Abstract
Memory has been the subject of scientific study for nearly 150 years. Because a broad range of studies have been done, we can now assess how effective memory is for a range of materials, from simple nonsense syllables to complex materials such as novels. Moreover, we can assess memory effectiveness for a variety of durations, anywhere from a few seconds up to decades later. Our aim here is to assess a range of factors that contribute to the patterns of retention and forgetting under various circumstances. This was done by taking a meta-analytic approach that assesses performance across a broad assortment of studies. Specifically, we assessed memory across 256 papers, involving 916 data sets (e.g., experiments and conditions). The results revealed that exponential-power, logarithmic, and linear functions best captured the widest range of data compared with power and hyperbolic-power functions. Given previous research on this topic, it was surprising that the power function was not the best-fitting function most often. Contrary to what would be expected, a substantial amount of data also revealed either stable memory over time or improvement. These findings can be used to improve our ability to model and predict the amount of information retained in memory. In addition, this analysis of a large set of memory data provides a foundation for expanding behavioral and neuroimaging research to better target areas of study that can inform the effectiveness of memory.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
An important aim of cognitive psychology is to understand the progress of memories over time (Ebbinghaus, 1885). One issue that memory researchers have struggled with for well over a century is how much is remembered and forgotten as time passes. This is important because understanding the progress of memory would allow us to predict memories at different points in the future. Ebbinghaus (1885) proposed a logarithmic function to capture memory retention over time. Other researchers have suggested other functions, including power (e.g., J. R. Anderson & Schooler, 1991; Averell & Heathcote, 2011; Rubin & Wenzel, 1996a, 1996b; Wickelgren, 1974; Wixted & Carpenter, 2007; Wixted & Ebbesen, 1991a, 1991b), exponential (e.g., Loftus, 1985a, 1985b), and linear functions (Fisher & Radvansky, 2019). The aim of this study was to analyze a very large corpus of data sets from well over a century of memory research to better understand which pattern(s) best captures the progress of memory retention and forgetting over time, and to assess whether the type of pattern observed may be influenced by other factors, such as the retention delay, the type of materials tested, the memory test used, and so on.
An earlier attempt at this was reported by Rubin and Wenzel (1996a, 1996b). They assessed how well 210 published data sets fit 105 two-parameter functions. Their criteria for data inclusion were that (a) there were five or more retention intervals, (b) the dependent measure conveyed how much was remembered, and (c) each included data set fit at least one of their 105 two-parameter functions with an r2 of .90 or better. Their assessment largely involved counting the number of times each of the 105 functions was one of the ten best for a given data set. These two-parameter functions were selected because they have (a) a measure of the rate of change over time and (b) a scaling measure. Using this approach, Rubin and Wenzel concluded that the four best-fitting two-parameter functions were (a) logarithmic, (b) power, (c) exponential in the square root of time (exponential-power), and (d) hyperbolic in the square root of time (hyperbolic-power).
Rubin and Wenzel (1996a, 1996b) classified and sorted the data based on the labs that the data came from (e.g., Wickelgren and Bahrick) and study types (e.g., autobiographical memory). Our study expands their assessment by considering the amount of data in a study (e.g., number of participants and observations). Our aim was to build upon the impressive effort of Rubin and Wenzel to systematically assess and compare different retention and forgetting patterns for various independent and dependent variables, such as retention delay and memory test types, and to examine why the pattern of retention and forgetting might vary as a function of such factors.
Like Rubin and Wenzel (1996a, 1996b), we take a largely atheoretical, exploratory approach to our analyses. That said, we do consider some theoretical issues that our analysis can inform and adjudicate. This assessment of patterns of retention and forgetting is motivated by recent work in our lab. One line of work showed that there are changes in the rate of forgetting for different periods of time (Radvansky et al., 2017a, 2017b, 2022). Another reports that some patterns of retention and forgetting do not conform to a negatively accelerating pattern but are better captured by a linear function (Fisher & Radvansky, 2019). This is important because with a negatively accelerating function, there is often a constant proportional loss over some unit of time (e.g., log time for a power function). In contrast, for a linear function there is a constant amount of loss over a given unit of time. Thus, there would be an increasing proportion of information lost with longer and longer periods of time.
For this study, we added more recent memory studies and used somewhat different criteria than Rubin and Wenzel (1996a, 1996b). This larger dataset allows for a more systematic assessment of a range of variables that may contribute to the observed memory performance. In what follows, we detail the inclusion criteria as well as the independent and dependent variables used for our analysis. Our analysis is composed of two phases. We first assess the goodness-of-fit of various functions to see which function(s) emerge as best descriptors of memory, similar to the procedure used by Rubin and Wenzel (1996a, 1996b). We then assess various factors that contribute to the emergence of those functions.
Memory functions
Our assessment of memory loss was done for five target functions (Table 1): logarithmic, power, hyperbolic-power, exponential-power, and linear. These are all two-parameterFootnote 1 functions in which memory, M, across time, t, is captured by a constant scale parameter, a, and a rate of change, b. Each of these functions is now considered in turn.
Logarithmic
Logarithmic functions have been attributed to EbbinghausFootnote 2 (1885; see also Woodworth, 1938) and were among the best-fitting functions in Rubin and Wenzel’s (1996a, 1996b) assessment. They suggested that these functions may be best for non-autobiographical memories. For this function, a boundary condition is necessary. Logarithmic loss functions predict that at some point people will remember less than nothing, which is nonsensical. Thus, there should be a constraint restricting values to be positive for the appropriate application of logarithmic functions to memory data.
Power
Power functions were also among the best-fitting functions reported by Rubin and Wenzel (1996a, 1996b) and are often preferred by many researchers studying patterns of retention and forgetting (e.g., Wixted & Ebbesen, 1991a, 1991b). As noted by Rubin and Wenzel, with a power function, there are equal ratios of retention time that are accompanied by unequal ratios of memory retrieval. It has been suggested that power function occur as a result of memory consolidation processes (Wixted, 2004a, 2004b) or to mirror the occurrence of events in the environment (Anderson & Schooler, 1991).
Exponential-power
The exponential in the square root of time, which is a special case of the Weibull function, was first suggested as a description of memory change by Wickelgren (1972). This function was also among the best-fitting functions reported by Rubin and Wenzel (1996a, 1996b).
Hyperbolic-power
The fourth function that did well in Rubin and Wenzel’s (1996a, 1996b) analysis was the hyperbola in the square root of time. However, it is rarely considered as a means of capturing data from human participants outside of their study. It has, as suggested by Rubin and Wenzel, been more popular in research with animals (e.g., Harnett et al., 1984a, 1984b; Staddon, 1983a, 1983b).
Linear
The last function that we consider is the linear function. Although it did not make Rubin and Wenzel’s (1996a, 1996b) top four list, recent work suggests that it may be observed under some circumstances (Fisher & Radvansky, 2019). Thus, we include it here to better understand the circumstances under which it would appear. Like logarithmic functions, linear functions also predict that at some point, memory becomes negative. Thus, these functions also need to be used with a boundary constraint so that they cannot drop below zero. That said, it has been suggested that a linear forgetting pattern may be reflective of curvilinear patterns “by assuming that it simply reflects a scaling/measurement artifact” (Wixted, 2022, p. 1785), and as such, linear patterns of forgetting could easily be dismissed.
These five functions describe memory loss over time. We also consider cases in which memory remains stable over extended retention intervals. For our purposes, we define data sets as falling into this category when the net proportion change across a data set’s retention intervals is between −.01 and .01. Although these data sets may not contribute to understanding the function of best fit (because the data may be so flat), it is useful to understand the situations in which stable memory arises. Following this logic, we also separately consider cases in which memory increases over time. Again, examining such cases helps identify the circumstances under which improvement rather than loss might be observed with the aim of contributing to a more robust understanding of memory retention.
Previous reviews of retention and forgetting, such as Rubin and Wenzel’s (1996a, 1996b) classic assessment, have concluded that different functions capture different patterns of loss that have been observed in the literature. Different retention and forgetting functions are likely to occur because different memory representations and processes are involved, leading to the observation of different functions. These can then be used to better explore and understand how memory retention, loss, and retrieval operate. As one example, Fisher and Radvansky (2022a, 2022b) found that less well-known information was better captured by a power function, whereas better-learned information was better captured by a linear function. This was explained using a computational model of memory retrieval, the RAFT model. In brief, the explanation is that better-learned information allows for more reconstruction of partial knowledge in memory, which can produce a linear forgetting function, even if the individual elements making up a representation are lost in a way that follows a power function. This is similar to the idea that the more components of an event that can be used as retrieval cues, the better memory performance will be (Jones, 1976). In comparison, with less well-learned information, reconstruction is more difficult, and a power function emerges more readily. Thus, with knowledge such as this, knowing which function best fits the data can provide some understanding of how information is represented and processed in memory.
Inclusion criteria
There are a number of criteria for each data set to meet to be included in our corpus and analyses. These are listed in Table 2 and are described here, along with a justification for each.
Study characteristics
English
We limited our search to studies that are either published in or translated into English to ensure an accurate understanding of the methods and results of each study. The one exception to this is a study by Radosavljevich (1907a, 1907b), which is included here because of its historical significance.
Number of retention intervals
For our analyses, we look at memory retention and forgetting over time, so the studies needed to measure memory after different retention delays. We used studies with three or more retention intervals because three data points are needed to fit the two-parameter functions currently considered. Rubin and Wenzel (1996a, 1996b) limited themselves to studies with five or more retention intervals to increase data stability. While including studies with three or four retention intervals does introduce some potential instability, we include them to increase the size of our corpus. The number of retention intervals is included as a factor in the analyses to account for any potential influences of it.
Measurement
Each included data set had a clear measure of the amount of information retained in memory, such as the amount recalled, recognized, or degree of savings. For our analyses this was uniformly represented as proportion correct.Footnote 3 We further restrict ourselves to studies in which declarative memory was measured. Procedural memory contains information that cannot be consciously recalled, such as how to ride a bicycle, and is thought to involve different neural mechanisms (e.g., Cohen et al., 1985; Squire, 1986). Moreover, it is much less clear how to quantify the proportion remembered for such data. Thus, studies of procedural memory retention were excluded.
We also exclude studies in which people made free memory associations to individual words, and the data are based on what people produced initially (e.g., Crovitz & Shiffman, 1974; Rubin & Wenzel, 1996a, 1996b). With this method, there is no targeted assessment of memories at different times. These were simply the first memories that come to mind when people hear those words. This is not an assessment of the proportion of memories accessible at different time periods. Thus, we do not know how much people do and do not remember from different periods of time.
Sample characteristics
For our analyses, we include studies using human adults, with no known psychopathologies.
Humans
Unlike Rubin and Wenzel (1996a, 1996b), who included animal studies, we confine our analyses to data sets that involve humans. Memory research with non-humans has been invaluable, as there are many points of convergence between the two (e.g., Squire, 2004). However, there are also notable dissimilarities that could complicate our analyses (e.g., Premack, 2007).
Age
We limit our analyses to data that do not involve children (younger than 18). There are both neurological developments and behavioral changes that occur throughout childhood that could complicate our analyses. Thus, we chose to be conservative and not include data from children here.
Psychopathology and cognitive state
We excluded samples that assessed people with a known psychopathology, such as amnesia or schizophrenia. We also excluded samples in which the participant’s cognitive state was altered by a substance such as alcohol or caffeine. Again, memory in these groups varies from normal (e.g., Aleman et al., 1999; Wickelgren, 1975). If there was a neurotypical control group in a study, we did include that data.
A note on fit characteristics and data reduction
Our analyses examine the influence of several factors on the goodness-of-fit, as indicated by the coefficient of determination, r2, of several functions to the retention data of each study. Rubin and Wenzel (1996a, 1996b) used a strict criterion of r2 ≥ .90 for at least one of their functions to focus on well-behaved data sets. We elected not to use this criterion because, although the fit of individual studies may not be as high, the various factors included in our analysis may account for such variance.
However, because our emphasis is on assessing which function(s) best fit the data, we removed any data sets that were so variable that none of the functions captured well the nature of changes in performance. After assembling our corpus, we took some steps to reduce the amount of data used. We first dropped any data sets, in cases of memory loss, in which the fit of any of the five functions was poor. Rather than dropping anything below r2 = .90, as Rubin and Wenzel (1996a, 1996b) had done, we took a more inclusive approach. We instead took the best-fitting r2 for each study, across the five retention functions, in addition to keeping any data showing no net change or improvement. We then chose to drop those data sets for which the best-fitting function was less than r2 = .5. This is around two standard deviations from the mean of the five best-fitting functions (M = .880, SD = .192), again excluding no net change and improving data sets. This resulted in the loss of 53 data sets, which had a low average best fit (M = .31, SD = .14). The reduced data set used for our analyses can be found in our online Supplement A (all supplements are available at https://osf.io/wq9ty). For interested readers, the removed data are provided in our online Supplement B. After dropping the poorly-fitting data, the data sets showing memory loss over time were better described by the five mathematical functions (M = .922, SD = .112).
In terms of prominent differences, relative to the retained data sets, the sets that were removed had slightly smaller average sample sizes (M = 98.2 vs. M = 110.0), smaller average groups sizes (M = 24.3 vs. M = 30.3), and fewer observations per person (M = 44.3 vs. M = 120.1). They also had a higher proportion of multiple study opportunity data sets (M = .71 vs. M = .44). Moreover, the trimmed data had more retention intervals (M = 5.5 and M = 4.7) and covered longer average periods of time (M = 18.7 years and M = 8.3 years). The trimmed data also had an average lower initial memory (M = .65 and M = .73), and importantly, the trimmed data sets were more likely to involve less change from one retention interval to the next (M = −.02 vs. M = −.07). Thus, it seems likely that the data sets that were trimmed out had poor retention function fits because they had less data overall, for information that was less well-learned, over longer periods of time, and were less likely to show much change over time.
