 0
Older adult (61⫹) 11 2 9, 567–572.
Mixed adult (18⫹) 14 9 *Bloom, K. C., & Shuell, T. J. (1981). Effects of massed and distributed
practice on the learning and retention of second-language vocabulary.
Note. Ages are in years. Journal of Educational Research, 74, 245–248.
 372 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

*Borkowski, J. G. (1967). Distributed practice in short-term memory. *D’Agostino, P. R., & DeRemer, P. (1973). Repetition effects as a function
Journal of Verbal Learning and Verbal Behavior, 6, 66 –72. of rehearsal and encoding variability. Journal of Verbal Learning and
*Braun, K. A. (1995). A retrieval model of the spacing effect (Doctoral Verbal Behavior, 12, 108 –113.
dissertation, Duke University, 1994). Dissertation Abstracts Interna- D’Amico, E. J., Neilands, T. B., & Zambarano, R. (2001). Power analysis
tional, 55, 3615. for multivariate and repeated measures designs: A flexible approach
*Braun, K., & Rubin, D. C. (1998). The spacing effect depends on an using the SPSS MANOVA procedure. Behavior Research Methods,
encoding deficit, retrieval, and time in working memory: Evidence from Instruments, and Computers, 33, 479 – 484.
once-presented words. Memory, 6, 37– 65. *Dellarosa, D., & Bourne, L. E., Jr. (1985). Surface form and the spacing
*Bray, J. F., Robbins, D., & Witcher, W. B., Jr. (1976). Encoding vari- effect. Memory & Cognition, 13, 529 –537.
ability theory and the spacing effect in associate learning. Memory & *Dempster, F. N. (1987a). Effects of variable encoding and spaced pre-
Cognition, 4, 548 –552. sentations on vocabulary learning. Journal of Educational Psychology,
*Bregman, A. S. (1967). Distribution of practice and between-trials inter- 79, 162–170.
ference. Canadian Journal of Psychology, 21, 1–14. Dempster, F. N. (1987b). Time and the production of classroom learning:
Briggs, D. C. (2005). Meta-analysis: A case study. Evaluation Review, 29, Discerning implications from basic research. Educational Psychologist,
87–127. 22, 1–21.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

*Brown, J., & Huda, M. (1961). Response latencies produced by massed Dempster, F. N. (1988). The spacing effect: A case study in the failure to
This document is copyrighted by the American Psychological Association or one of its allied publishers.

and spaced learning of a paired-associates list. Journal of Experimental apply the results of psychological research. American Psychologist, 43,
Psychology, 61, 360 –364. 627– 634.
*Brown, W. (1924). Effects of interval on recall. Journal of Experimental Dempster, F. N. (1989). Spacing effects and their implications for theory
Psychology, 7, 469 – 474. and practice. Educational Psychology Review, 1, 309 –330.
*Burtt, H. E., & Dobell, E. M. (1925). The curve of forgetting for *DeRemer, P., & D’Agostino, P. R. (1974). Locus of distributed lag effect
advertising material. Journal of Applied Psychology, 9, 5–21. in free recall. Journal of Verbal Learning and Verbal Behavior, 13,
*Carter, J. F., & Carrier, C. (1976). Prose organization and recall. Con- 167–171.
temporary Educational Psychology, 1, 329 –345. Donovan, J. J., & Radosevich, D. J. (1999). A meta-analytic review of the
*Cepeda, N. J., Mozer, M. C., Coburn, N., Rohrer, D., Wixted, J. T., & distribution of practice effect. Journal of Applied Psychology, 84, 795–
Pashler, H. (2005). Optimizing distributed practice: Theoretical analysis 805.
and practical implications. Manuscript submitted for publication. Doré, L. R., & Hilgard, E. R. (1938). Spaced practice as a test of Snoddy’s
*Cermak, L. S., Verfaellie, M., Lanzoni, S., Mather, M., & Chase, K. A. two processes in mental growth. Journal of Experimental Psychology,
(1996). Effect of spaced repetitions on amnesia patients’ recall and 23, 359 –374.
recognition performance. Neuropsychology, 10, 219 –227. Dunlap, W. P., Cortina, J. M., Vaslow, J. B., & Burke, M. J. (1996).
*Challis, B. H. (1993). Spacing effects on cued-memory tests depend on Meta-analysis of experiments with matched groups or repeated measures
level of processing. Journal of Experimental Psychology: Learning, designs. Psychological Methods, 1, 170 –177.
Memory, and Cognition, 19, 389 –396. *Durgunoğlu, A. Y., Mir, M., & Ariño-Martı́, S. (1993). Effects of re-
*Childers, J. B., & Tomasello, M. (2002). Two-year-olds learn novel peated readings on bilingual and monolingual memory for text. Con-
nouns, verbs, and conventional actions from massed or distributed temporary Educational Psychology, 18, 294 –317.
exposures. Developmental Psychology, 38, 967–978. Ebbinghaus, H. (1964). Memory: A contribution to experimental psychol-
*Ciccone, D. S., & Brelsford, J. W. (1976). Spacing repetitions in paired- ogy (H. A. Ruger, C. E. Bussenius, & E. R. Hilgard, Trans.). New York:
associate learning: Experimenter versus subject control. Journal of Ex- Dover Publications. (Original work published 1885)
perimental Psychology: Human Learning and Memory, 2, 446 – 455. *Edwards, A. S. (1917). The distribution of time in learning small amounts
*Clark, B. E. (1928). The effect upon retention of varying lengths of study of material. In Studies in Psychology Contributed by Colleagues and
periods and rest intervals in distributed learning time. Journal of Edu- Former Students of Edward Bradford Titchener (pp. 209 –213). Worces-
cational Psychology, 19, 552–559. ter, MA: Louis N. Wilson.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences *Elmes, D. G., Chapman, P. F., & Selig, C. W. (1984). Role of mood and
(2nd ed.). Hillsdale, NJ: Erlbaum. connotation in the spacing effect. Bulletin of the Psychonomic Society,
Cortina, J. M., & Nouri, H. (2000). Effect sizes for ANOVA designs. 22, 186 –188.
Thousand Oaks, CA: Sage. *Elmes, D. G., Dye, C. J., & Herdelin, N. J. (1983). What is the role of
Crowder, R. G. (1976). Principles of learning and memory. Hillsdale, NJ: affect in the spacing effect? Memory & Cognition, 11, 144 –151.
Erlbaum. *Elmes, D. G., Greener, W. I., & Wilkinson, W. C. (1972). Free recall of
*Cuddy, L. J., & Jacoby, L. L. (1982). When forgetting helps memory: An items presented after massed- and distributed-practice items. American
analysis of repetition effects. Journal of Verbal Learning and Verbal Journal of Psychology, 85, 237–240.
Behavior, 21, 451– 467. *Elmes, D. G., Sanders, L. W., & Dovel, J. C. (1973). Isolation of massed-
*Cull, W. L. (1995). How and when should information be restudied? and distributed-practice items. Memory & Cognition, 1, 77–79.
(Doctoral dissertation, Loyola University Chicago, 1995). Dissertation *Evans, J. D. (1975). The lag effect in free recall: Differential encoding or
Abstracts International, 56, 549. differential rehearsal? (Doctoral dissertation, Iowa State University,
*Cull, W. L. (2000). Untangling the benefits of multiple study opportuni- 1974). Dissertation Abstracts International, 35, 4142.
ties and repeated testing for cued recall. Applied Cognitive Psychology, *Fishman, E. J., Keller, L., & Atkinson, R. C. (1968). Massed versus
14, 215–235. distributed practice in computerized spelling drills. Journal of Educa-
*Cull, W. L., Shaughnessy, J. J., & Zechmeister, E. B. (1996). Expanding tional Psychology, 59, 290 –296.
understanding of the expanding-pattern-of-retrieval mnemonic: Toward *Folarin, B. A. (1983). The effect of spacing category members on chil-
confidence in applicability. Journal of Experimental Psychology: Ap- dren’s memory. Journal of Psychology, 114, 167–177.
plied, 2, 365–378. *Foos, P. W., & Smith, K. H. (1974). Effects of spacing and spacing
*D’Agostino, P. R. (1974). Repetition and recall in an incidental-memory patterns in free recall. Journal of Experimental Psychology, 103, 112–
paradigm. Canadian Journal of Psychology, 28, 468 – 473. 116.
*D’Agostino, P. R., & DeRemer, P. (1972). Item repetition in free and cued *Gagné, R. M. (1969). Context, isolation, and interference effects on the
recall. Journal of Verbal Learning and Verbal Behavior, 11, 54 –58. retention of fact. Journal of Educational Psychology, 60, 408 – 414.
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 373

