Retention

Abstract

Retention is the persistence of learned material across time: what survives in memory between the moment of learning and a later attempt to recover it. It is the middle term of memory, bracketed by encoding before and retrieval after, and it is measured by how much of an original acquisition is still available after a retention interval. The experimental study of retention began with Ebbinghaus, whose forgetting curve first described the orderly decline of the amount retained, and whose savings method remains the template for quantifying it. A century of work has since mapped the shape of that curve, debated whether forgetting reflects decay or interference, and identified the conditions, spacing, retrieval practice, and consolidation, that make retention durable. This article surveys how retention is measured, how it changes over time, and what determines whether learning lasts.

Keywords: retention, forgetting curve, retention interval, savings method, testing effect

Retention denotes the amount of previously learned material that remains available after an interval has elapsed (Ebbinghaus, 2013). It is not a separate faculty but a property of memory measured over time: the fraction of an acquisition that endures between study and test. Because retention can only be inferred from a later act of recall, recognition, or relearning, its study is inseparable from the methods used to probe it and from the retention interval across which forgetting is allowed to run.

What Retention Is

Retention is the persistence over time of the effects of learning. In the three-stage framing of memory, encoding registers an experience, retention holds it across an interval, and retrieval recovers it; retention is the phase during which a memory either survives or is lost. The complementary quantity is forgetting, the portion of an acquisition no longer recoverable, so that retention and forgetting sum to the original learning by definition. A retention function plots the amount retained against the length of the interval, and its downward slope is the rate of forgetting.

The construct is deliberately neutral about mechanism. To say that ninety percent of a list is retained after an hour is to state a measured fact about behaviour, not a claim about where the memory resides or why the other ten percent is gone. This separation of the measured quantity from its causes is what allowed retention to become an experimental variable in the first place: Ebbinghaus could chart how much survived long before anyone could say what the survivor was made of (Murre & Dros, 2015).

Measuring Retention

Three methods dominate, and they are ordered by sensitivity. Free recall asks the learner to reproduce the material with no cues and detects only memories strong enough to be regenerated unaided. Recognition presents the material and asks whether it was studied, crediting fainter traces that recall would miss, so recognition scores typically exceed recall for the same acquisition. The savings method, Ebbinghaus's own, is the most sensitive of the three: it relearns the material to the original criterion and measures the reduction in trials or time required. Savings often reveals retention that both recall and recognition report as zero, because a trace too weak to recover can still speed its own reacquisition.

Because the three methods tap different amounts of surviving memory, a retention interval that looks empty by recall may be substantially full by savings. This is why claims about how much has been forgotten are meaningless without naming the measure: the same head, at the same delay, retains different amounts depending on how the question is asked (Ebbinghaus, 2013).

The Shape of Retention Over Time

Ebbinghaus's central discovery was that forgetting is orderly. Plotting savings against delay, he found retention dropping sharply in the first hours after learning and then declining ever more slowly, a curve steep at the origin and nearly flat at long intervals. Replications with modern controls recover the same shape from his original nonsense-syllable paradigm (Murre & Dros, 2015).

The precise mathematical form has been contested. A review of more than two hundred retention datasets found that no single function fits universally, but that logarithmic and power functions describe the decline more accurately than the exponential a pure-decay account would predict (Rubin & Wenzel, 1996). The slow tail matters: some material, once learned well enough, resists forgetting for decades. Bahrick's study of Spanish learned in school found a stable residue, a permastore, that survived up to fifty years with little further loss once the first few years of forgetting had passed (Bahrick, 1984).

The curve is therefore not a single exponential fall to zero but a fast early loss followed by a durable remainder, whose size depends on how thoroughly the material was learned and how it has been rehearsed since.

What Determines Retention

Two families of explanation compete for why the curve falls at all. Decay theory holds that traces fade with the passage of time itself; interference theory holds that forgetting reflects competition from other learning, whether earlier material blocking later (proactive) or later material overwriting earlier (retroactive). Underwood's demonstration that much of what had been attributed to decay was in fact proactive interference from prior lists shifted the field decisively toward interference for a generation (Underwood, 1957). The question was sharpened at the short end by the Brown-Peterson paradigm, in which a few consonants are lost within roughly eighteen seconds once rehearsal is blocked, a rapid decline that looked like pure decay until interference accounts were extended to it (Peterson & Peterson, 1959). The debate remains genuinely open for these shortest intervals, where recent work has revived a role for time-based decay in immediate memory (Ricker, Vergauwe, & Cowan, 2016).