Literature search
Many studies of memory with multiple retention intervals are not labeled as such, so it is far too limiting to do a literature search merely using key terms. We have found published work examining forgetting over time (e.g., Rubin & Wenzel, 1996a, 1996b) and have sifted through the references of each to find additional papers. We also included any papers and data sets that we have happened upon during this process. We hope that the reports that are included in this analysis are representative of the population of studies that fit our inclusion criteria. From this set of criteria, we have developed our corpus. This includes data from 256 papers, involving 916 data sets (e.g., multiple experiments and/or conditions within articles).Footnote 4 That said, this corpus is almost certainly incomplete.
Data coding
There were a range of variables considered for our analyses (Table 3). These are about general characteristics of a study, characteristics of the materials, learning characteristics, nature of the retrieval tasks, and aspects of retention. The summary statistics for these variables are provided in Table 4. The data used here as well as the syntax for analysis are publicly available on the Open Science Framework (https://osf.io/wq9ty).
General characteristics
Publication year
Research practices in the study of human memory have changed over the decades. These changes may contribute to the observed data in ways that are not captured by our other factors. To allow for this, we include the year of publication as a dependent variable. The description of the various years in our corpus is provided in Table 4.
Sample size/observations per participant/amount of data
All else being equal, studies with larger sample size, more observations per participant, and more data in general, are likely to have more stable data that more accurately reflect the population (Cohen, 1992a, 1992b; Cronbach et al., 1972; Marcoulides, 1993; VanVoorhis & Morgan, 2007). Studies with fewer data points may overestimate true effect sizes and, as such, may be less replicable (e.g., Button et al., 2013). Specifically, if certain functions are better fit by small data sets, this would bring into question the accuracy of such functions. The description of the sample sizes, observations per participant, and amount of data in our data set is provided in Table 4.
Material characteristics
Material type
There are different types of memory for different types of information (e.g., nonsense syllables and novels). Thus, we also coded for the types of materials used. We identified many different material types. These are listed in Table 5.
Material complexity
The level of complexity of the material was coded to convey the degree to which the materials likely activate prior world knowledge and invite inferences. The levels of complexity are shown alongside the material types in Table 5. Level 1 (n = 124 data sets) includes materials that have very little to no semantic meaning, such as nonsense syllables, and are presented in isolation with no reference to other items. Level 2 (n = 252) includes materials that have semantic meaning, either through prior knowledge or individual experiences. However, the items are presented in isolation with little to no relation to one another. Thus, while elaboration is possible, it is likely to be limited in scope.
Level 3 (n = 37) includes materials in which there is some type of association. However, while one of the items may be meaningful, the other is not, at least from the perspective of the participant. Again, while elaboration is possible, it is likely to be limited. Level 4 (n = 145) includes materials for which there is an association between two or more meaningful items, however, not enough to form a complete proposition. Some elaboration might be possible, but it would likely be subjective and initiated by the person.
Level 5 (n = 91) includes materials that convey at least one complete idea or proposition. These materials are much more likely to encourage some elaborative processing. Level 6 (n = 24) includes materials that clearly go beyond a single propositional idea unit and involve multiple ideas. Thus, there is more opportunities for elaborative processing. However, more complex information is less likely to convey a coherent situation or event, or a collection of situations or events. The faces and events material types are placed here because they involve elements of simpler and more complex materials. Poems are placed here because they typically do not convey elaborative descriptions of events as prose does. Level 7 (n = 243) includes materials that clearly involve an understanding of situations and events, that often span across time, with many different elements and inter-relations, such as novels. Given recent findings of linear forgetting in complex materials, it is expected that material complexity may be related to linear function fit (Fisher & Radvansky, 2019).
Given that our complexity measure provides an index that can be more readily used to compare with other measures, and the different memoranda were sorted into the seven complexity classifications, we elected not to use any coding of the specific memoranda in any of our analyses. This information is available for any readers wishing to explore such issues.
Learning characteristics
Multiple study opportunities
Memory can vary depending on whether information was presented once, or if there were multiple study opportunities (e.g., Ebbinghaus, 1885). This was coded in our corpus as 0 for single (n = 471 data sets) and 1 for multiple (n = 445).
Degree of learning
Another factor that can influence the consolidation of information into memory is the degree of learning. The better the information is learned, the more likely that it has been stored in memory (e.g., Craik & Lockhart, 1972). This could have consequences for the nature of the pattern of retention and forgetting that is observed. To quantify this for the purposes of our analyses, we came up with a rough measure to code these studies by identifying four levels. Level 1 are materials from Complexity Levels 1 to 4 that likely involved rote rehearsal and were only explicitly processed once (n = 310). Level 2 are materials from Levels 1 to 4 and were explicitly processed more than once (n = 248). Level 3 are materials from Complexity Levels 5 to 7 that likely involved elaborative rehearsal and were explicitly processed once (n = 178). Finally, Level 4 are materials from Levels 5 to 7 that were explicitly processed more than once (n = 180).
Distractor task
Studies vary in whether there was an experimenter-imposed distractor task used to encourage forgetting. Memory retention requires that traces go through a process of consolidation (e.g., McGaugh, 1966, 2000). Prior to this, traces might be disrupted through distraction, leading to faster forgetting. Thus, it is possible that the pattern of forgetting would differ in studies in which a distractor task was present (n = 150) versus those in which it was not (n = 766).
Test characteristics
Assessment type
This is the type of memory test used. In our data set, these include free recall (n = 432 data sets), cued recall (n = 121), yes–no recognition (n = 141), forced choice recognition (n = 160), savings (n = 14), stem completion (n = 22), fragment completion (n = 12), anagram solution (n = 1), matching (n = 7), problem solutions (n = 4), and source monitoring (n = 2).
Number of retention intervals
We coded for the number of retention intervals in each data set. Data with more retention intervals may have more stable patterns than data with fewer, and this needs to be considered. This variable allows us to assess whether certain retention functions are more likely with different numbers of intervals, perhaps because of the stability (or lack thereof) of the data. The descriptive information about the number of retention intervals is shown in Table 4.
Study design
Studies of retention and forgetting use either a between-subjects design (with a different group of participants at each retention interval), or a within-subjects design (measuring memory in all people at all retention intervals). This is important to account for because practice effects can occur with within-subjects’ designs, although it does reduce some error variance (Greenwald, 1976). In our corpus, within participant designs were coded as 0 and between participant designs as 1. There were more within-participant designs (n = 489) than between (n = 294).
Number of observations
We also include the number of observations per participant. This includes the number of trials. The more observations there were per person, the more stable the data are likely to be. Thus, more observations are more likely to capture retention, and provide more replicable and stable results. The number of observations per person can compensate for low levels on other factors, such as the number of participants (Smith & Little, 2018).
Amount of data
We combined information about sample size and the number of observations per person to calculate the overall amount of data in each study. This is the number of participants in the study times the number of observations per person. Descriptive information about both the number of observations per person and the amount of data are provided in Table 4. However, for all further analyses, number of participants and number of observations per participant are not considered, as the amount of data measure was derived from these.
Retention characteristics
Retention intervals
Memory exhibits different properties at different periods of time. Traditionally, there is a distinction between short-term/working memory and long-term memory (e.g., Atkinson & Shiffrin, 1968; Cowan, 2008). There have also been suggestions of other divisions, such as between long-term memory and long-lasting memory (McGaugh, 2000). These phases of memory may have different neurological and behavioral properties. For example, there is a shift in neurological processes over the span of retention, where the hippocampus is more active in early memory consolidation but becomes less active as memories are transferred to the neocortex (e.g., Squire & Alvarez, 1995). The neurological shift is reflected in changing speeds of forgetting (e.g., Radvansky et al., 2022), where forgetting speed decreases up to about a day, increases between 1 and 9 days, and remains somewhat stable thereafter.
Because there are shifts in the neural mechanisms supporting memory retention, it is possible that forgetting can be best described by different functions over the course of retention. As a hypothetical example, it could be that retention is better captured by a logarithmic function prior to one day, but a power function at longer delays. Thus, we include delay to allow us to discover any such differences. The issue of retention delay, while seemingly straight-forward, becomes somewhat thorny when considering the fit of data to retention and forgetting functions. There are many ways to quantify delay in studies with multiple delay intervals, so we capture delay in four ways: (1) the shortest retention interval, (2) the longest retention interval, (3) the average retention interval, and (4) the retention range (longest–shortest). For consistency, all studies are coded into the nearest reasonable number of seconds. Descriptive data on each of these is provided in Table 4.
Initial memory
Somewhat related to degree of learning is initial memory. Initial memory could affect the fit and rate of forgetting (Slamecka & McElree, 1983a, 1883b; Wixted, 2022). For example, Anderson and Schooler (1991) suggested that a logarithmic function has a steeper slope when there is a higher initial memory, but the power function does not depend on the initial memory. Thus, the best-fitting function may be related to initial memory levels. The descriptive data for initial memory levels are provided in Table 4.
Note that for some studies, the first memory assessment is reported as being made “immediately” or after 0 s. Rather than setting an initial delay time as 0 in these cases, which would literally mean that the information and the memory test were perfectly coexistent in time, and because some functions require a non-zero value to calculate, we took two approaches. The first was the acknowledgement that an immediate memory test was not actually instantaneous. Instead, there was some nominal delay, even if it is something like the refresh rate of a computer screen. To provide an estimate of what the actual delay time was we either used an estimate based on information about actual display durations or provided an educated estimate of how long instructions for a memory test would take (e.g., 30 s). We note in our corpus how these initial estimates were made in applicable cases. In some cases, a 0-second delay is more in line with what actually happened, such as in cases of short-term memory testing. This is a problem for some memory functions, such as the power function, because it is mathematically undefined at 0 seconds. To address this, we set the number of seconds for this initial memory at a small value (e.g., .01 s) that is not likely to be psychologically meaningful in the context of these data sets.
Confirmatory and exploratory analyses
While our approach is largely atheoretical, there are some theoretical implications for what we find. Our analyses are part confirmatory and part exploratory. The confirmatory analyses address issues or findings that are already reported in the literature. In comparison, the exploratory analyses are those for which there is no strong, a priori, theoretical expectation or prediction about the outcome. However, these are of interest because they have the potential to provide insight to guide future research.
Confirmatory analyses
Best-fitting functions
Of most interest to us here, we consider predictions for which of the retention and forgetting functions will best capture the data. Previously, Rubin and Wenzel (1996a, 1996b), in their extensive analysis, reported that logarithmic functions fit the data best most often, followed by power, exponential-power, and hyperbolic power, and with linear patterns doing the worst by far. This would be in line with some of the earliest work on memory retention and forgetting (i.e., Ebbinghaus, 1885). Moreover, they also suggest that this will be more likely to be the case for simpler materials, compared with complex memories, such as autobiographical memories.
A competing prediction, based on a report by Wixted and Ebbesen (1991a, 1991b), is that the data will be best fit by a power function. This is almost a default assumption of researchers studying retention and forgetting (e.g., Anderson & Schooler, 1991; Averell & Heathcote, 2011; Carpenter et al., 2008a, 2008b; Wickelgren, 1974; Wixted & Carpenter, 2007). An attraction of the power function is the idea that forgetting functions “almost invariably exhibit a decreasing relative rate of forgetting, as noted long ago by Jost (1897)” (Wixted, 2022, p. 1779). This invariability is assessed here.
Material characteristics
Rubin and Wenzel’s (1996a, 1996b) report predicts that autobiographical memories will be better fit by power functions than other types of materials. We can tentatively expand this to memory for any type of complex set of materials (e.g., stories). Rubin and Wenzel arrived at this conclusion largely based on a consideration of autobiographical memory studies using the Galton–Crovitz technique of eliciting memory reports (Crovitz & Quina-Holland, 1976). We exclude these here because this approach does not assess how much is remembered from a given time period, but only provide information about the memories initially retrieved in response to a cue.
A competing prediction is that autobiographical memories will be more likely to be well-fit by linear functions (Linton, 1982a, 1982b). This is based on the pattern of data observed for long-term memory for an extensive autobiographical memory study. A more general prediction is that as material complexity increases, the pattern of forgetting will be more linear (Fisher & Radvansky, 2019). This is based on research that has found clear and stable evidence of linear forgetting. Fisher and Radvansky’s (2019) report highlighted the fact that when a linear pattern of retention and forgetting is observed, the studies involved used more complex materials (e.g., narratives).
Learning characteristics
Another observed methodological factor involved in the pattern of retention and forgetting is the degree of learning. One prediction is that when there is a higher degree of learning, the pattern of forgetting will most likely be linear (Fisher & Radvansky, 2019). A review of the literature showed that when clear linear patterns of forgetting are observed, it is not unusual for such studies to involve a higher degree of learning, as with overlearning (e.g., Burtt & Dobell, 1925a, 1925b).
Having said all of this, we explicitly note here that we are aware that there may be other reports or accounts in the literature that provide a basis for confirmatory predictions that we have missed.
Exploratory analyses
Year of publication
There are no strong a priori expectations that the year a study was published will influence the pattern of data. However, we do include this as a factor in our analysis to address the possibility that something, perhaps methodologically, has changed over the years to yield different patterns of memory retention and forgetting.
Sample size
There is no question that sample sizes can influence the patterns of data observed in psychological studies. Studies with small sample sizes may provide distorted views of the mind. Thus, it is reasonable to expect that differences in sample size may lead to different patterns of results, some of which may be more distorted than others. For example, if some memory functions are only seen with smaller sample sizes, then this would be an indication that that function is a result of more random fluctuations in the observed pattern of data, and not reflective of underlying memory mechanisms. Our analyses allow for this assessment.