*Gartman, L. M., & Johnson, N. F. (1972). Massed versus distributed Hintzman, D. L. (1974). Theoretical implications of the spacing effect. In
repetition of homographs: A test of the differential-encoding hypothesis. R. L. Solso (Ed.), Theories in cognitive psychology: The Loyola sym-
Journal of Verbal Learning and Verbal Behavior, 11, 801– 808. posium (pp. 77–97). Potomac, MD: Erlbaum.
Gibbons, R. D., Hedeker, D. R., & Davis, J. M. (1993). Estimation of effect Hintzman, D. L., Block, R. A., & Summers, J. J. (1973). Modality tags and
size from a series of experiments involving paired comparisons. Journal memory for repetitions: Locus of the spacing effect. Journal of Verbal
of Educational Statistics, 18, 271–279. Learning and Verbal Behavior, 12, 229 –238.
*Gilliland, T. R. (2000). Intervening tasks and the spacing effect: Theo- *Hintzman, D. L., & Stern, L. D. (1977). Failure to confirm Elmes,
retical and practical implications (Doctoral dissertation, Texas A&M Greener, and Wilkinson’s finding on the spacing effect. American Jour-
University, 1999). Dissertation Abstracts International, 60, 4920. nal of Psychology, 90, 489 – 497.
*Glanzer, M. (1969). Distance between related words in free recall: Trace Hintzman, D. L., Summers, J. J., Eki, N. T., & Moore, M. D. (1975).
of the STS. Journal of Verbal Learning and Verbal Behavior, 8, 105– Voluntary attention and the spacing effect. Memory & Cognition, 3,
111. 576 –580.
*Glanzer, M., & Duarte, A. (1971). Repetition between and within lan- *Hohn, R. L. (1972). Effects of massed vs. distributed practice and word
guages in free recall. Journal of Verbal Learning and Verbal Behavior, frequency on young children’s free recall (Rep. No. PS005604). Re-
10, 625– 630. trieved March 20, 2006, from ERIC database: http://eric.ed.gov/
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Glass, G. V., McGaw, B., & Smith, M. L. (1981). Meta-analysis in social Hollingworth, H. L. (1913). Advertising and selling: Principles of appeal
This document is copyrighted by the American Psychological Association or one of its allied publishers.