Beyond why memory is lost, a parallel literature identifies what makes it last. Wixted's integration of the psychology and neuroscience of forgetting frames retention as governed by consolidation: newly formed traces are labile and stabilise over time, so protecting the consolidation window improves durability (Wixted, 2004). Two manipulations reliably raise long-term retention. Distributing study across spaced sessions beats massing it into one, an advantage confirmed across hundreds of comparisons (Cepeda, Pashler, Vul, Wixted, & Rohrer, 2006). And testing beats restudy: retrieving material, rather than reading it again, produces markedly better retention at a delay, the testing effect (Roediger & Karpicke, 2006).

The testing effect is large and counterintuitive. In its canonical demonstration, learners who studied a passage once and then took repeated recall tests retained far more after a week than learners who restudied the passage the same number of times, even though the restudy group judged their own learning to be better (Karpicke & Roediger, 2008).

Worked Example

Consider the savings method applied to a single list. A learner masters a list of nonsense syllables to one perfect recitation, requiring twenty trials to reach that criterion. A day later, free recall of the list yields nothing: by the recall measure, retention is zero. The learner then relearns the same list to the same one-perfect-recitation criterion, and this time reaches it in twelve trials.

Savings is the proportional reduction in effort: the original twenty trials less the twelve now needed, divided by the original twenty, giving eight over twenty, or forty percent. So although recall reported zero retention, the savings method reveals that forty percent of the original learning in fact survived the day, latent but real, detectable only because it accelerated relearning. This gap between a recall score of zero and a savings score of forty percent is the concrete meaning of the claim that the measure defines the amount retained.

Current Directions

Three active fronts extend the classical picture. The first reinterprets the testing effect through consolidation: retrieval is now argued to be not merely a readout of memory but a fast route to stabilising it, engaging the same hippocampal-neocortical dynamics that slow consolidation exploits over sleep (Antony, Ferreira, Norman, & Wimber, 2017). On this view the durability that testing confers is a consolidation effect, not a separate mechanism.

The second is the accumulation of classroom evidence. A systematic meta-analytic review of quizzing in real courses confirms that the testing effect scales out of the laboratory, boosting durable classroom learning across grade levels and materials, though its size depends on feedback and quiz format (Yang, Luo, Vadillo, Yu, & Shanks, 2021). The third revisits the oldest question at the shortest timescale: whether the loss of information from immediate memory over a few seconds reflects decay or interference remains unresolved, and the modern reappraisal of decay theory argues the evidence does not cleanly exclude a time-based component (Ricker et al., 2016).

Key Researchers

Nelson Cowan (b. 1951). University of Missouri; he advanced the decay-versus-interference debate over what limits immediate memory and co-authored the modern reappraisal of decay theory. ORCID - Wikipedia

Hermann Ebbinghaus (1850-1909). University of Berlin and Breslau; he founded the experimental study of retention, introducing nonsense syllables, the savings method, and the forgetting curve. Wikipedia

Jeffrey D. Karpicke (b. 1978). Purdue University; he demonstrated that retrieval practice produces markedly better long-term retention than repeated study. ORCID

Henry L. Roediger III (b. 1947). Washington University in St. Louis; he established test-enhanced learning as a robust route to durable retention. Wikipedia

Benton J. Underwood (1915-1994). Northwestern University; he reframed forgetting as largely the product of interference from other learning rather than passive decay.