Memory test characteristics
While there has been some suggestion that memory test types may influence the pattern of observed retention and forgetting (e.g., Haist et al., 1992), there may be aspects of the memory testing process itself that influence this pattern of which we are not aware. These aspects may include study design (such as whether it is a within- or between-participants design) and the number of observations per person.
Retention characteristics
The influence of various aspects of retention, such as the number of retention intervals, the shortest and longest intervals, the average interval, and the range of the retention interval, are largely unknown. Our analysis will provide some insight into this.
Analyses
We have two basic analyses. The first is in line with the approach taken by Rubin and Wenzel (1996a, 1996b). Specifically, we compare the fits of the top four functions from Rubin and Wenzel’s work (logarithmic, power, hyperbolic-power, and exponential-power), along with the linear function in cases in which there is no net change over time or which memory is increasing, to assess how often each of these fit the data better than the others. Our second major set of analyses was to assess which of a wide range of factors are more likely to produce one pattern of data over another. That is, what aspects of the sample, the materials, the memory assessment, and so on, lead to different patterns of retention and forgetting.
Best-fitting functions analysis
We first consider the assessment of how often each of our functions was the best fit for each of the data sets. As a reminder, Rubin and Wenzel (1996a, 1996b) fit all the data sets in their corpus to 105 functions as part of their primary analysis. They then tallied how often a given function was among the best 10 fitting functions for a given data set. From this, they concluded that the best-fitting four functions were logarithmic, power, exponential-power, and hyperbolic-power. We take a similar approach here, with some changes. Again, we deviate from Rubin and Wenzel in that we are not comparing 105 functions, as they did (they rejected many of them). We also deviated from Rubin and Wenzel in that while they only considered data when r2 ≥ .90 for at least one function, we allow for poorer fits. Second, while they treated each data set equivalently, we adjusted for the number of observations in a data set by using weighted means. It is important to account for the size of the study, in terms of the number of participants and the number of observations. Otherwise, studies with few participants/observations can place undue weight on the results and skew our conclusions. For those readers interested in the patterns of results when data sets were limited to r2 ≥ .90 and/or the data were not adjusted for the number of observations, these are available in our online Supplement C.
The results of this analysis are in Fig. 1. As can be seen, the logarithmic function was the most successful because it best fit the largest proportion of data sets, followed closely by the linear and exponential-power functions. The hyperbolic-power function did worse, and the power function again did the worst. Finally, the relative proportion of studies that conformed to either no net change or increasing was substantial.
Overall, this analysis failed to support two confirmatory predictions. The first, based on work by Rubin and Wenzel (1996a, 1996b), was that logarithmic functions would do best, followed by power, exponential-power, and hyperbolic power, and with linear patterns doing the worst. Instead, we observed very different patterns of effectiveness of the various functions at capturing memory retention and forgetting.
The second unsupported prediction was that, based on a report by Wixted and Ebbesen (1991a, 1991b), as well as others, the dominant equation would be a power function. They might have observed those results for no other reason than because the averaging of various other underlying functions can be well fit by a power function (e.g., Anderson, 2001), particularly if there is greater variability among the individual forgetting functions due to variability in participants and memory for different material items. Based on this alone, one might expect power functions to do very well, even if the underlying pattern of memory loss is not a power function. However, with our corpus, the power function was the best to the smallest degree. Thus, prior implications of the dominance of the power function as the best description of memory retention and forgetting was not supported. At a minimum, given that power functions can be observed from an averaging of other functions with higher levels of variability in their loss rates, within studies, there is relatively little variability across participants and items.Footnote 5
One other point to note is that goodness of fit values for the loss functions are correlated with one another, suggesting that some of them produce patterns that are difficult to distinguish with some memory data. The correlation matrix is shown in Table 6. As can be seen, the power and logarithmic function fits were highly correlated, as were the exponential-power and hyperbolic-power functions. The linear function differed the most from the other functions. Thus, there seem to be three families of functions here: (1) logarithmic/power, (2) exponential-power/hyperbolic-power, and (3) linear. One could drop consideration of the power and hyperbolic-power functions and still provide a fairly accurate characterization of most forgetting curves.
After assessing which functions fit the data most often, we also assessed how well the various functions did when they were the best fit for a data set. First, the exponential-power (M = .968, SE = .004; min = .764, max = 1.00) and hyperbolic-power functions (M = .966, SE = .006; min = .522, max = 1.00) fits did the best. This was followed by the power (M = .934, SE = .008; min = .651, max = 1.00), linear (M = .888, SE = .010; min = .510, max = 1.00), and logarithmic functions (M = .883, SE = .010; min = .514, max = 1.00).
Typical functions
To get a feel for the different categories of memory retention and forgetting for our five functions, as well as no net change and increasing data sets, we plotted each of these using the median value for the a and b formula values. For the no net change data sets, we used the median initial performance level, and then extended that throughout. For the increasing data sets, we used the median a and b values for the best-fitting linear functions. The plots for these typical patterns are shown in Fig. 2. As can be seen, the logarithmic and power functions both show more dramatic loss over log time, especially for earlier retention intervals, with the power function showing more rapid loss. Also, the exponential-power and hyperbolic-power functions closely resembled each other, and the linear pattern was closer to these as well. Finally, the increasing data sets started out at a lower initial level of performance, overall, and there was very little change over time for the typical data set. Thus, many of these data sets could be grouped with the no net change data sets, without much loss in the accuracy of the characterization of the data.
Memory factors analysis
Our next aim was to determine which factors contribute to different patterns of retention and forgetting. The first step was to assess whether any variables were strongly correlated with one another, and drop those that were to reduce redundancy. The next step was to assess the characteristics of the data sets for our seven patterns of performance over time. This may provide some insight into which study characteristics, when present, are likely to lead to a particular retention and forgetting pattern.
Correlation analysis
Our first step was to assess whether any of our 13 numeric variables are strongly correlated with one another. The results of a correlation matrix are shown in Table 7. First, the correlation between Longest Retention Interval and Range was nearly perfect (r = .99). We elected to drop Range because, conceptually, we are interested in how long people remember things.
Moreover, we elected to identify variables that are correlated .70 or greater as collinear. To reduce collinearity, we also reduced the number of variables that met this standard criterion. Not surprisingly, the remaining three variables related to the length of the retention interval (Shortest, Longest, and Average Retention Interval), were highly inter-correlated (r = .72 to .98) and collinear. We elected to keep Longest Retention Interval. Another indicator of collinearity was between Complexity and Degree of Learning (r = .77). To address this, we elected to drop the Degree of Learning because it was also highly correlated with Multiple Study Opportunities (r = .52).
Finally, there was still one large correlation, between Amount of Data with Number of Retention intervals (r = .55). This is sensible. In general, the more retention intervals that were tested, the more data there was. This is largely unavoidable, and we retain these variables for our analyses but with an awareness of this relationship. Overall, we reduced the number of independent variables from 13 to 9.
Curve category characteristics
Our next step was to assess how our study characteristics differed across our seven categories. We did this in two ways. We first tested for any differences for each of the individual factors using ANOVA and Tukey tests. We then did logistic regressions in which we assessed the ability of our factors to predict when a given function would be the best at capturing the data sets.Footnote 6 The factor means and standard errors for each of the memory patterns is shown in Table 8. Here, we do not report data from studies using anagram solution, matching, problem solving, and source monitoring measures because there were so few of them. Moreover, we collapsed stem and fragment completion studies because they were methodologically and conceptually so similar.
For the data in Table 8, we performed an ANOVA comparing performance on each of the factors (Table 9). If the ANOVA was significant (at least marginally), we also report any significant pairwise Tukey comparisons. One concern may be that five of these patterns are loss functions, and the other two (No net change and Increasing) are not. It might be that there are qualitative differences that result in memory loss versus not. Thus, we did each of our analyses twice, first including the No net Loss and Increasing pattern data sets, and then without. We report the first here, and the results are shown in Table 9. The second are available in our on-line Supplement D for interested readers. There were no major differences between the two analysis approaches.
The results of the regression analyses are shown in Table 10.
Logarithmic
According to the logistic regression, the logarithmic function was more likely to be the best fit for studies using a between-participants design. This was also only mildly observed with the individual characteristics analysis, in which the only significant difference was with hyperbolic-power functions, and that was marginal. Still, the fact that logarithmic functions were quite frequent, and many studies used between-participant designs, leads to the conclusion that this may be driving this high frequency.
The regression also revealed that logarithmic functions were likely to have more retention intervals. However, in the individual factor analyses, the only difference is with hyperbolic-power functions. Thus, it may be the case that logarithmic functions are more likely to emerge when researchers test a broader range of retention intervals, however the evidence is not strong.
Finally, it seems as though logarithmic functions are more likely when savings is used, but this is not seen with the individual characteristics analyses. Conversely, there is some evidence that logarithmic functions are more likely with word stem/fragment completion tasks, but this is only seen in the individual comparison analyses, and it is weak there.
Power
When power functions best fit a data set, the factor that was most strongly associated was Multiple Study, when people had been exposed to the materials once rather than multiple times, as seen in both the individual characteristic and regression analyses. Thus, power functions may be capturing weaker memories that a person has only experienced once.
This is supported by the finding that the data sets best fit by a power function also involved less-complex materials, which was also supported to some degree in the individual characteristics analysis. Thus, power functions are more likely to be seen with simplistic materials.
Moreover, like the logarithmic function, there was some weak evidence that power functions were likely to be the best-fitting functions when savings was the measure of memory, such as Ebbinghaus’s work. That said, this should be taken with a grain of salt given that there are so few studies in our corpus that used savings. Conversely, like logarithmic functions, there is some evidence that power functions are more likely with word stem/fragment completion tasks, but this is only seen in the individual comparison analyses, and it is weak there. Finally, there was some evidence that power functions were more likely to be the best when the data come from older studies, but, again, evidence for this is weak.
Exponential-power
A higher level of initial memory was by far the clearest factor to lead to exponential-power functions succeeding as the best. This was true both in the regression and individual characteristic comparisons. Thus, this sort of function is more likely to emerge with well-learned materials.
Furthermore, exponential-power functions were more likely to be the best when there was a distractor task just after learning. Again, there was support for this in both regression and individual characteristics comparisons, although weaker in the latter. A distractor task immediately after presentation of high initial memory likely disrupts consolidation processes.
In addition, exponential-power functions were the best-fitting ones when the retention intervals were relatively short, according to the regression. However, in the individual comparisons, exponential-power functions do not strikingly differ from the rest. This may be because there is so much overlap with other, especially curvilinear, functions.
Finally, there was some evidence that an exponential-power function was likely to be the best when there was more data in a data set, but this evidence was weak, and not evident in the individual characteristics analyses.
Hyperbolic-power
The factor that was most clearly associated with hyperbolic-power functions best-fitting a data set is Longest RI, when the retention intervals were relatively short, at least for the regression. There was a small amount of support for this in the individual characteristics analyses. Thus, these sorts of functions are more likely to be the best ones over shorter periods of time.
In addition, these functions were more likely to be the best when there were fewer retention intervals. This was evident in both the logistic regression and individual comparison results. It may be the case that this type of function, which is rarely discussed, may be more likely to be the best one when there are fewer data points to fit the curve, which lowers our confidence in it.
There was also some suggestion that this pattern was more likely to be the best when a repeated-measures design was used with the same people being tested at multiple time points. However, this was only significant in the regression. Finally, there was some suggestion that hyperbolic-power functions were more likely to be best fitting when recognition was used, but this finding was weak.
Linear
The factor that was most strongly associated with linear functions being the best fit for a data set was Longest RI, when the retention intervals were relatively long. This was the case both for regression and individual characteristic comparisons. Thus, those memories that are particularly long-lasting, such as autobiographical and event model memories, have a different memory retention and loss profile than other types of material. This finding also works against the idea that linear forgetting is a scaling artifact in which curvilinear forgetting would be observed if data across a longer time scale were collected (Wixted, 2022). If this were the case, then one would expect linear forgetting to be observed with shorter retention intervals, not longer.
This is supported to some degree by the finding that in the individual characteristic analyses, data sets that were best fit by linear functions tended to also involve more complex materials. That is, there was significant difference when compared against all the other loss functions. That said, this was not significant in the regression.
Another outcome was the effect of year in the regression, with the individual characteristic comparisons highlighting the difference between when linear and power functions were the best fits. This suggests that there is some aspect of those studies that is not well captured by other factors used here and which would need to be resolved by further research.
Also, there was some evidence that a linear function was likely to be the best-fitting one when there was greater initial memory, but only marginally so, but not for the individual characteristic comparisons. Thus, the evidence here is weak. Finally, there was a significant difference in the individual characteristic comparisons suggesting that linear forgetting patterns were more likely to be the best when multiple choice measures were used, although this was not significant in the regression, the reasons for this pattern not clear.
No net change
The factor that was most strongly associated with no net change across retention intervals, like linear patterns, was Longest RI, with relatively long retention intervals. This was the case in both the regression and individual comparison analyses. Thus, those memories that are longer-enduring are also less likely to show evidence of forgetting. This is akin to the idea of a memory permastore for long-lasting memories (e.g., Bahrick, 1984a, 1984b). Moreover, no net change was more likely to be observed in studies with more retention intervals, consistent with the idea that memories that are likely to be observed across multiple time periods are also likely to be more durable.
There was also some suggestion in the regression analysis that no net change was more likely to be the best solution when initial memory levels were relatively low, perhaps because most of forgetting processes had already occurred. This, however, was not supported by the individual characteristic comparisons. Finally, although not significant in the regression, the individual characteristic comparisons suggested that these patterns of data were more likely to involve more complex memories.