research. Beverly Hills, CA: Sage. and response. New York: D. Appleton.
*Glenberg, A. M. (1976). Monotonic and nonmonotonic lag effects in *Houston, J. P. (1966). List differentiation and distributed practice. Journal
paired-associate and recognition memory paradigms. Journal of Verbal of Experimental Psychology, 72, 477– 478.
Learning and Verbal Behavior, 15, 1–16. *Houston, J. P., & Reynolds, J. H. (1965). First-list retention as a function
*Glenberg, A. M. (1977). Influences of retrieval processes on the spacing of list differentiation and second-list massed and distributed practice.
effect in free recall. Journal of Experimental Psychology: Human Learn- Journal of Experimental Psychology, 69, 387–392.
ing and Memory, 3, 282–294. *Hovland, C. I. (1940). Experimental studies in rote-learning theory: VI.
*Glenberg, A. M. (1979). Component-levels theory of the effects of Comparison of retention following learning to same criterion by massed
spacing of repetitions on recall and recognition. Memory & Cognition, 7, and distributed practice. Journal of Experimental Psychology, 26, 568 –
95–112. 587.
*Glenberg, A. M., & Lehmann, T. S. (1980). Spacing repetitions over 1 *Hser, Y.-I., & Wickens, T. D. (1989). The effects of the spacing of test
week. Memory & Cognition, 8, 528 –538. trials and study trials in paired-association learning. Educational Psy-
*Glenberg, A. M., & Melton, A. W. (1978). An analysis of the free recall chology, 9, 99 –120.
spacing effect using the overt rehearsal technique. Unpublished manu- Hull, C. L. (1943). Principles of behavior. New York: Appleton-Century.
script, University of Wisconsin—Madison and University of Michigan, Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis: Cor-
Ann Arbor. recting error and bias in research findings. Newbury Park, CA: Sage.
*Glenberg, A. M., & Smith, S. M. (1981). Spacing repetitions and solving Irion, A. L. (1966). A brief history of research on the acquisition of skill.
problems are not the same. Journal of Verbal Learning and Verbal In E. A. Bilodeau (Ed.), Acquisition of skill (pp. 1– 46). New York:
Behavior, 20, 110 –119. Academic Press.
Gleser, L. J., Olkin, I. (1994). Stochastically dependent effect sizes. In H. *Izawa, C. (1970). List versus items in distributed practice in paired-
Cooper & L. V. Hedges (Eds.), The handbook of research synthesis (pp. associate learning. Proceedings of the Annual Convention of the Amer-
261–281). New York: Russell Sage Foundation. ican Psychological Association, 5, 87– 88.
*Gordon, K. (1925). Class results with spaced and unspaced memorizing. *Izawa, C. (1971). Massed and spaced practice in paired-associate leaning:
Journal of Experimental Psychology, 7, 337–343. List versus item distributions. Journal of Experimental Psychology, 89,
*Greene, R. L. (1989). Spacing effects in memory: Evidence for a two- 10 –21.
process account. Journal of Experimental Psychology: Learning, Mem- *Izawa, C. (1979). Toward a comprehensive theory of variable perfor-
ory, and Cognition, 15, 371–377. mance differences and differential spaced practice effects between an-
*Greene, R. L. (1990). Spacing effects on implicit memory tests. Journal ticipation and study-test methods. Journal of General Psychology, 100,
of Experimental Psychology: Learning, Memory, and Cognition, 16, 63– 83.
1004 –1011. *Jacoby, L. L. (1978). On interpreting the effects of repetition: Solving a
Greene, R. L. (1992). Human memory: Paradigms and paradoxes. Hills- problem versus remembering a solution. Journal of Verbal Learning and
dale, NJ: Erlbaum. Verbal Behavior, 17, 649 – 667.
*Hall, J. W. (1992). Unmixed effects of spacing on free recall. Journal of Janiszewski, C., Noel, H., & Sawyer, A. G. (2003). A meta-analysis of the
Experimental Psychology: Learning, Memory, and Cognition, 18, 608 – spacing effect in verbal learning: Implications for research on advertis-
614. ing repetition and consumer memory. Journal of Consumer Research,
*Hall, J. W., Smith, T. A., Wegener, S. L., & Underwood, B. J. (1981). 30, 138 –149.
Rate and frequency as determinants of learning with complete and *Jensen, T. D., & Freund, J. S. (1981). Persistence of the spacing effect in
discrete list presentation. Memory & Cognition, 9, 360 –367. incidental free recall: The effect of external list comparisons and inter-
*Harzem, P., Lee, I., & Miles, T. R. (1976). The effects of pictures on task correlations. Bulletin of the Psychonomic Society, 18, 183–186.
learning to read. British Journal of Educational Psychology, 46, 318 – Johnson, B. T., & Eagly, A. H. (2000). Quantitative synthesis of social
322. psychological research. In H. T. Reis & C. M. Judd (Eds.), Handbook of
*Hayes-Roth, B. (1976). Effects of repetitions and questions at varying research methods in social and personality psychology (pp. 496 –528).
lags during self-paced learning from text (Rand Paper P-5925). Santa New York: Cambridge University Press.
Monica, CA: Rand Corporation. *Johnston, W. A., Coots, J. H., & Flickinger, R. G. (1972). Controlled
*Hecht, D. J. (1982). Another ten years of massed practice on distributed semantic encoding and the effect of repetition lag on free recall. Journal
practice (Doctoral dissertation, University of Texas at Austin, 1982). of Verbal Learning and Verbal Behavior, 11, 784 –788.
Dissertation Abstracts International, 43, 899. *Johnston, W. A., & Uhl, C. N. (1976). The contributions of encoding
Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. effort and variability to the spacing effect on free recall. Journal of
Orlando, FL: Academic Press. Experimental Psychology: Human Learning and Memory, 2, 153–160.
 374 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