John T. Wixted (b. 1959). University of California, San Diego; he gave a quantitative account of the retention function and its dependence on consolidation. ORCID

Discussion

Retention matters because durability, not initial acquisition, is what distinguishes learning that lasts from learning that evaporates. A striking recurring finding is that the conditions which make acquisition feel easy, massing, rereading, and cramming, are precisely those that leave retention fragile, while the conditions that feel effortful, spacing and self-testing, are those that make it durable. This is the distinction between learning and performance: manipulations that boost performance during acquisition often fail to boost the retention that is the real object, and the effortful manipulations that depress momentary performance are frequently the ones that make learning last, so-called desirable difficulties (Soderstrom & Bjork, 2015). The dissociation between the subjective fluency of learning and its actual retention is why learners systematically mispredict what they will remember, and why evidence-based study advice so often contradicts intuition.

The practical reach is broad. The spacing and testing effects are among the most robust findings in the science of learning, and they translate directly into classroom and clinical practice. At the same time, the theoretical core, whether forgetting is decay or interference, and how the retention function should be modelled, remains genuinely unsettled, especially at the two extremes of the timescale: the seconds of immediate memory and the decades of permastore.

Glossary

Consolidation.

The time-dependent process by which a newly formed memory trace becomes progressively more stable and resistant to disruption.

Decay theory.

The account that memory traces weaken and fade as a function of the passage of time itself, independent of intervening events.

Desirable difficulties.

Conditions that slow or depress performance during learning yet improve long-term retention, such as spacing and retrieval practice.

Distributed practice.

Study spread across separated sessions, which yields better long-term retention than the same amount of study massed together.

Encoding.

The initial registration of an experience into memory, the stage that precedes retention.

Forgetting curve.

The function relating the amount retained to the length of the retention interval, steep at first and flattening thereafter.

Forgetting.

The loss over time of previously learned material, the complement of retention across a retention interval.

Interference theory.

The account that forgetting reflects competition from other learned material rather than the passage of time.

Permastore.

The durable residue of very well-learned material that survives for decades with little further loss once early forgetting has passed.

Proactive interference.

The disruption of retention for newer material by material learned earlier.

Recall.

A retention measure requiring the learner to reproduce studied material without cues, detecting only relatively strong traces.

Recognition.

A retention measure in which studied material is presented and judged as old or new, crediting fainter traces than recall.

Retention interval.

The elapsed time between the end of learning and the test of memory, across which forgetting is measured.

Retroactive interference.

The disruption of retention for earlier material by material learned later.

Savings.

Ebbinghaus's measure of retention as the proportional reduction in trials or time needed to relearn material to its original criterion.

Testing effect.

The finding that retrieving material through a test produces better long-term retention than an equivalent amount of restudy.

Frequently Asked Questions

What is the difference between retention and memory?
Memory is the general capacity to store and use information, while retention is the specific property of how much of a given acquisition persists across a retention interval (Wixted, 2004).

What is the difference between retention and retrieval?
Retention is the surviving trace held across the interval; retrieval is the later act of recovering it. A memory can be retained yet momentarily unretrievable, which is why the savings method finds retention that recall misses (Ebbinghaus, 2013).

Why do we forget so quickly at first?
The forgetting curve is steepest just after learning because the newest traces are the least consolidated and the most exposed to interference; once the fragile fraction is lost, the durable remainder declines far more slowly (Murre & Dros, 2015).

Is forgetting caused by decay or interference?
Both are supported in part. Interference explains much long-term forgetting, as Underwood showed, while a time-based decay component has been revived for the immediate memory of a few seconds; the question is not settled (Underwood, 1957).

What is the best way to make learning last?
Study is best spread across separated sessions and driven by self-testing rather than rereading. Distributed practice and retrieval practice are the two most reliable levers on durable retention (Cepeda, Pashler, Vul, Wixted, & Rohrer, 2006).

Why does testing help more than restudying?
Retrieving material strengthens and helps consolidate the trace in a way that passive rereading does not, so tested material is retained better at a delay even when it felt harder to study (Roediger & Karpicke, 2006).

Can something forgotten still be retained?
Yes. Material that yields nothing on free recall can still show substantial savings when relearned, meaning a latent trace survived even though it could not be recovered directly (Bahrick, 1984).

Does very well-learned material ever stop being forgotten?
Largely, yes. Bahrick found that knowledge learned to a high level, such as school Spanish, reaches a permastore that persists for decades with little further loss once the first few years have passed (Bahrick, 1984).

References