Increasing
The factor that was most strongly associated with increasing memory over time, like linear forgetting and no net change, was Longest RI, when the retention intervals were relatively long. This was found in both the regression and individual characteristic analyses. This is again consistent with the idea that this is more likely with materials that are durable in memory.
Unlike linear forgetting and no net change, data that were best described as increasing were more likely to have initial memory levels that were lower and included fewer retention intervals. This was found in both the regression and individual characteristic analyses. This makes sense in that these are likely to involve cases where there is more room for improvement. On top of this, this was more likely to happen, based on the regression, when there were fewer retention intervals and with repeated-measures designs, suggesting that some element of hypermnesia may be involved (e.g., Payne, 1987). That said, there was less support for this in the individual characteristics analyses.
Finally, there were several individual comparisons that were not significant in the regression. Specifically, increasing data patterns were more likely to involve more complex materials, multiple study exposures, the presence of a distractor task (from which recovery after initial encounter might be possible), and with recall tests (which is also where hypermnesia effects are more likely to occur).
Overview summary
At this point, let’s take a step back and consider why different factors may lead to different retention functions. All of these ideas are speculative, and would require explicit experimental investigation to support or refute them. Logarithmic and power functions are highly correlated. They are more likely to involve older studies with simpler materials and a single exposure. These sorts of forgetting functions may emerge because less complex memory traces can degrade more quickly, leading to clearly curvilinear forgetting functions that would be better captured by logarithmic and, especially, power functions. They were also more likely to be observed when the memory tasks involved savings or completion tasks, both of which place a heavier emphasis on implicit memory.
Much of this is reinforced by the regressions, especially for power functions, which were the best fits when there were single study opportunities, less complex materials, shorter retention intervals, and savings was used as the memory measure. In comparison, logarithmic functions were the best fit when between participants designs were used (which reduces the impact of learning the task, and the benefit of prior retrievals), and more retention intervals (perhaps because it would be more likely that an asymptote would be approached).
Exponential-power and hyperbolic-power functions were also similar, with exponential-power functions having clearer characteristics. Specifically, they were more likely to involve distractor tasks and higher levels of initial memory. The distractor tasks may disrupt consolidation processes, leading to more rapid forgetting earlier on. With the higher levels of initial memory, this may provide some resistance to forgetting processes earlier on, but which exert themselves after a period, leading to curvilinear patterns more like to those observed with data best fit by logarithmic and power functions.
The logistic analysis supported the idea that shorter retention intervals are more important for these functions. On top of this, exponential-power functions were also more likely to be the best fits when a distractor task was involved and higher initial levels of memory were recorded. They were also more likely with more data in a study and multiple study opportunities. Hyperbolic-power functions were more likely to be best-fitting functions when repeated-measures designs were used (allowing some influence of prior retrieval attempts), along with fewer retention intervals in the data set.
Linear functions were characterized by many of the same factors that also characterized stable and increasing data sets. These factors closely match those suggested in other work (Fisher & Radvansky, 2019). Specifically, it has been suggested that linear forgetting is more likely to be observed with complex materials that have been learned well. The additional finding that data sets with longer retention intervals fit along nicely with this given that more complex materials, such as memories for events, are likely to be more enduring. The additional characteristic that increasing data sets have lower initial memory is consistent with the idea that in order for memory to improve over time, the worse it is to begin with and the more room there is for improvement. Stable data sets might have more retention intervals on average because they are extended into longer periods of time, when memory traces are more likely to have been well-established.
In terms of the regressions, while complexity was not a significant predictor for any of these three patterns, all of them had longer retention intervals as a significant predictor. However, while linear forgetting was more likely to be observed with higher initial memory, stable and increasing data sets were more likely to have lower initial memory. Moreover, while stable memory data sets were more likely with more retention intervals, increasing memory data sets were more likely to involve fewer. Also, while stable data sets were more likely with single study opportunity data sets, increasing data sets were more likely to involve repeated-measures designs, suggesting that some element of hypermnesia may be involved. Finally, the finding that linear best-fitting functions were more likely to be observed for more recent studies suggest that there are aspects of these studies that are not well-captured by the factors that we identified here.
As a further step, we elected to create a guide for predicting the function of best fit for future work. As previously mentioned, one of our goals is to improve our ability to model and predict the amount of information retained in memory. This model allows researchers to input the characteristics of a study and obtain the expected function that would best predict memory over time for the study’s specific set of characteristics. Importantly, the results of the model should be approached with caution, as it is trained using datasets that are potentially underpowered, noisy, or biased. The model was created using the ChefBoost: C4.5 machine learning model, which works as a statistical classifier (Quinlan, 2014; Serengil & Teknoloji, 2021). The resulting model is available as Supplement E.
Overall, it is clear from our analyses that different functions describe changes in memory over time (or not, if memory is stable). These different functions tend to be associated with different material and task properties. Thus, as we continue to develop our understanding of when and why different functions are likely to be observed over time, we can better understand the mental representations and processes involved.
General discussion
One of the most fundamental features we know about memory is that it changes over time (Ebbinghaus, 1885). If we hope to be able to predict future memory, we need to be able to provide an adequate description of the nature of that change over different types of information and over different time scales. Our aim was to identify the shape of the patterns of memory change over time, and to begin to identify characteristics that bring about different patterns of memory change.
There were a number of major findings to come out of our analyses. The first is, consistent with Rubin and Wenzel (1996a, 1996b), that there is no single function that captures the progress of memory retention and forgetting over time. Second, the functions varied noticeably in terms of the proportion of data accounted for best by a function, with logarithmic, exponential-power, and linear functions accounting for the bulk of the cases in which there was forgetting. This is consistent with the idea that different types of memory representations and processes can lead to different changes in memory over time (or even no change at all). Thus, such patterns can be used to further assess the nature of memory. Third, there was a strong similarity between logarithmic and power functions, as well as between exponential-power and hyperbolic functions. Linear functions differed from these. Next, let us consider some of the confirmatory and exploratory analyses discussed in the introduction.
The analysis of our corpus for retention and forgetting provides a means of addressing a wide range of confirmatory and exploratory analyses. If we know how various factors contribute to the goodness-of-fit for the various functions, it helps us predict memory from a subset of retention intervals. For example, if a linear function is best for factors X, Y, and Z, and a study has those factors, a linear function should be used to fit performance and predict future memory.
Confirmatory analyses
Best-fitting functions
The first issue our analyses revealed is the relative prevalence of different function types in our corpus. Our assessment of the common function types is at odds with what has been reported in the literature. In the earlier meta-analysis by Rubin and Wenzel (1996a, 1996b), it was suggested that logarithmic functions would fit the data the best, followed by power, exponential-power, and hyperbolic-power, and with linear patterns doing the worst. However, we found that a different order emerged. While we are in line with Rubin and Wenzel in finding that logarithmic functions did the best, with exponential-power functions also doing well, we found that linear functions also accounted for a substantially larger proportion of the data. Moreover, power and hyperbolic-power functions, while accounting for some of the data sets, did more poorly relative to the first three.
Part of the reason for this may lie in how Rubin and Wenzel (1996a, 1996b) analyzed the data compared with how we did. Specifically, Rubin and Wenzel gave credit to a function if it was one of the top 10 (out of 105) best-fitting for a data set. In comparison, we are strictly looking at which was the best. Because logarithmic and power function fits are so highly correlated, when one of these does a good job, the other is likely to do so as well. Thus, they are both likely to make the top 10 with Rubin and Wenzel’s approach, but only one will be the actual best, as in our approach. The same can be said for exponential-power and hyperbolic-power functions.
Furthermore, Rubin and Wenzel (1996a, 1996b) did not consider whether data sets had data that either had no net change or were increasing, whereas we did. Such data sets may have been more likely to be excluded from their analysis because of their criterion of r2 ≥ .90 for at least one function. Such data sets, because they deviate from a traditional loss pattern, are less likely to fit one of these functions well. Thus, there is a substantial subset of data that are producing patterns over time that are not being well accounted for.
We now turn to other issues raised in the introduction. It has also been suggested that power functions best capture the pattern of forgetting (e.g., Wixted & Ebbesen, 1991a, 1991b). However, while power functions did the best for some data sets and were highly correlated with logarithmic functions, which did the best overall, they accounted for a smaller proportion of data sets compared with our other functions. This is particularly surprising given that there have been several studies showing that power functions should be quite common because averaging across other types of functions (as might occur when averaging across multiple memory traces) generally produces a power function as an artifact (e.g., R. B. Anderson, 2001). Moreover, the acceptance of a power function in the literature may be a consequence of unfortunate sampling error of a small set of studies from those assessments that have looked at multiple data sets (e.g., Wixted & Ebbesen, 1991a, 1991b).
Overall, our analyses suggest that there are clear regularities in how memory operates that in turn produces other types of common functions such as logarithmic, exponential-power, and linear functions. Logarithmic functions are intuitively sensible. Specifically, a simple logarithmic function conveys a constant proportion loss of information across a given unit of time. Similarly, a simple linear function conveys a constant amount of loss of information across a given unit of time.
The exponential-power function is less intuitive, and deserves further consideration. At the outset, it should be noted that the square root of time that is used in this function reflects a setting of the exponent of the power function of this formula to .5 (after Wickelgren, 1974). The more expanded version has an additional parameter: \(M=a{e}^{-b*{t}^{-c}}\). Thus, this three-parameter function puts it outside of the set of two-parameter functions that were explored here and by Rubin and Wenzel (1996a, 1996b). That said, further research may show that such an approach better captures memory over time, despite the increase in the number of parameters. For the time being, like Rubin and Wenzel, we considered the most preferred two-factor functions.
Material characteristics
Our analysis helps clarify why, in some cases, forgetting seems to follow one pattern over another. That is there are some factors that influence the observed patterns of change. Rubin and Wenzel (1996a, 1996b) suggested that autobiographical memories will be more likely to be well fit by power functions, and we expanded this hypothesis to memory for complex materials more generally. Our analyses failed to support this prediction. This discrepancy may result from the fact that Rubin and Wenzel considered studies that largely used the Galton–Crovitz technique in which people respond with the first autobiographical memory they retrieve in response to a cue word (e.g., cake). These first retrieved memories may exhibit different properties than assessments of accuracy over various retention intervals.
In contrast to this, our analyses found that complex materials (Fisher & Radvansky, 2019), including autobiographical memory studies (e.g., Linton, 1982a, 1982b) are more likely to be better fit by a linear function. We were able to provide some confirmatory support for this finding.
Learning characteristics
Our analyses also partially supported the idea that when there are higher degrees of learning there are more linear patterns of forgetting (Fisher & Radvansky, 2019). This finding was present, and provided some confirmatory support, albeit this was weak.
Exploratory analyses
Year of publication
Although we did not expect to find any influence of the year of publication on the patterns of forgetting, there were some significant effects. Specifically, it was found that older publications were more likely to conform to power functions than more recent work, which were more likely to conform to linear functions. This may be because older studies tended to emphasize simpler materials and newer studies more complex materials. This may contribute to differences in findings between our study and the one by Rubin and Wenzel (1996a, 1996b).
Amount of data
We saw very small influences of the amount of data on the observed pattern of retention and forgetting. Thus, there does not appear to be any strong evidence for the idea that some patterns of data may be observed because of fewer data points. This may be the case with hyperbolic-power functions. There was also some evidence that studies with more data were more likely to either be best fit by an exponential-power function or have no net change. However, this evidence was weak, and we would caution against using this as a basis of prediction.
Memory test characteristics
In our data set, there was some suggestion of an association of free recall tests with increasing functions and recognition with hyperbolic-power functions, although this evidence is very weak. There was also some suggestion that studies that used multiple choice tests were more likely to be associated with linear functions, and a strong suggestion that studies that used savings were more likely to be associated with power functions. Thus, there is some evidence to suggest that different memory tests may lead to different patterns of results. However, this was really only observed in the individual characteristic comparisons, and not when other factors were considered, as with the regression analyses.
Retention characteristics
Finally, our exploratory analysis of retention characteristics assessed the number and length of the retention intervals. In terms of the number of retention intervals, more retention intervals were associated with logarithmic and no net change patterns. In the first case, this may be because more retention intervals make it more likely that an asymptote would have been approached, making it difficult to detect any kind of changes over those time intervals. In comparison, data sets with fewer intervals were associated with hyperbolic-power and increasing power functions. The first, our most unusual function, may be reflective of studies with fewer time points, allowing for an unusual function to fit best by chance. The second is likely to emerge under more constrained circumstances, making it more likely for random variation to emerge as an increase in memory over time.
In terms of the longest retention interval, this is probably the one factor that was the clearest indicator of which pattern would be observed. Specifically, shorter retention intervals were better fit by curvilinear functions, such as logarithmic, power, exponential-power, and hyperbolic-power functions, whereas longer retention intervals were more likely to be better fit by linear functions, as well as being more likely to involve either no net change or be an increasing data set. Why might this be the case? It may be that, consistent with work by Ebbinghaus (1885), the largest changes in memory are more likely to occur early on after learning. Thus, when data sets are more likely to involve short retention intervals, these large changes are more likely to be observed. However, when studies focus on longer delays, this would not be revealed, with the data falling in a more consistent way, or even perhaps having reached an asymptote. If the information from these studies were sampled at shorter time intervals, more curvilinear patterns would be observed.
However, there are some important things to note. The first is that if a function holds true across time, it should not matter where along the curve it is sampled. So, a change in best-fitting pattern for different time periods is important. It suggests that memory processes are changing over time. Second, it is possible to observe patterns of data shifting from curvilinear to linear within the same time frame. For example, in a study by Fisher and Radvansky (2022a, 2022b), people memorized lists of sentences that varied in terms of the degree of learning. Although the same memory test delays were used in all conditions, the less the degree of learning, the better the data were fit by a power function, and the greater the degree of learning, the better the data were fit by a linear function. The same can be said for studies of the testing effect (e.g., Roediger & Karpicke, 2006a, 2006b). Thus, while length of retention interval was an important factor in our analyses of our corpus, it is not definitive.