Jost, A. (1897). Die Assoziationsfestigkeit in ihrer Abhängigkeit von der terms on retention in paired-associate learning. Psychological Reports,
Verteilung der Wiederholungen [The strength of associations in their 33, 307–311.
dependence on the distribution of repetitions]. Zeitschrift für Psycholo- *McAllister, E. W. C. (1976). Effects of associative reaction time and
gie und Physiologie der Sinnesorgane, 14, 436 – 472. spaced presentations of stimulus-test items, response-test items, and
*Kahana, M. J., & Greene, R. L. (1993). Effects of spacing on memory for stimulus-response repetitions on retention in paired associate learning.
homogenous lists. Journal of Experimental Psychology: Learning, Mem- Bulletin of the Psychonomic Society, 7, 205–207.
ory, and Cognition, 19, 159 –162. *McAllister, E. W. C., & Ley, R. (1972). Effects of spaced presentations
*Kahana, M. J., Kamins, B. J., & Howard, M. W. (2003). Spacing and lag of stimulus terms, response terms, and stimulus-response pairs on re-
effects are obtained in free recall of pure lists. Unpublished manuscript. tention in paired-associate learning. Journal of Experimental Psychol-
*Keppel, G. (1964). Facilitation in short- and long-term retention of paired ogy, 94, 68 –73.
associates following distributed practice in learning. Journal of Verbal McBride, D. M., & Dosher, B. A. (1997). A comparison of forgetting in an
Learning and Verbal Behavior, 3, 91–111. implicit and explicit memory task. Journal of Experimental Psychology:
*Keppel, G. (1967). A reconsideration of the extinction-recovery theory. General, 126, 371–392.
Journal of Verbal Learning and Verbal Behavior, 6, 476 – 487. *McCormack, P. D., & Carboni, N. L. (1973). Lag invariance with forced
*Kitao, N., & Inoue, T. (1998). The effects of spaced repetition on explicit encodings in free recall. Canadian Journal of Psychology, 27, 144 –151.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

and implicit memory. Psychologia, 41, 114 –119. *McCullough, M. H. (1980). The role of repetition and context in the free
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Kitson, H. D. (1921). The mind of the buyer: A psychology of selling. New recall of pictures and words. Dissertation Abstracts International, 40 (7),
York: MacMillan. 3452B. (UMI No. AAT 8000423)
*Kraft, R. N., & Jenkins, J. J. (1981). The lag effect with aurally presented McDaniel, M. A., & Pressley, M. (1984). Putting the keyword method in
passages. Bulletin of the Psychonomic Society, 17, 132–134. context. Journal of Educational Psychology, 76, 598 – 609.
*Krug, D., Davis, B., & Glover, J. A. (1990). Massed versus distributed *McFarland, C. E., Jr., Rhodes, D. D., & Frey, T. J. (1979). Semantic-
repeated reading: A case of forgetting helping recall? Journal of Edu- feature variability and the spacing effect. Journal of Verbal Learning
cational Psychology, 82, 366 –371. and Verbal Behavior, 18, 163–172.
*Landauer, T. K. (1967). Interval between item repetitions and free recall McGeoch, J. A., & Irion, A. L. (1952). The distribution of practice and
memory. Psychonomic Science, 8, 439 – 440. reminiscence. In The psychology of human learning (2nd ed., pp. 138 –
*Landauer, T. K., & Bjork, R. A. (1978). Optimum rehearsal patterns and 193). New York: David McKay.
name learning. In M. M. Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), *Melton, A. W. (1970). The situation with respect to the spacing of
Practical aspects of memory (pp. 625– 632). London: Academic Press. repetitions and memory. Journal of Verbal Learning and Verbal Behav-
*Landauer, T. K., & Eldridge, L. (1967). Effect of tests without feedback ior, 9, 596 – 606.
and presentation-test interval in paired-associate learning. Journal of *Modigliani, V. (1967). Effects on a later recall by delaying initial recall.
Experimental Psychology, 75, 290 –298. Journal of Experimental Psychology: Human Learning and Memory, 2,
*Landauer, T. L., & Rubin, N. (1966). Spacing of item repetitions in 609 – 622.
paired-associate learning. Proceedings of the Annual Convention of the Morris, S. B. (2000). Distribution of the standardized mean change effect
American Psychological Association, 91–92. size for meta-analysis on repeated measures. British Journal of Mathe-
*Lee, C. L. (1976). Repetition and acoustic contrast in short-term memory matical and Statistical Psychology, 53, 17–29.
for letter sequences. Journal of Experimental Psychology: Human Morris, S. B., & DeShon, R. P. (2002). Combining effect size estimates in
Learning and Memory, 2, 695–704. meta-analysis with repeated measures and independent-groups designs.
Lee, T. D., & Genovese, E. D. (1988). Distribution of practice in motor Psychological Methods, 7, 105–125.
skill acquisition: Learning and performance effects reconsidered. Re- Moss, V. D. (1996). The efficacy of massed versus distributed practice as
search Quarterly for Exercise and Sport, 59, 277–287. a function of desired learning outcomes and grade level of the student
*Leng, N. R. C., & Parkin, A. J. (1989). Aetiological variation in the (Doctoral dissertation, Utah State University, 1995). Dissertation Ab-
amnesic syndrome: Comparisons using the Brown-Peterson task. Cor- stracts International, 56, 5204.
tex, 25, 251–259. *Murray, J. T. (1983). Spacing phenomena in human memory: A study-
*Madigan, S. A. (1970). Intraserial repetition and coding processes in free phase retrieval interpretation (Doctoral dissertation, University of Cali-
recall (Doctoral dissertation, University of Toronto, 1968). Dissertation fornia, Los Angeles, 1982). Dissertation Abstracts International, 43,
Abstracts International, 31, 933–934. 3058.
*Madsen, M. C. (1963). Distribution of practice and level of intelligence. *Nelson, T. O. (1977). Repetition and depth of processing. Journal of
Psychological Reports, 13, 39 – 42. Verbal Learning and Verbal Behavior, 16, 151–171.
*Madsen, M. C. (1966). Individual differences and temporal factors in *Paivio, A. (1974). Spacing of repetitions in the incidental and intentional
memory consolidation. American Journal of Mental Deficiency, 71, free recall of pictures and words. Journal of Verbal Learning and Verbal
501–507. Behavior, 13, 497–511.
Maki, R. H., & Hasher, L. (1975). Encoding variability: A role in imme- *Pashler, H., & Cepeda, N. J. (2003). [Within-subjects replication of
diate and long term memory? American Journal of Psychology, 88, Glenberg and Lehmann (1980), Experiment 2]. Unpublished raw data.
217–231. Pashler, H., Cepeda, N. J., Wixted, J. T., & Rohrer, D. (2005). When does
*Marmurek, H. H. C., Holt, P. D., & Coe, K. (1978). Presentation mode feedback facilitate learning of words? Journal of Experimental Psychol-
and repetition effects in free recall. American Journal of Psychology, 91, ogy: Learning, Memory, and Cognition, 31, 3– 8.
483– 490. *Pashler, H., Zarow, G., & Triplett, B. (2003). Is temporal spacing of tests
*Maskarinec, A. S., & Thompson, C. P. (1976). The within-list distributed helpful even when it inflates error rates? Journal of Experimental
practice effect: Tests of the varied context and varied encoding hypoth- Psychology: Learning, Memory, and Cognition, 29, 1051–1057.
eses. Memory & Cognition, 4, 741–746. *Patten, E. F. (1938). The influence of distribution of repetitions on certain
*Mathews, R. C., & Tulving, E. (1973). Effects of three types of repetition rote learning phenomena. Journal of Psychology, 5, 359 –374.
on cued and noncued recall of words. Journal of Verbal Learning and *Pavlik, P. I., & Anderson, J. R. (2003). An ACT-R model of the spacing
Verbal Behavior, 12, 707–721. effect. In F. Detje, D. Dorner, & H. Schaub (Eds.), Proceedings of the
*McAllister, E. W. C. (1973). Effects of type of recall test and spaced Fifth International Conference of Cognitive Modeling (pp. 177–182).
presentations of stimulus terms, response terms and stimulus-response Bamberg, Germany: Universitats-Verlag Bamberg.
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 375

*Perruchet, P. (1989). The effect of spaced practice on explicit and implicit *Shaughnessy, J. J. (1976). Persistence of the spacing effect in free recall
memory. British Journal of Psychology, 80, 113–130. under varying incidental learning conditions. Memory & Cognition, 4,
*Peterson, L. R., & Gentile, A. (1965). Proactive interference as a function 369 –377.
of time between tests. Journal of Experimental Psychology, 70, 473– *Shaughnessy, J. J. (1977). Long-term retention and the spacing effect in
478. free-recall and frequency judgments. American Journal of Psychology,
*Peterson, L. R., Hillner, K., & Saltzman, D. (1962). Supplementary 90, 587–598.
report: Time between pairings and short-term retention. Journal of *Shaughnessy, J. J., Zimmerman, J., & Underwood, B. J. (1972). Further
Experimental Psychology, 64, 550 –551. evidence on the MP-DP effect in free-recall learning. Journal of Verbal
*Peterson, L. R., Saltzman, D., Hillner, K., & Land, V. (1962). Recency Learning and Verbal Behavior, 11, 1–12.
and frequency in paired-associate learning. Journal of Experimental *Shaughnessy, J. J., Zimmerman, J., & Underwood, B. J. (1974). The
Psychology, 63, 396 – 403. spacing effect in the learning of word pairs and the components of word
*Peterson, L. R., Wampler, R., Kirkpatrick, M., & Saltzman, D. (1963). pairs. Memory & Cognition, 2, 742–748.
Effect of spacing presentations on retention of a paired associate over *Shuell, T. J. (1981). Distribution of practice and retroactive inhibition in
short intervals. Journal of Experimental Psychology, 66, 206 –209. free-recall learning. Psychological Record, 31, 589 –598.
Postman, L., & Knecht, K. (1983). Encoding variability and retention. *Siegel, M. A., & Misselt, A. L. (1984). Adaptive feedback and review
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Journal of Verbal Learning and Verbal Behavior, 22, 133–152. paradigm for computer-based drills. Journal of Educational Psychology,
This document is copyrighted by the American Psychological Association or one of its allied publishers.