Unresolved issues
Our analysis here covered a wide range of issues. However, there are still many left unresolved. One of these is that recent work has suggested that memory retention and forgetting is not continuous, but goes through different phases (Radvansky et al., 2022). It might be the case that different functions are more appropriate for different time intervals after learning. Another is that different types of materials and methods may lead to differences in the rate with which information is forgotten (or not). This is a long-standing issue (Loftus, 1985a, 1985b; Rivera-Lares et al., 2022; Slamecka, 1985; Slamecka & McElree, 1983a, 1983b; Wixted, 2022) that data sets such as ours may be able to provide important insight into. We are actively pursuing both issues with our data set. It is also almost certainly the case that there may be aspects of memoranda and methods that influence patterns of retention and forgetting that have not been considered here.
Conclusions
This work provided an opportunity to expand on the work originally reported by Rubin and Wenzel (1996a, 1996b). Our attempt uses a wider range of studies that is, in some ways, more inclusive, allowing for a broader assessment of memory retention and forgetting. We explored the degree to which different functions fit the available data, and the factors that contribute to those functions. This approach allows use to perform a wide range of confirmatory and exploratory analyses. This work has the potential to have a broad-ranging impact on psychological science.
Perhaps the most important issue addressed here is changes in memory over time. As is emphatically clear, like Rubin and Wenzel (1996a, 1996b), we did not find any evidence for a single function that does a good job with all of the data. Instead, we found the data when there was forgetting were best captured by (a) logarithmic/power functions, (b) exponential-power/hyperbolic-power functions, and (c) linear functions. Moreover, a sizable minority of studies in our corpus showed either no evidence of a net change across all of the retention intervals, or actually showed more than trivial improvement in performance over time. These different patterns of performance are almost certainly due to different underlying memory processes. Just what these are is a task left to future research.
Having identified which pattern best captured each of our data sets, we then assessed what study characteristics were more or less likely to lead to different patterns of performance. While we found that there were many characteristics associated with each pattern, they varied in the degree to which they served to distinguish one from another. Perhaps the strongest characteristic was the duration of the retention intervals, with shorter intervals being more likely to produce a curvilinear loss function, and longer intervals being more likely to produce linear, no net change, and improvement patterns. Our hope is that future work can use this as an inspiration to explore various theoretical ideas about the different patterns that cognitive and neurological processes are likely to produce and why they would do so.
Data Availability
All data used for our analyses are availble on-line at https://osf.io/wq9ty.
Notes
Like Rubin and Wenzel (1996a, 1996b), we only consider two-parameter functions. Wickelgren (1974) also proposed a three-parameter function that is a variation of the power function. In addition to the a and b parameters, there is a third parameter, c, that is set at the level of memory performance at time 0. The equation is M = c(at + 1)-b . With an extra parameter, this formula better fits some data sets than any two-parameter function. We do not consider it here for several reasons. First, this requires additional information, namely memory at time 0, which is simply not available for many of our data sets. Second, in some cases, memory at time 0 can be set to 1, but not in all cases. Finally, in some cases where an estimate of memory is provided at time 0, we can address this using a very small value for time close to, but not equaling 0, as with the power function.
Many studies reported data in a figure. To extract the data, we used Web Plot Digitizer (Rohatgi, 2019).
Note that for the studies with poor fits that were removed from our analyses, the power function was the best-fit one most often.
Memory test types (e.g., free recall, cued recall) were dummy coded for these analyses.
References
Aleman, A., Hijman, R., De Haan, E. H., & Kahn, R. S. (1999). Memory impairment in schizophrenia: A meta-analysis. American Journal of Psychiatry, 156(9), 1358–1366.
Anderson, J. R., & Schooler, L. J. (1991). Reflections of the environment in memory. Psychological Science, 2(6), 396–408.
Anderson, R. B. (2001). The power law as an emergent property. Memory & Cognition, 29(7), 1061–1068.
Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. The Psychology of Learning and Motivation, 2, 89–195.
Averell, L., & Heathcote, A. (2011). The form of the forgetting curve and the fate of memories. Journal of Mathematical Psychology, 55(1), 25–35.
Bahrick, H. P. (1984). Semantic memory content in permastore: Fifty years of memory for Spanish learned in school. Journal of Experimental Psychology: General, 113(1), 1–29.
Bauer, P. J., & Larkina, M. (2014). Childhood amnesia in the making: Different distributions of autobiographical memories in children and adults. Journal of Experimental Psychology: General, 143(2), 597–611.
Burtt, H. E., & Dobell, E. M. (1925). The curve of forgetting for advertising material. Journal of Applied Psychology, 9(1), 5–21.
Button, K. S., Ioannidis, J., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S., & Munafò, M. R. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5), 365–376.
Carpenter, S. K., Pashler, H., Wixted, J. T., & Vul, E. (2008). The effects of tests on learning and forgetting. Memory & Cognition, 36(2), 438–448.
Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155–59. https://doi.org/10.1037/0033-2909.112.1.155
Cohen, J. (1992). Statistical power analysis. Current Directions in Psychological Science, 1(3), 98–101.
Cohen, N., Eichenbaum, H., Deacedo, B., & Corkin, S. (1985). Different memory systems underlying acquisition of procedural and declarative knowledge. Annals of the New York Academy of Sciences, 444, 54–71.
Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Progress in Brain Research, 169, 323–338.
Craik, F. I., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11(6), 671–684.
Cronbach, L., Gleser, G., Nanda, H., & Rajaratnam, N. (1972). The dependability of behavioral measurement: Theory of generalizability for scores and profiles. Wiley.
Crovitz, H. F., & Quina-Holland, K. (1976). Proportion of episodic memories from early childhood by years of age. Bulletin of the Psychonomic Society, 7(1), 61–62.
Crovitz, H. F., & Schiffman, H. (1974). Frequency of episodic memories as a function of their age. Bulletin of the Psychonomic Society, 4(5), 517–518.
Dudai, Y., Kami, A., & Born, J. (2005). The consolidation and transformation of memory. Neuron, 88(1), 20–32.
Ebbinghaus, H. (1885). Memory: A contribution to experimental psychology (H. A. Ruger & C. E. Bussenius, Trans.). Teachers College, Columbia University.
Elliott, G., Isaac, C. L., & Muhlert, N. (2014). Measuring forgetting: A critical review of accelerated long-term forgetting studies. Cortex, 54, 16–32.
Fioravanti, M., & Di Cesare, F. (1992). Forgetting curves in long-term memory: Evidence for a multistage model of retention. Brain and Cognition, 18(2), 116–124.
Fisher, J. S., & Radvansky, G. A. (2019). Linear forgetting. Journal of Memory and Language, 108, 104035.
Fisher, J. S., & Radvansky, G. A. (2022). Degree of learning and linear forgetting. Quarterly Journal of Experimental Psychology, 75(8), 1483–1496.
Geurts, S., van der Werf, S. P., & Kessels, R. P. (2015). Accelerated forgetting? An evaluation on the use of long-term forgetting rates in patients with memory problems. Frontiers in Psychology, 6, 752.
Greenwald, A. G. (1976). Within-subjects designs: To use or not to use? Psychological Bulletin, 83(2), 314–320.
Haist, F., Shimamura, A. P., & Squire, L. R. (1992). On the relationship between recall and recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18(4), 691–702.
Harnett, P., McCarthy, D., & Davison, M. (1984). Delayed signal detection, differential reinforcement, and short-term memory in the pigeon. Journal of the Experimental Analysis of Behavior, 42(1), 87–111.
Hollingworth, H. L. (1913). Characteristic differences between recall and recognition. American Journal of Psychology, 24(4), 532–544.
Jones, G. V. (1976). A fragmentation hypothesis of memory: Cued recall of pictures and of sequential position. Journal of Experimental Psychology: General, 105(3), 277–293.
Kintsch, W., Welsch, D., Schmalhofer, F., & Zimny, S. (1990). Sentence memory: A theoretical analysis. Journal of Memory and Language, 29(2), 133–159.
Linton, M. (1982). Transformations of memory in everyday life. In U. Neisser (Ed.), Memory observed: Remembering in natural contexts (pp. 77–91). Freeman.
Loftus, G. R. (1985). Evaluating forgetting curves. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11(2), 397–406.
Loftus, G. R. (1985). Consistency and confoundings: Reply to Slamecka. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11(4), 817–820.
Marcoulides, G. A. (1993). Maximizing power in generalizability studies under budget constraints. Journal of Educational Statistics, 18(2), 197–206.
McGaugh, J. L. (1966). Time-dependent processes in memory storage. Science, 153(3742), 1351–1358.
McGaugh, J. L. (2000). Memory–A century of consolidation. Science, 287(5451), 248–251.
Paivio, A. (1971). Imagery and verbal processes. Holt, Rinehart & Winston.
Payne, D. G. (1987). Hypermnesia and reminiscence in recall: A historical and empirical review. Psychological Bulletin, 101(1), 5–27.
Premack, D. (2007). Human and animal cognition: Continuity and discontinuity. Proceedings of the National Academy of Sciences, 104(35), 13861–13867.
Quinlan, J. (2014). C4. 5: Programs for machine learning. Elsevier.
Radosavljevich, P. R. (1907). Das Behalten und Vergessen bei Kindern und Erwachsenen nach experimentellen Untersuchungen [An experimental investigation of the retention and forgetting of children and adults]. Nemnich.
Radvansky, G. A., Andrea, E. O., & Fisher, J. S. (2017). Event models and the fan effect. Memory & Cognition, 45(6), 1028–1044.
Radvansky, G. A., Doolen, A. C., Pettijohn, K. A., & Ritchie, M. (2022). A new look at memory retention and forgetting. Journal of Experimental Psychology: Learning, Memory, and Cognition, 48(11), 1698–1723.
Ranganath, C., & Blumenfeld, R. S. (2005). Doubts about double dissociations between short-and long-term memory. Trends in Cognitive Sciences, 9(8), 374–380.
Ricker, T. J., Sandry, J., Vergauwe, E., & Cowan, N. (2019). Do familiar memory items decay? Journal of Experimental Psychology: Learning, Memory, and Cognition, 46(1), 60–76.
Rivera-Lares, K., Logie, R., Baddeley, A., & Della Sala, S. (2022). Rate of forgetting is independent of initial degree of learning. Memory & Cognition, 50(8), 1706–1718.
Roediger, H. L., III., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
Rohatgi, A. (2019). WebPlotDigitizer (Version 4.2) [Computer software]. https://apps.automeris.io/wpd/
Rohrer, D., & Taylor, K. (2006). The effects of overlearning and distributed practise on the retention of mathematics knowledge. Applied Cognitive Psychology, 20(9), 1209–1224.
Rubin, D. C., & Wenzel, A. E. (1996). One hundred years of forgetting: A quantitative description of retention. Psychological Review, 103(4), 734–760.
Serengil, S. I., & Teknoloji, Y. K. (2021). ChefBoost: A lightweight boosted decision tree framework [Computer software]. https://doi.org/10.5281/zenodo.5576202
Slamecka, N. J. (1985). On comparing rates of forgetting: Comment on Loftus (1985). Journal of Experimental Psychology: Learning, Memory, and Cognition, 11(4), 812–816.
Slamecka, N. J., & McElree, B. (1983). Normal forgetting of verbal lists as a function of their degree of learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(3), 384–397.
Smith, P. L., & Little, D. R. (2018). Small is beautiful: In defense of the small-N design. Psychonomic Bulletin & Review, 25(6), 2083–2101.
Squire, L. R. (1986). Mechanisms of memory. Science, 232(4758), 1612–1619.
Squire, L. R. (2004). Memory systems of the brain: A brief history and current perspective. Neurobiology of Learning and Memory, 82(3), 171–177.
Squire, L. R., & Alvarez, P. (1995). Retrograde amnesia and memory consolidation: A neurobiological perspective. Current Opinion in Neurobiology, 5(2), 169–177.
Staddon, J. E. R. (1983). Adaptive behavior and learning. Cambridge University Press.
Staugaard, S. R., & Berntsen, D. (2019). Retrieval intentionality and forgetting: How retention time and cue distinctiveness affect involuntary and voluntary retrieval of episodic memories. Memory & Cognition, 47(5), 893–905.
Vallar, G., & Baddeley, A. D. (1982). Short-term forgetting and the articulatory loop. Quarterly Journal of Experimental Psychology Section, 34A(1), 53–60.
VanVoorhis, C. W., & Morgan, B. L. (2007). Understanding power and rules of thumb for determining sample sizes. Tutorials in Quantitative Methods for Psychology, 3(2), 43–50.
Wickelgren, W. A. (1972). Trace resistance and the decay of long-term memory. Journal of Mathematical Psychology, 9(4), 418–455.
Wickelgren, W. A. (1974). Single-trace fragility theory of memory dynamics. Memory & Cognition, 2(4), 775–780.
Wickelgren, W. (1975). Alcoholic intoxication and memory storage dynamics. Memory & Cognition, 3(4), 385–389.
Wixted, J. T. (2004). On common ground: Jost’s (1897) law of forgetting and Ribot’s (1881) law of retrograde amnesia. Psychological Review, 111(4), 864–879.
Wixted, J. T. (2022). Absolute versus relative forgetting. Journal of Experimental Psychology: Learning, Memory, and Cognition, 48(12), 1775–1786.
Wixted, J. T., & Carpenter, S. K. (2007). The Wickelgren power law and the Ebbinghaus savings function. Psychological Science, 18(2), 133–134. https://doi.org/10.1111/j.1467-9280.2007.01862.x
Wixted, J. T., & Ebbesen, E. B. (1991). On the form of forgetting. Psychological Science, 2(6), 409–415.
Woodworth, R. S. (1938). Experimental psychology. Henry Holt.
Papers included in our analyses
Abel, M., Haller, V., Kock, H., Potshke, S., Hieb, D., Schabus, M., & Bauml, K. (2019). Sleep reduces the testing effect—but not after corrective feedback and prolonged retention interval. Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(2), 272–287.
Albert, M. S., Butters, N., & Brandt, J. (1981). Development of remote memory loss in patients with Huntington’s disease. Journal of Clinical Neuropsychology, 3(1), 1–12.
Albert, M. S., Butters, N., & Brandt, J. (1981). Patterns of remote memory in amnesic and demented patients. Archives of Neurology, 38(8), 495–500.
Albert, M. S., Butters, N., & Levin, J. (1979). Temporal gradients in the retrograde amnesia of patients with alcoholic Korsakoff’s disease. Archives of Neurology, 36(4), 211–216.
Austin, S. D. M. (1921). A study in logical memory. The American Journal of Psychology, 32(3), 370–403.
Averell, L., & Heathcote, A. (2009). Long term implicit and explicit memory for briefly studied words. In A. Taatgen & H. van Rijn (Eds.), Proceedings of the 31st Annual Conference of the Cognitive Science Society (pp. 267–281). Cognitive Science Society.
Bahrick, H. P. (1983). The cognitive map of a city: Fifty years of learning and memory. Psychology of Learning and Motivation, 17, 125–163.
Bahrick, H. P. (1984). Semantic memory content in permastore: Fifty years of memory for Spanish learned in school. Journal of Experimental Psychology: General, 113(1), 1–29.
Bahrick, H. P., Bahrick, P. O., & Wittlinger, R. P. (1975). Fifty years of memory for names and faces: A cross-sectional approach. Journal of Experimental Psychology: General, 104(1), 54–75.
Bahrick, H. P., Hall, L. K., & Da Costa, L. A. (2008). Fifty years of memory of college grades: Accuracy and distortions. Emotion, 8(1), 13–22.
Barclay, C. R., & Wellman, H. M. (1986). Accuracies and inaccuracies in autobiographical memories. Journal of Memory and Language, 25(1), 93–103.
Barr, W. B., Goldberg, E., Wasserstein, J., & Novelly, R. A. (1990). Retrograde amnesia following unilateral temporal lobectomy. Neuropsychologia, 28(3), 243–255.
Bauer, P. J., & Larkina, M. (2016). Predicting remembering and forgetting of autobiographical memories in children and adults: a 4-year prospective study. Memory, 24(10), 1345–1368.
Bauer, P. J., Stweart, R., Sirkin, R. E., & Larkina, M. (2017). Robust memory of where from way back when: Evidence from behaviour and visual attention. Memory, 25(8), 1089–1109.
Bean, C. H. (1912). The curve of forgetting. Press of the New Era Printing Company.
Beatty, W. W., Bailly, R. C., & Fisher, L. (1989). Korsakoff-like amnesic syndrome in a patient with anorexia and vomiting. International Journal of Clinical Neuropsychology, 11(2), 55–65.
Beatty, W. W., Goodkin, D. E., Monson, N., Beatty, P. A., & Hertsgaard, D. (1988). Anterograde and retrograde amnesia in patients with chronic progressive multiple sclerosis. Archives of Neurology, 45(6), 611–619.
Beatty, W. W., Salmon, D. P., Bernstein, N., & Butters, N. (1987). Remote memory in a patient with amnesia due to hypoxia. Psychological Medicine, 17(3), 657–665.
Beatty, W. W., Salmon, D. P., Butters, N., Heindel, W. C., & Granholm, E. L. (1988). Retrograde amnesia in patients with Alzheimer’s disease or Huntington’s disease. Neurobiology of Aging, 9, 181–186.
Begg, I., & Wickelgren, W. A. (1974). Retention functions for syntactic and lexical vs semantic information in sentence recognition memory. Memory & Cognition, 2(2), 353–359.
Bell, B. D. (2006). WMS-III logical memory performance after a two-week delay in temporal lobe epilepsy and control groups. Journal of Clinical and Experimental Neuropsychology, 28(8), 1435–1443.
Bell, B. D., Fine, J., Dow, C., Seidenberg, M., & Hermann, B. P. (2005). Temporal lobe epilepsy and the selective reminding test: The conventional 30-minute delay suffices. Psychological Assessment, 17(1), 103–109.
Blake, R. V., Wroe, S. J., Breen, E. K., & McCarthy, R. A. (2000). Accelerated forgetting in patients with epilepsy: Evidence for an impairment in memory consolidation. Brain, 123(3), 472–483.
Bregman, A. S. (1968). Forgetting curves with semantic, phonetic, graphic, and contiguity cues. Journal of Experimental Psychology, 78(4), 539–546.
Briggs, G. E. (1954). Acquisition, extinction, and recovery functions in retroactive inhibition. Journal of Experimental Psychology, 47(5), 285–293.
Burt, C. D. B. (1992). Retrieval characteristics of autobiographical memories: Event and date information. Applied Cognitive Psychology, 6(5), 389–404.
Burtt, H., & Dobell, E. (1925). The curve of forgetting for advertising material. Journal of Applied Psychology, 9(1), 5–21.
Butler, C., Kapur, N., Zeman, A., Weller, R., & Connelly, A. (2012). Epilepsy-related long-term amnesia: Anatomical perspectives. Neuropsychologia, 50(13), 2973–2980.
Butler, C. R., Graham, K. S., Hodges, J. R., Kapur, N., Wardlaw, J. M., & Zeman, A. Z. (2007). The syndrome of transient epileptic amnesia. Annals of Neurology, 61(6), 587–598.
Butter, M. J. (1970). Differential recall of paired associates as a function of arousal and concreteness-imagery levels. Journal of Experimental Psychology, 84(2), 252–256.
Butters, N. (1984). Alcoholic Korsakoff’s syndrome: An update. Seminars in Neurology, 4(2), 226–244.
Butters, N., & Albert, M. S. (1982). Processes underlying failures to recall remote events. In L. S. Cermak (Ed.), Human memory and amnesia (pp. 257–274). Erlbaum.
Butters, N., Cermak, L. S., Montgomery, K., & Adinolfi, A. (1977). Some comparisons of the memory and visuoperceptive deficits of chronic alcoholics and patients with Korsakoff’s disease. Alcoholism: Clinical and Experimental Research, 1(1), 73–80.
Carlesimo, G. A., Sabbadini, M., Fadda, L., & Caltagirone, C. (1995). Forgetting from long-term memory in dementia and pure amnesia: Role of task, delay of assessment and aetiology of cerebral damage. Cortex, 31(2), 285–300.
Carpenter, S. K., Pashler, H., Wixted, J. T., & Vul, E. (2008). The effects of tests on learning and forgetting. Memory & Cognition, 36(2), 438–448.
Catal, L. L., & Fitzgerald, J. M. (2004). Autobiographical memories in two older adults over a twenty-year retention interval. Memory & Cognition, 32(2), 311–323.
Cepeda, N., Vul, E., Rohrer, D., Wixted, J., & Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science, 19(11), 1095–1102.
Cepeda, N. J., Coburn, N., Rohrer, D., Wixted, J. T., Mozer, M. C., & Pashler, H. (2009). Optimizing distributed practice: Theoretical analysis and practical implications. Experimental Psychology, 56(4), 236–246.
Cermak, L. S., & O’Connor, M. (1983). The anterograde and retrograde retrieval ability of a patient with amnesia due to encephalitis. Neuropsychologia, 21(3), 213–234.
Chan, J. C. K. (2010). Long-term effects of testing on the recall of nontested materials. Memory, 18(1), 49–57.
Chance, J. E., & Goldstein, A. G. (1987). Retention interval and face recognition: Response latency measures. Bulletin of the Psychonomic Society, 25(6), 415–418.
Cohen, N. J., & Squire, L. R. (1981). Retrograde amnesia and remote memory impairment. Neuropsychologia, 19(3), 337–356.
Conway, M. A., Cohen, G., & Stanhope, N. (1991). On the very long-term retention of knowledge acquired through formal education: Twelve years of cognitive psychology. Journal of Experimental Psychology: General, 120(4), 395–409.
Courtois, M. R., & Mueller, J. H. (1981). Target and distractor typicality in facial recognition. Journal of Applied Psychology, 66(5), 639–645.
Craig, C., Sternthal, B., & Olshan, K. (1972). The effect of overlearning on retention. Journal of General Psychology, 87, 85–94.
Cronel-Ohayon, S., Zesiger, P., Davidoff, V., Boni, A., Roulet, E., & Deonna, T. (2006). Deficit in memory consolidation (abnormal forgetting rate) in childhood temporal lobe epilepsy: Pre and postoperative long-term observation. Neuropediatrics, 37(06), 317–324.
Davidson, D. (1994). Recognition and recall of irrelevant and interruptive atypical actions in script-based stories. Journal of Memory and Language, 33(6), 757–775.
D’Esposito, M., Alexander, M. P., Fischer, R., Mcglinchey-Berroth, R., & O’connor, M. (1996). Recovery of memory and executive function following anterior communicating artery aneurysm rupture. Journal of the International Neuropsychological Society, 2(6), 565–570.
Dewar, M., Alber, J., Butler, C., Cowan, N., & Della Salla, S. (2012). Brief wakeful resting boosts new memories over the long term. Psychological Science, 23(9), 955–960.
Doolen, A. C., & Radvansky, G. A. (2021). A novel study: Long-lasting event memory. Memory, 29(8), 963–982. https://doi.org/10.1080/09658211.2021.1953079
Ebbinghaus, H. (1913). Memory: A contribution to experimental psychology (H. A. Ruger & C. E. Bussenius, Trans.). Columbia University, Teacher’s College. (Original publication 1885; Reprinted 1964, Dover).
Ellis, J. A., Semb, G. B., & Cole, B. (1998). Very long-term memory for information taught in school. Contemporary Educational Psychology, 23(4), 419–433.
Epstein, M. L., & Phillips, W. D. (1976). Delayed recall of paired associates as a function of processing level. The Journal of General Psychology, 95(1), 127–132.
Evans, S. J., Elliott, G., Reynders, H., & Isaac, C. L. (2014). Can temporal lobe epilepsy surgery ameliorate accelerated long-term forgetting? Neuropsychologia, 53, 64–74.
Finkenbinder, E. O. (1913). The curve of forgetting. American Journal of Psychology, 24(1), 8–32.
Fioravanti, M., & Di Cesare, F. (1992). Forgetting curves in long-term memory: Evidence for a multistage model of retention. Brain and Cognition, 18(2), 116–124.
Fisher, J. S., & Radvansky, G. A. (2018). Patterns of forgetting. Journal of Memory and Language, 102, 130–141.
Fisher, J. S., & Radvansky, G. A. (2022). Degree of learning and linear forgetting. Quarterly Journal of Experimental Psychology, 75(8), 1483–1496.
Fitzgerald, Z., Thayer, Z., Mohamed, A., & Miller, L. A. (2013). Examining factors related to accelerated long-term forgetting in epilepsy using ambulatory EEG monitoring. Epilepsia, 54(5), 819–827.
Flores, S., Bailey, H. R., Eisenberg, M. L., & Zacks, J. M. (2017). Event segmentation improves event memory up to one month later. Journal of Experimental Psychology: Learning, Memory, and Cognition, 43(8), 1183–1202.
Freedman, M., Rivoira, P., Butters, N., Sax, D. S., & Feldman, R. G. (1984). Retrograde amnesia in Parkinson’s disease. Canadian Journal of Neurological Sciences, 11(2), 297–301.
Gade, A., & Mortensen, E. L. (1990). Temporal gradient in the remote memory impairment of amnesic patients with lesions in the basal forebrain. Neuropsychologia, 28(9), 985–1001.
Gaudino, E. A., Chiaravalloti, N. D., DeLuca, J., & Diamond, B. J. (2001). A comparison of memory performance in relapsing–remitting, primary progressive and secondary progressive, multiple sclerosis. Cognitive and Behavioral Neurology, 14(1), 32–44.
Gehring, R. E., Toglia, M. P., & Kimble, G. A. (1976). Recognition memory for words and pictures at short and long retention intervals. Memory & Cognition, 4(3), 256–260.
Gilbert, T. (1957). Overlearning and the Retention of Meaningful Prose. The Journal of General Psychology, 56(2), 281–289.
Gilliland, A. R. (1948). The rate of forgetting. Journal of Educational Psychology, 39(1), 19–26.
Giovagnoli, A. R., Casazza, M., & Avanzini, G. (1995). Visual learning on a selective reminding procedure and delayed recall in patients with temporal lobe epilepsy. Epilepsia, 36(7), 704–711.
Glasnapp, D. R., Poggio, J. P., & Ory, J. C. (1978). End-of-course and long-term retention outcomes for mastery and nonmastery learning paradigms. Psychology in the Schools, 15(4), 595–603.
Goetschalckx, L., Moors, P., & Wagemans, J. (2018). Image memorability across longer time intervals. Memory, 26(5), 581–588.
Goldsmith, M., Koriat, A., & Pansky, A. (2005). Strategic regulation of grain size in memory reporting over time. Journal of Memory and Language, 52, 505–525.
Graf, P., & Mandler, G. (1984). Activation makes words more accessible, but not necessarily more retrievable. Journal of Verbal Learning and Verbal Behavior, 23(5), 553–568.
Graves, E. A. (1936). The effect of sleep upon retention. Journal of Experimental Psychology, 19(3), 316–322.