*Potts, G. R., & Shiffrin, R. M. (1970). Repetitions, blank trials, and the 76, 310 –317.
vonRestorff effect in free recall memory. Journal of Experimental Psy- *Silverstein, L. D. (1978). Repetition and distribution effects on memory:
chology, 86, 128 –130. A psychophysiological analysis (Doctoral dissertation, University of
Pyle, W. H. (1913). Economical learning. Journal of Educational Psychol- Florida, 1977). Dissertation Abstracts International, 38, 5618.
ogy, 4, 148 –158. *Simon, J. L. (1979). What do Zielske’s real data really show about
Raaijmakers, J. G. W. (2003). Spacing and repetition effects in human pulsing? Journal of Marketing Research, 16, 415– 420.
memory: Application of the SAM model. Cognitive Science, 27, 431– *Singh, S. N., Mishra, S., Bendapudi, N., & Linville, D. (1994). Enhancing
452. memory of television commercials through message spacing. Journal of
*Rea, C. P., & Modigliani, V. (1985). The effect of expanded versus Marketing Research, 31, 384 –392.
massed practice on the retention of multiplication facts and spelling lists. *Spitzer, H. F. (1939). Studies in retention. Journal of Educational Psy-
Human Learning, 4, 11–18. chology, 30, 641– 656.
*Rea, C. P., & Modigliani, V. (1987). The spacing effect in 4- to 9-year-old *Strong, E. C. (1973). The effects of repetition in advertising: A field
children. Memory & Cognition, 15, 436 – 443. experiment (Doctoral dissertation, Stanford University, 1972). Disser-
Reed, A. V. (1977). Quantitative prediction of spacing effects in learning. tation Abstracts International, 33, 4615.
Journal of Verbal Learning and Verbal Behavior, 16, 693– 698. *Strong, E. K., Jr. (1916). The factors affecting a permanent impression
*Robbins, D., & Bray, J. F. (1974). The spacing effect and the A-B, A-C developed through repetition. Journal of Experimental Psychology, 1,
paradigm: Evidence for retroactive facilitation. Journal of Experimental 319 –338.
Psychology, 103, 420 – 425. *Stroud, J. B., & Johnson, E. (1941). /1942). The temporal position of
*Robinson, E. S. (1921). The relative efficiencies of distributed and con- reviews. Journal of Educational Research, 35, 618 – 622.
centrated study in memorizing. Journal of Experimental Psychology, 4, SuperMemo World. (n.d.). Super memory: Forget about forgetting. Re-
327–343. trieved March 16, 2006, from http://www.supermemo.com/
*Rohrer, D., Taylor, K., Pashler, H., Wixted, J. T., & Cepeda, N. J. (2005). Taylor, M. J., & White, K. R. (1992). An evaluation of alternative methods
The effect of overlearning on long-term retention. Applied Cognitive for computing standardized mean difference effect size. Journal of
Psychology, 19, 361–374. Experimental Education, 61, 63–72.
*Rose, R. J. (1984). Processing time for repetitions and the spacing effect. *Thios, S. J. (1972a). The effect of spacing of repetitions and variation of
Canadian Journal of Psychology, 83, 537–550. context on memory for discrete events (Doctoral dissertation, University
*Rose, R. J. (1992). Degree of learning, interpolated tests, and rate of of Virginia, 1971). Dissertation Abstracts International, 32, 4909.
forgetting. Memory & Cognition, 20, 621– 632. *Thios, S. J. (1972b). Memory for words in repeated sentences. Journal of
Rosenthal, R. (1979). The file drawer problem and tolerance for null Verbal Learning and Verbal Behavior, 11, 789 –793.
results. Psychological Bulletin, 86, 638 – 641. *Thios, S. J., & D’Agostino, P. R. (1976). Effects of repetition as a
*Ross, B. H., & Landauer, T. K. (1978). Memory for at least one of two function of study-phase retrieval. Journal of Verbal Learning and Verbal
items: Test and failure of several theories of spacing effects. Journal of Behavior, 15, 529 –536.
Verbal Learning and Verbal Behavior, 17, 669 – 680. Thorndike, E. L. (1912). The curve of work. Psychological Review, 19,
Ruch, T. C. (1928). Factors influencing the relative economy of massed 165–194.
and distributed practice in learning. Psychological Review, 35, 19 – 45. *Toppino, T. C. (1991). The spacing effect in young children’s free recall:
*Rumelhart, D. E. (1968). The effects of interpresentation intervals on Support for automatic-process explanations. Memory & Cognition, 19,
performance in a continuous paired-associate task (Doctoral dissertation, 159 –167.
Stanford University, 1967). Dissertation Abstracts International, 28, *Toppino, T. C. (1993). The spacing effect in preschool children’s free
4782. recall of pictures and words. Bulletin of the Psychonomic Society, 31,
*Russo, R., Parkin, A. J., Taylor, S. R., & Wilks, J. (1998). Revising 27–30.
current two-process accounts of spacing effects in memory. Journal of *Toppino, T. C., & DeMesquita, M. (1984). Effects of spacing repetitions
Experimental Psychology: Learning, Memory, and Cognition, 24, 161– on children’s memory. Journal of Experimental Child Psychology, 37,
172. 637– 648.
*Schwartz, M. (1975). The effect of constant vs. varied encoding and *Toppino, T. C., & DiGeorge, W. (1984). The spacing effect in free recall
massed vs. distributed presentations on recall of paired associates. Mem- emerges with development. Memory & Cognition, 12, 118 –122.
ory & Cognition, 3, 390 –394. *Toppino, T. C., & Gracen, T. F. (1985). The lag effect and differential
Shadish, W. R., & Haddock, C. K. (1994). Combining estimates of effect organization theory: Nine failures to replicate. Journal of Experimental
size. In H. Cooper & L. V. Hedges (Eds.), The handbook of research Psychology: Learning, Memory, and Cognition, 11, 185–191.
synthesis (pp. 261–281). New York: Russell Sage Foundation. *Toppino, T. C., Hara, Y., & Hackman, J. (2002). The spacing effect in the
 376 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

free recall of homogenous lists: Present and accounted for. Memory & tice: XVII. Interlist interference and the retention of paired consonant
Cognition, 30, 601– 606. syllables. Journal of Experimental Psychology, 54, 274 –279.
*Toppino, T. C., & Schneider, M. A. (1999). The mix-up regarding mixed *von Wright, J. M. (1971). Effects of distributed practice and distributed
and unmixed lists in spacing-effect research. Journal of Experimental recall tests on later recall of paired associates. Journal of Verbal Learn-
Psychology: Learning, Memory, and Cognition, 25, 1071–1076. ing and Verbal Behavior, 10, 311–315.
*Tsai, L.-S. (1927). The relation of retention to the distribution of relearn- *Watts, D., & Chatfield, D. (1976). Response availability as a function of
ing. Journal of Experimental Psychology, 10, 30 –39. massed and distributed practice and list differentiation. Psychological
*Tzeng, O. J. L. (1973). Stimulus meaningfulness, encoding variability, Record, 26, 487– 493.
and the spacing effect. Journal of Experimental Psychology, 99, 162– *Waugh, N. C. (1967). Presentation time and free recall. Journal of
166. Experimental Psychology, 73, 39 – 44.
*Underwood, B. J. (1951). Studies of distributed practice: II. Learning and *Waugh, N. C. (1970). On the effective duration of a repeated word.
retention of paired-adjective lists with two levels of intra-list similarity. Journal of Verbal Learning and Verbal Behavior, 9, 587–595.
Journal of Experimental Psychology, 42, 153–161. *Welborn, E. L. (1933). A study of logical learning in college classes. 20th
*Underwood, B. J. (1952a). Studies of distributed practice: VI. The influ- Annual Conference on Educational Measurement, Indiana University
ence of rest-interval activity in serial learning. Journal of Experimental School of Education Bulletin, 10, 12–20.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

*Wells, J. E., & Kirsner, K. (1974). Repetition between and within mo-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Psychology, 43, 329 –340.