Haist, F., Gore, J. B., & Mao, H. (2001). Consolidation of human memory over decades revealed by functional magnetic resonance imaging. Nature Neuroscience, 4(11), 1139–1145.
Harnett, P., McCarthy, D., & Davison, M. (1984). Delayed signal detection, differential reinforcement, and short-term memory in the pigeon. Journal of the Experimental Analysis of Behavior, 42(1), 87–111.
Hasher, L., & Griffin, M. (1978). Reconstructive and reproductive processes in memory. Journal of Experimental Psychology: Human Learning and Memory, 4(4), 318–330.
Hashtroudi, S., Ferguson, S., Rappold, V., Chrosniak, L., & Roediger, Henry L. (1988). Data-driven and conceptually driven processes in partial-word identification and recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(4), 749–757.
Hellyer, S. (1962). Supplementary report: Frequency of stimulus presentation and short-term decrement in recall. Journal of Experimental Psychology, 64(6), 650.
Hirst, W., Phelps, E. A., Meksin, R., Vaidya, C. J., Johnson, M. K., Mitchell, K. J., Buckner, R. L., Budson, A. E., Gabrieli, J. D. E., Lustig, C., Mather, M., Ochsner, K. N., Schacter, D., Simons, J. S., Lyle, K. B., Cuc, A. F., & Olsson, A. (2015). A ten-year follow-up of a study of memory for the attack of September 11, 2001: Flashbulb memories and memories for flashbulb events. Journal of Experimental Psychology: General, 144(3), 604–623.
Howarth, E., & Eysenck, H. J. (1969). Extraversion, arousal, and paired-associate recall. Journal of Experimental Research in Personality, 3, 114–116.
Huber, S. J., Shuttleworth, E. C., & Paulson, G. W. (1986). Dementia in Parkinson’s disease. Archives of Neurology, 43(10), 987–990.
Hunkin, N. M., Parkin, A. J., Bradley, V. A., Burrows, E. H., Aldrich, F. K., Jansari, A., & Burdon-Cooper, C. (1995). Focal retrograde amnesia following closed head injury: A case study and theoretical account. Neuropsychologia, 33(4), 509–523.
Huppert, F. A., & Piercy, M. (1978). Dissociation between learning and remembering in organic amnesia. Nature, 275(5678), 317–318.
Huppert, F. A., & Piercy, M. (1979). Normal and abnormal forgetting in organic amnesia: Effect of locus of lesion. Cortex, 15(3), 385–390.
Isaac, C. L., & Mayes, A. R. (1999). Rate of forgetting in amnesia: I. Recall and recognition of prose. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(4), 942–962.
Jackson, H. F., & Bentall, R. P. (1991). Operant conditioning in amnesic subjects: Response patterning and sensitivity to schedule changes. Neuropsychology, 5(2), 89–105.
Jansari, A. S., Davis, K., McGibbon, T., Firminger, S., & Kapur, N. (2010). When “long-term memory” no longer means “forever”: Analysis of accelerated long-term forgetting in a patient with temporal lobe epilepsy. Neuropsychologia, 48(6), 1707–1715.
Jenkins, J. G., & Dallenbach, K. M. (1924). Obliviance during sleep and waking. American Journal of Psychology, 35(4), 605–612.
Jeunehomme, O., Folville, A., Stawarczyk, D., Van der Linden, M., & D’Argembeau, A. (2018). Temporal compression in episodic memory for real-life events. Memory, 26(6), 759–770.
Jones, H. E. (1923). Experimental studies of college teaching: The effect of examination on the permanence of forgetting. Archives of Psychology, 10, 1–70.
Kaplan, R., & Kaplan, S. (1969). The arousal-retention interval interaction revisited: The effects of some procedural changes. Psychonomic Science, 15(2), 84–85.
Kapur, N., Heath, P., Meudell, P., & Kennedy, P. (1986). Amnesia can facilitate memory performance: Evidence from a patient with dissociated retrograde amnesia. Neuropsychologia, 24(2), 215–221.
Kapur, N., Millar, J., Colbourn, C., Abbott, P., Kennedy, P., & Docherty, T. (1997). Very long-term amnesia in association with temporal lobe epilepsy: Evidence for multiple-stage consolidation processes. Brain and Cognition, 35(1), 58–70.
Kapur, N., Scholey, K., Moore, E., Barker, S., Brice, J., Thompson, S., Shiel, A., Carn, R., Abbott, P., & Fleming, J. (1996). Long-term retention deficits in two cases of disproportionate retrograde amnesia. Journal of Cognitive Neuroscience, 8(5), 416–434.
Kapur, N., Young, A., Bateman, D., & Kennedy, P. (1989). Focal retrograde amnesia: A long term clinical and neuropsychological follow-up. Cortex, 25(3), 387–402.
Kemp, S., Illman, N. A., Moulin, C. J., & Baddeley, A. D. (2012). Accelerated long-term forgetting (ALF) and transient epileptic amnesia (TEA): Two cases of epilepsy-related memory disorder. Epilepsy and Behavior, 24(3), 382–388.
Kleinsmith, L. J., & Kaplan, S. (1963). Paired-associate learning as a function of arousal and interpolated interval. Journal of Experimental Psychology, 65(2), 190–193.
Kleinsmith, L. J., & Kaplan, S. (1964). Interaction of arousal and recall interval in nonsense syllable paired-associate learning. Journal of Experimental Psychology, 67(2), 124–126.
Komatsu, S. I., & Ohta, N. (1985). Priming effects in word-fragment completion for short-and long-term retention intervals. Japanese Psychological Research, 26(4), 194–200.
Kopelman, M. D. (1985). Rates of forgetting in Alzheimer-type dementia and Korsakoff’s syndrome. Neuropsychologia, 23(5), 623–638.
Kopelman, M. D. (1989). Remote and autobiographical memory, temporal context memory and frontal atrophy in Korsakoff and Alzheimer patients. Neuropsychologia, 27(4), 437–460.
Kopelman, M. D., Stanhope, N., & Kingsley, D. (1999). Retrograde amnesia in patients with diencephalic, temporal lobe or frontal lesions. Neuropsychologia, 37(8), 939–958.
Koppenaal, R. J. (1963). Time changes in the strengths of A-B, A–C lists; spontaneous recovery? Journal of Verbal Learning and Verbal Behavior, 2(4), 310–319.
Kristo, G., Janssen, S. M., & Murre, J. M. (2009). Retention of autobiographical memories: An Internet-based diary study. Memory, 17(8), 816–829.
Krueger, W., & Bentley, M. (1929). The effect of overlearning on retention. Journal of Experimental Psychology, 12(1), 71–78.
Lavach, J. F. (1971). The effects of emotional arousal on short vs. long term retention of continuously presented information (Final report). College of William and Mary.
Lawless, H. T. (1978). Recognition of common odors, pictures, and simple shapes. Perception & Psychophysics, 24(6), 493–495.
Legenfelder, J., Chiaravalloti, N. D., & DeLuca, J. (2007). The efficacy of the generation effect in improving new learning in persons with traumatic brain injury. Rehabilitation Psychology, 52(3), 290–296.
LePort, A. K., Stark, S. M., McGaugh, J. L., & Stark, C. E. (2016). Highly superior autobiographical memory: Quality and quantity of retention over time. Frontiers in Psychology, 6, 2017.
Levin, H. S., High, W. M., & Eisenberg, H. M. (1988). Learning and forgetting during posttraumatic amnesia in head injured patients. Journal of Neurology, Neurosurgery and Psychiatry, 51(1), 14–20.
Linton, M. (1982). Transformations of memory in everyday life. In U. Neisser (Ed.), Memory observed: Remembering in natural contexts (pp. 77–91). Freeman.
Loftus, E. F., Miller, D. G., & Burns, H. J. (1978). Semantic integration of verbal information into a visual memory. Journal of Experimental Psychology: Human Learning and Memory, 4(1), 19–31.
Lucchelli, F., & Spinnler, H. (1998). Ephemeral new traces and evaporated remote engrams: A form of neocortical temporal lobe amnesia? A preliminary case report. Neurocase, 4(6), 447–459.
Luh, C. W. (1922). The conditions of retention. Psychological Monographs, 31(3), i–87.
MacLeod, C. (1988). Forgotten but not gone: Savings for pictures and words in long-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(2), 195–212.
Mameniskiene, R., Jatuzis, D., Kaubrys, G., & Budrys, V. (2006). The decay of memory between delayed and long-term recall in patients with temporal lobe epilepsy. Epilepsy and Behavior, 8(1), 278–288.
Mammarella, N., Farifield, B., & Cornoldi, C. (2007). Reality monitoring and resistance to forgetting under short delay intervals. Quarterly Journal of Experimental Psychology, 60(4), 551–570.
Manes, F., Graham, K. S., Zeman, A., de Lujan Calcagno, M., & Hodges, J. R. (2005). Autobiographical amnesia and accelerated forgetting in transient epileptic amnesia. Journal of Neurology, Neurosurgery and Psychiatry, 76(10), 1387–1391.
Markowitsch, H. J., Calabrese, P., Liess, J., Haupts, M., Durwen, H. F., & Gehlen, W. (1993). Retrograde amnesia after traumatic injury of the fronto-temporal cortex. Journal of Neurology, Neurosurgery and Psychiatry, 56(9), 988–992.
Markowitsch, H. J., Weber-Luxemburger, G., Ewald, K., Kessler, J., & Heiss, W. D. (1997). Patients with heart attacks are not valid models for medial temporal lobe amnesia. A neuropsychological and FDG-PET study with consequences for memory research. European Journal of Neurology, 4(2), 178–184.
Marslen-Wilson, W. D., & Teuber, H. L. (1975). Memory for remote events in anterograde amnesia: Recognition of public figures from news photographs. Neuropsychologia, 13(3), 353–364.
Martin, R. C., Loring, D. W., Meador, K. J., Lee, G. P., Thrash, N., & Arena, J. G. (1991). Impaired long-term retention despite normal verbal learning in patients with temporal lobe dysfunction. Neuropsychology, 5(1), 3–12.
Martini, M., Martini, C., Maran, T., & Sachse, P. (2018). Effects of post-encoding wakeful rest and study time on long-term memory performance. Journal of Cognitive Psychology, 30(5–6), 558–569.
Mayes, A. R., Daum, I., Markowisch, H. J., & Sauter, B. (1997). The relationship between retrograde and anterograde amnesia in patients with typical global amnesia. Cortex, 33(2), 197–217.
Mayes, A. R., Isaac, C. L., Holdstock, J. S., Cariga, P., Gummer, A., & Roberts, N. (2003). Long-term amnesia: A review and detailed illustrative case study. Cortex, 39(4/5), 567–603.
Mayes, A. R., Meudell, P. R., Mann, D., & Pickering, A. (1988). Location of lesions in Korsakoff’s syndrome: Neuropsychological and neuropathological data on two patients. Cortex, 24(3), 367–388.
McBride, D. M., & Dosher, B. A. (1997). A comparison of forgetting in an implicit and explicit memory task. Journal of Experimental Psychology: General, 126(4), 371–392.
McBride, D. M., & Dosher, B. A. (1999). Forgetting rates are comparable in conscious and automatic memory: A process-dissociation study. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(3), 583–607.
McGibbon, T., & Jansari, A. S. (2013). Detecting the onset of accelerated long-term forgetting: Evidence from temporal lobe epilepsy. Neuropsychologia, 51(1), 114–122.
McKee, R. D., & Squire, L. R. (1992). Equivalent forgetting rates in long-term memory for diencephalic and medial temporal lobe amnesia. Journal of Neuroscience, 12(10), 3765–3772.
Meeter, M., Murre, J. M., & Janssen, S. M. (2005). Remembering the news: Modeling retention data from a study with 14,000 participants. Memory & Cognition, 33(5), 793–810.
Meudell, P. R., Northen, B., Snowden, J. S., & Neary, D. (1980). Long term memory for famous voices in amnesic and normal subjects. Neuropsychologia, 18(2), 133–139.
Montaldi, D., & Parkin, A. J. (1989). Retrograde amnesia in Korsakoff’s syndrome: An experimental and theoretical analysis. In J. R. Crawford & D. M. Parker (Eds.), Developments in clinical and experimental neuropsychology (pp. 213–218). Springer.
Muhlert, N., Grünewald, R. A., Hunkin, N. M., Reuber, M., Howell, S., Reynders, H., & Isaac, C. L. (2011). Accelerated long-term forgetting in temporal lobe but not idiopathic generalised epilepsy. Neuropsychologia, 49(9), 2417–2426.
Muhlert, N., Milton, F., Butler, C. R., Kapur, N., & Zeman, A. Z. (2010). Accelerated forgetting of real-life events in transient epileptic amnesia. Neuropsychologia, 48(11), 3235–3244.
Murdock, B. B. (1961). The retention of individual items. Journal of Experimental Psychology, 62(6), 618–625.
Mure, J. M. J., & Dross, J. (2015). Replication and analysis of Ebbinghaus’ Forgetting curve. PLOS ONE, 10(7), 1–23.
Nelson, T. O., Shimamura, A. P., & Leonesio, R. J. (1980). Large effects on long-term retention after standard list learning vs. adjusted learning. Behavior Research Methods & Instrumentation, 12(1), 42–44.
Nunoi, M., & Yoshikawa, S. (2016). Deep processing makes stimuli more preferable over long durations. Journal of Cognitive Psychology, 28(6), 756–763.
O’Connor, M., Sieggreen, M. A., Ahern, G., Schomer, D., & Mesulam, M. (1997). Accelerated forgetting in association with temporal lobe epilepsy and paraneoplastic encephalitis. Brain and Cognition, 35(1), 71–84.