*Underwood, B. J. (1952b). Studies of distributed practice: VII. Learning dalities in free recall. Bulletin of the Psychonomic Society, 2, 395–397.
and retention of serial nonsense lists as a function of intralist similarity. *Wenger, S. K. (1979). The within-list distributed practice effect: More
Journal of Experimental Psychology, 44, 80 – 87. evidence for the inattention hypothesis. American Journal of Psychol-
ogy, 92, 105–113.
*Underwood, B. J. (1953). Studies of distributed practice: XI. An attempt
Wickelgren, W. A. (1972). Trace resistance and the decay of long-term
to resolve conflicting facts on retention of serial nonsense lists. Journal
memory. Journal of Mathematical Psychology, 9, 418 – 455.
of Experimental Psychology, 45, 355–359.
*Wilson, J. T. (1949). The formation and retention of remote associations
*Underwood, B. J. (1954). Studies of distributed practice: XII. Retention
in rote learning. Journal of Experimental Psychology, 39, 830 – 838.
following varying degrees of original learning. Journal of Experimental
*Wilson, W. P. (1976). Developmental changes in the lag effect: An
Psychology, 47, 294 –300.
encoding hypothesis for repeated word recall. Journal of Experimental
Underwood, B. J. (1961). Ten years of massed practice on distributed
Child Psychology, 22, 113–122.
practice. Psychological Review, 68, 229 –247.
Wixted, J. T. (2004). On common ground: Jost’s (1897) law of forgetting
*Underwood, B. J. (1969). Some correlates of item repetition in free-recall
and Ribot’s (1881) law of retrograde amnesia. Psychological Review,
learning. Journal of Verbal Learning and Verbal Behavior, 8, 83–94. 111, 864 – 879.
*Underwood, B. J. (1970). A breakdown of the total-time law in free-recall Wixted, J. T., & Ebbesen, E. B. (1997). Genuine power curves in forget-
learning. Journal of Verbal Learning and Verbal Behavior, 9, 573–580. ting: A quantitative analysis of individual subject forgetting functions.
Underwood, B. J., & Ekstrand, B. R. (1967). Effect of distributed practice Memory & Cognition, 25, 731–739.
on paired-associate learning. Journal of Experimental Psychology, 73, Woźniak, P. A., & Gorzelańczyk, E. J. (1994). Optimization of repetition
1–21. spacing in the practice of learning. Acta Neurobiologiae Experimentalis,
*Underwood, B. J., Kapelak, S. M., & Malmi, R. A. (1976). The spacing 54, 59 – 62.
effect: Additions to the theoretical and empirical puzzles. Memory & *Wright, J., & Brelsford, J. (1978). Changes in the spacing effect with
Cognition, 4, 391– 400. instructional variables in free recall. American Journal of Psychology,
Underwood, B. J., & Keppel, G. (1963). Retention as a function of degree 91, 631– 643.
of learning and letter-sequence interference. Psychological Mono- *Young, J. L. (1966). Effects of intervals between reinforcements and test
graphs: General and Applied, 77 (4, Whole No. 567). trials in paired-associate learning (Doctoral dissertation, Stanford Uni-
*Underwood, B. J., Keppel, G., & Schulz, R. W. (1962). Studies of versity, 1966). Dissertation Abstracts International, 27, 3699.
distributed practice: XXII. Some conditions which enhance retention. *Zechmeister, E. B., & Shaughnessy, J. J. (1980). When you know that you
Journal of Experimental Psychology, 64, 355–363. know and when you think that you know but you don’t. Bulletin of the
*Underwood, B. J., & Richardson, J. (1955). Studies of distributed prac- Psychonomic Society, 15, 41– 44.
tice: XIII. Interlist interference and the retention of serial nonsense lists. *Zimmerman, J. (1975). Free recall after self-paced study: A test of the
Journal of Experimental Psychology, 50, 39 – 46. attention explanation of the spacing effect. American Journal of Psy-
*Underwood, B. J., & Richardson, J. (1957). Studies of distributed prac- chology, 88, 277–291.
 REVIEW OF THE DISTRIBUTED PRACTICE EFFECT 377

Appendix
Effects of Task Type on Distributed Practice

One lingering concern with our lag analyses is whether task type plays The typical primacy and recency buffers have been removed:
a role in the expression of joint effects between ISI and retention interval.
Put another way, is it reasonable to expect the joint effects of ISI difference i1 i2 i1 i2 i3 i4 i3 i4 i5 i6 i5 i6
and retention interval to be constant, regardless of task type? We can think
retention interval (time ⫽ x)
of no a priori reason to expect lag effects to vary on the basis of task type.
On the other hand, different experimental methodologies, which vary recall test (unlimited time given to complete test).
consistently with task type, might reduce our ability to glean the joint
effects of ISI difference and retention interval. Specifically, some para- The first feature to notice is that retention interval for items i1 and i2 is
digms provided consistent and accurate manipulation of ISI difference and
longer than retention interval for i5 and i6. This problem becomes worse
retention interval, and these well-controlled paradigms were used in most
when list length is long and retention interval is short. Also, we have
of the experiments with paired associate tasks. In most experiments with
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

presented a best-case scenario. Many list recall paradigms present items

This document is copyrighted by the American Psychological Association or one of its allied publishers.

paired associate tasks, items separated by a given lag were almost always
i1–i6, and then rerandomize item order before re-presenting the entire
followed by exactly the same retention interval. Thus, there is no question
list. This introduces even more variability, as ISI difference is then
that ISI and retention interval values used in this meta-analysis were
variable, as is retention interval. An additional, smaller, problem is that
accurate. In contrast, list recall paradigms did not accurately control ISI
difference and retention interval, so there is some degree of incorrectness giving unlimited time to recall means that retention interval becomes
in the ISI difference and retention interval values we used. To illustrate the more variable than if recall time were fixed, as occurs in many paired
problem, say items are represented by ix. The following is a sample list associate paradigms.
recall paradigm. Lag is always 1 item, and there are no filler items. (text continues on page 380)

Figure A1. For paired associate studies in the accuracy difference lag analyses, accuracy difference between
all adjacent pairs of interstudy interval (ISI) values from each study, binned by difference in ISI and retention
interval and averaged across studies. When surrounded by ISI bins with lower accuracy values, the ISI bin
showing the highest accuracy value at each retention interval bin is indicated with an asterisk. Error bars
represent one standard error of the mean.

(Appendix continues)
 378 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure A2. For list recall studies in the accuracy difference lag analyses, accuracy difference between all
adjacent pairs of interstudy interval (ISI) values from each study, binned by difference in ISI and retention
interval and averaged across studies. When surrounded by ISI bins with lower accuracy values, the ISI bin
showing the highest accuracy value at each retention interval bin is indicated with an asterisk. Error bars
represent one standard error of the mean.

Table A1
For Paired Associate Data, Shorter and Longer Interstudy Interval (ISI) Range, Retention
Interval Range, Percentage Correct at the Shorter and Longer ISI Range, and Statistical
Analyses, for Accuracy Difference Lag Analyses

ISI range % Correct at ISI range

Retention
Shorter Longer interval range Shorter Longer SEM Statistical analysis

1–10 s 11–29 s 4–59 s 1.1 2.5 1.6 t(53) ⫽ 0.6, p ⫽ .522

11–29 s 1–15 min 4–59 s 2.5 ⫺2.2 1.8 t(23) ⫽ 1.1, p ⫽ .280
1–10 s 11–29 s 1 min–2 hr 1.0 2.8 1.5 t(35) ⫽ 0.6, p ⫽ .554
11–29 s 2–28 days 1 min–2 hr 2.8 ⫺13.7 3.8 t(28) ⫽ 3.0, p ⬍ .01
30–59 s 1–15 min 1 day 1.3 8.3 3.4 t(8) ⫽ 1.5, p ⫽ .162
1–15 min 1 day 2–28 days 4.5 11.0 3.3 t(9) ⫽ 1.4, p ⫽ .194
1 day 2–28 days 2–28 days 11.0 0.2 2.8 t(10) ⫽ 2.5, p ⬍ .05
1 day 2–28 days 30–2,900 days 6.5 9.7 2.7 t(14) ⫽ 1.0, p ⫽ .356
2–28 days 29–84 days 30–2,900 days 9.7 ⫺0.6 2.6 t(16) ⫽ 3.2, p ⬍ .01
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure A3. For paired associate studies in the absolute lag analyses, accuracy, binned by interstudy interval
(ISI) and retention interval and averaged across studies. When surrounded by ISI bins with lower accuracy
values, the ISI bin showing the highest accuracy value at each retention interval bin is indicated with an asterisk.
Error bars represent one standard error of the mean.

Figure A4. For list recall studies in the absolute lag analyses, accuracy, binned by interstudy interval (ISI) and
retention interval and averaged across studies. When surrounded by ISI bins with lower accuracy values, the ISI
bin showing the highest accuracy value at each retention interval bin is indicated with an asterisk. Error bars
represent one standard error of the mean.
(Appendix continues)
 380 CEPEDA, PASHLER, VUL, WIXTED, AND ROHRER

Table A2
For Paired Associate Data, Shorter and Longer Interstudy Interval (ISI) Range, Retention
Interval Range, Percentage Correct at the Shorter and Longer ISI Range, and Statistical
Analyses, for Absolute Lag Analyses

ISI range % Correct at ISI Range

Retention
Shorter Longer interval range Shorter Longer SEM Statistical analysis

1–10 s 30–59 s 2–59 s 51.4 60.1 4.7 t(60) ⫽ 1.3, p ⫽ .183

30–59 s 1 min–3 hr 2–59 s 60.1 41.9 4.9 t(33) ⫽ 2.3, p ⬍ .05
1–10 s 1 min–3 hr 1 min–2 hr 35.9 56.3 3.8 t(67) ⫽ 3.8, p ⬍ .001
1 min–3 hr 2–28 days 1 min–2 hr 56.3 50.7 8.9 t(46) ⫽ 0.4, p ⫽ .664
11–29 s 2–28 days 2–28 days 29.0 55.5 9.8 t(16) ⫽ 1.9, p ⫽ .073
1 min–3 hr 29–168 days 30–2,900 days 27.0 50.3 11.5 t(12) ⫽ 1.4, p ⫽ .180
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

To assess the impact of these paradigmatic issues, we have reana- In Figures A3 and A4 we present joint effects of absolute ISI and
lyzed lag data, separating by task type. Figures A1 and A2 show joint retention interval, for paired associate and list recall data, respectively.
effects of ISI difference and retention interval, for paired associate and Table A2 provides quantitative analyses of joint effects of absolute ISI and
list recall data, respectively. Table A1 provides quantitative analyses of retention interval joint effects. The data once again support an increase in
joint effects of ISI difference and retention interval, for paired associate optimal ISI as retention interval increases.
data. As would be predicted by paradigmatic differences, paired asso-
ciate data paint a much cleaner qualitative picture of joint effects
between ISI difference and retention interval. Unfortunately, this Received April 14, 2004
cleaner qualitative picture comes with a less clean quantitative picture, Revision received July 20, 2005
because sample size, and thus power, is reduced as well. Accepted September 9, 2005 䡲