Parkin, A. J., & Hunkin, N. M. (1991). Memory loss following radiotherapy for nasal pharyngeal carcinoma—an unusual presentation of amnesia. British Journal of Clinical Psychology, 30(4), 349–357.
Parkin, A. J., Montaldi, D., Leng, N. R., & Hunkin, N. M. (1990). Contextual cueing effects in the remote memory of alcoholic Korsakoff patients and normal subjects. Quarterly Journal of Experimental Psychology, 42(3), 585–596.
Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58(3), 193–198.
Postman, L., & Phillips, L. W. (1965). Short-term temporal changes in free recall. Quarterly Journal of Experimental Psychology, 17(2), 132–138.
Postman, L., Stark, K., & Fraser, J. (1968). Temporal changes in interference. Journal of Verbal Learning and Verbal Behavior, 7(3), 672–694.
Radosavljevich, P. R. (1907). Das Behalten und Vergessen bei Kindern und Erwachsenen nach experimentellen Untersuchungen [An experimental investigation of the retention and forgetting of children and adults]. Nemnich.
Radvansky, G., O’Rear, A., & Fisher, A. (2017). Event models and the fan effect. Memory & Cognition, 45(6), 1028–1044.
Raymaekers, L., Otgaar, H., & Smeets, T. (2014). The longevity of adaptive memory: Evidence for mnemonic advantages of survival processing 24 and 48 hours later. Memory, 22(1), 19–25.
Reed, J. M., & Squire, L. R. (1998). Retrograde amnesia for facts and events: Findings from four new cases. Journal of Neuroscience, 18(10), 3943–3954.
Reinvang, I., & Gjerstad, L. (1998). Focal retrograde amnesia associated with vascular headache. Neuropsychologia, 36(12), 1335–1341.
Ricker, T. J., & Cowan, N. (2010). Loss of visual working memory within seconds: The combined use of refreshable and non-refreshable features. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36(6), 1355–1368.
Ricker, T. J., & Cowan, N. (2014). Differences between presentation methods in working memory procedures: A matter of working memory consolidation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40(2), 417–428.
Ricker, T. J., Spiegel, L. R., & Cowan, N. (2014). Time-based loss in visual short-term memory is from trace decay, not temporal distinctiveness. Journal of Experimental Psychology: Learning, Memory, and Cognition, 40(6), 1510–1523.
Ricker, T. J., Sandry, J., Vergauwe, E., & Cowan, N. (2020). Do familiar memory items decay? Journal of Experimental Psychology: Learning, Memory, and Cognition, 46(1), 60–76.
Roediger, H., & Karpicke, J. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
Roediger, H., Weldon, M., Stadler, M., Riegler, G., & Rayner, Keith. (1992). Direct comparison of two implicit memory tests: Word fragment and word stem completion. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18(6), 1251–1269.
Rohrer, D., & Taylor, K. (2006). The effects of overlearning and distributed practise on the retention of mathematics knowledge. Applied Cognitive Psychology, 20(9), 1209–1224.
Rohrer, D., Taylor, K., Pashler, H., Wixted, J. T., & Cepeda, N. J. (2005). The effect of overlearning on long-term retention. Applied Cognitive Psychology, 19(3), 361–374.
Rowland, C., & Delosh, E. (2014). Mnemonic benefits of retrieval practice at short retention intervals. Memory, 23(3), 1–17.
Rubin, D. C. (1978). A unit of analysis of prose memory. Journal of Verbal Learning and Verbal Behavior, 17(5), 599–620.
Rubin, D. C., Hinton, S., & Wenzel, A. (1999). The precise time course of retention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(5), 1161–176.
Rubin, D. C., & Wenzel, A. E. (1996). One hundred years of forgetting: A quantitative description of retention. Psychological Review, 103(4), 734–760.
Runquist, W. N. (1983). Some effects of remembering on forgetting. Memory & Cognition, 11(6), 641–650.
Ryback, R. S., Weinert, J., & Fozard, J. L. (1970). Disruption of short-term memory in man following consumption of ethanol. Psychonomic Science, 20(6), 353–354.
Sagar, H. J., Cohen, N. J., Sullivan, E. V., Corkin, S., & Growdon, J. H. (1988). Remote memory function in Alzheimer’s disease and Parkinson’s disease. Brain, 111(1), 185–206.
Salmon, D. P., Lasker, B. R., Butters, N., & Beatty, W. W. (1988). Remote memory in a patient with circumscribed amnesia. Brain and Cognition, 7(2), 201–211.
Seamon, J. G., Luo, C. R., Kopecky, J. J., Price, C. A., Rothschild, L., Fung, N. S., & Schwartz, M. A. (2002). Are false memories more difficult to forget than accurate memories? The effect of retention interval on recall and recognition. Memory & Cognition, 30(7), 1054–1064.
Shapiro, L. (2006). Remembering September 11th: The role of retention interval and rehearsal on flashbulb and event memory. Memory, 14(2), 129–147.
Shepherd, J. W., & Ellis, H. D. (1973). The effect of attractiveness on recognition memory for faces. American Journal of Psychology, 86(3), 627–633.
Shepherd, J. W., Ellis, H. D., & Davies, G. M. (1982). Identification evidence: A psychological evaluation. Aberdeen University Press.
Siler, J., & Benjamin, A. S. (2019). Long-term inference and memory following retrieval practice. Memory & Cognition, 48, 645–654.
Simon, H. (1966). A note on Jost’s law and exponential forgetting. Psychometrika, 31(4), 505–506.
Slamecka, N. J., & McElree, B. (1983). Normal forgetting of verbal lists as a function of their degree of learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(3), 384–97.
Sloman, S. A., Hayman, C. A. G., Ohta, N., Law, J., & Tulving, E. (1988). Forgetting in primed fragment completion. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(2), 223–239.
Smith, D. A., & Graesser, A. C. (1981). Memory for actions in scripted activities as a function of typicality, retention interval, and retrieval task. Memory & Cognition, 9(6), 550–559.
Söderlund, H., Moscovitch, M., Kumar, N., Daskalakis, Z. J., Flint, A., Herrmann, N., & Levine, B. (2014). Autobiographical episodic memory in major depressive disorder. Journal of Abnormal Psychology, 123(1), 51–60.
Spikman, J. M., Berg, I. J., & Deelman, B. G. (1995). Spared recognition capacity in elderly and closed-head-injury subjects with clinical memory deficits. Journal of Clinical and Experimental Neuropsychology, 17(1), 29–34.
Squire, L. R. (1989). On the course of forgetting in very long-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15(2), 241–245.
Squire, L. R., & Cohen, N. J. (1982). Remote memory, retrograde amnesia, and the neuropsychology of memory. In L. Cermak (Ed.), Human memory and amnesia (pp. 275–303). Taylor & Francis.
Squire, L. R., Haist, F., & Shimamura, A. P. (1989). The neurology of memory: Quantitative assessment of retrograde amnesia in two groups of amnesic patients. Journal of Neuroscience, 9(3), 828–839.
Squire, L. R., Shimamura, A. P., & Graft, P. (1987). Strength and duration of priming effects in normal subjects and amnesic patients. Neuropsychologia, 25(1), 195–210.
Squire, L. R., & Slater, P. C. (1975). Forgetting in very long-term memory as assessed by an improved questionnaire technique. Journal of Experimental Psychology: Human Learning and Memory, 1(1), 50–54.
Squire, L. R., & Slater, P. C. (1978). Anterograde and retrograde memory impairment in chronic amnesia. Neuropsychologia, 16(3), 313–322.
Squire, L. R., Zola-Morgan, S., Cave, C. B., Haist, F., Musen, G., & Suzuki, W. A. (1990). Memory: Organization of brain systems and cognition. Cold Spring Harbor Symposia on Quantitative Biology, 55, 1007–1023.
Staats, A.W., Minke, K. A., & Stalling, R. B. (1970). Long-term retention of conditioned attitudes (No. TR-6). Hawaii University Honolulu, Department of Psychology.
Staddon, J. E. R. (1983). Adaptive behavior and learning. Cambridge University Press.
Stanhope, N., Cohen, G., & Conway, M. (1993). Very long-term retention of a novel. Applied Cognitive Psychology, 7(3), 239–256.
Staugaard, S. R., & Berntsen, D. (2019). Retrieval intentionality and forgetting: How retention time and cue distinctiveness affect involuntary and voluntary retrieval of episodic memories. Memory & Cognition, 47(5), 893–905.
Strong, E. K. (1913). The effect of time-interval upon recognition memory. Psychological Review, 20(5), 339–372.
Stuss, D. T., Guberman, A., Nelson, R., & Larochelle, S. (1988). The neuropsychology of paramedian thalamic infarction. Brain and Cognition, 8(3), 348–378.
Takashima, A., Petersson, K. M., Rutters, F., Tendolkar, I., Jensen, O., Zwarts, M. J., McNaughton, B. L., & Fernandez, G. (2006). Declarative memory consolidation in humans: A prospective functional magnetic resonance imaging study. Proceedings of the National Academy of Sciences of the United States of America, 103(3), 756–761.
Talamini, L. M., & Gorree, E. (2012). Aging memories: Differential decay of episodic memory components. Learning & Memory, 19(6), 239–246.
Tanaka, Y., Miyazawa, Y., Hashimoto, R., Nakano, I., & Obayashi, T. (1999). Postencephalitic focal retrograde amnesia after bilateral anterior temporal lobe damage. Neurology, 53(2), 344–344.
Thapar, A., & McDermott, K. (2001). False recall and false recognition induced by presentation of associated words: Effects of retention interval and level of processing. Memory & Cognition, 29(3), 424–432.
Thompson, C. P., Skowronski, J. J., Larsen, S. F., & Beltz, A. L. (1996). Autobiographical memory: Remembering what and remembering when. Erlbaum.
Tramoni, E., Felician, O., Barbeau, E. J., Guedj, E., Guye, M., Bartolomei, F., & Ceccaldi, M. (2011). Long-term consolidation of declarative memory: Insight from temporal lobe epilepsy. Brain, 134(3), 816–831.
Tunney, R. J., & Bezzina, G. (2007). Effects of retention intervals on receiver operating characteristics in artificial grammar learning. Acta Psychologica, 125(1), 37–50.
Turvey, M. T., & Weeks, R. A. (1979). Effects of proactive interference and rehearsal on the primary and secondary components of short-term retention. Quarterly Journal of Experimental Psychology, 27(1), 47–62.
Vallar, G., & Baddeley, A. D. (1982). Short-term forgetting and the articulatory loop. Quarterly Journal of Experimental Psychology Section, 34A(1), 53–60.
van der Werf, S. P., Geurts, S., & de Werd, M. M. (2016). Subjective memory ability and long-term forgetting in patients referred for neuropsychological assessment. Frontiers in Psychology, 7, 605.
Van Ormer, E. B. (1932). Sleep and Retention. Archives of. Psychology, 137, 49.
Von Cramon, D. Y., Markowitsch, H. J., & Schuri, U. (1993). The possible contribution of the septal region to memory. Neuropsychologia, 31(11), 1159–1180.
Wagenaar, W. A. (1986). My memory: A study of autobiographical memory over six years. Cognitive Psychology, 18(2), 225–252.
Walker, E. L., & Tarte, R. D. (1963). Memory storage as a function of arousal and time with homogeneous and heterogeneous lists. Journal of Verbal Learning and Verbal Behavior, 2(1), 113–119.
Wang, B., & Ren, Y. (2020). Time-dependent effects of discrete post-encoding emotions on item memory and source memory. Memory, 28(3), 417–440.
Warrington, E. K. (1982). The double dissociation of short- and long-term memory deficits. In L. S. Cermak (Ed.), Human memory and amnesia (pp. 61–76). Erlbaum.
Weaver, C. A., III., & Krug, K. S. (2004). Consolidation-like effects in flashbulb memories: Evidence from September 11, 2001. American Journal of Psychology, 117(4), 517–530.
White, K. G. (2012). Dissociation of short-term forgetting from the passage of time. Journal of Experimental Psychology: Learning, Memory, and Cognition, 38(1), 255–259.
Wilkinson, H., Holdstock, J. S., Baker, G., Herbert, A., Clague, F., & Downes, J. J. (2012). Long-term accelerated forgetting of verbal and non-verbal information in temporal lobe epilepsy. Cortex, 48(3), 317–332.
Wilson, R. S., Kaszniak, A. W., & Fox, J. H. (1981). Remote memory in senile dementia. Cortex, 17(1), 41–48.
Wixted, J. T. (2004). On Common Ground: Jost’s (1897) law of forgetting and Ribot’s (1881) law of retrograde amnesia. Psychological Review, 111(4), 864–879.
Wixted, J. T., & Ebbesen, E. B. (1991). On the form of forgetting. Psychological Science, 2(6), 409–415.
Wylie, M. (1926). Recognition of Chinese symbols. American Journal of Psychology, 37(2), 224–232.
Yarmey, A. D., Yarmey, M. J., & Yarmey, A. L. (1996). Accuracy of eyewitness identifications in showups and lineups. Law and Human Behavior, 20(4), 459–477.
Youtz, A. C. (1941). An experimental evaluation of Jost’s laws. Psychological Monographs, 238, 1–54.
Zhang, W., & Luck, S. J. (2009). Sudden death and gradual decay in visual working memory. Psychological Science, 20(4), 423–428.
Acknowledgments
We would like to thank Kyle Pettijohn, Andrea O’Rear, Jerry Fisher, and Kristina Krasich for their assistance in searching the literature and gathering memory data from the articles found.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Radvansky, G.A., Parra, D. & Doolen, A.C. Memory from nonsense syllables to novels: A survey of retention. Psychon Bull Rev (2024). https://doi.org/10.3758/s13423-024-02514-3
Accepted:
Published:
DOI: https://doi.org/10.3758/s13423-024-02514-3