Memory as Stages of Information Processing: Short-Term and Long-Term Memory Models

The idea that human thought processes are analogous to computer programs (e.g., Newell & Simon, 1972) had an important influence on the cognitive revolution in the 1950s and 1960s, shifting the dominant approach in modern psychology from behaviorism to cognitivism. Contrary to the behaviorist approach, which mainly focuses on studying observable external stimulus–response behaviors—for example, the classic studies of Pavlov (1927) and Skinner (1953, 1957)—the cognitivist approach focuses on internal mental processes that may occur between the stimulus and the response. Under this approach, an individual is considered to be a processor of information, similar to how a computer is programmed to process (e.g., code, store, and retrieve) information to produce an output. This analogy has greatly contributed to the analyses of perception, attention, and memory by providing theorists with a new framework to study the black box of human mental processing systems, in which an environmental stimulus (information) can be entered, transformed, stored, and retrieved to respond to the stimulus (Brown & Craik, 2000).

One important concept that is relevant to this information-processing model is the notion of limited capacity, which posits that humans—as information processors—can process only a limited amount of information at one time without becoming overloaded (Broadbent, 1958; Miller, 1956). Under this information-processing framework with the limited capacity perspective, memory is structured as a multi-stage process, in which people are seen as taking information into a perceptual system through various modalities, selectively attending to (parts of) the information, transforming it for use by their available cognitive abilities, storing it in memory, and later retrieving it from memory when an appropriate retrieval cue is activated (Bower, 2000).

There are three essential aspects of memory that constitute this multistage information processing: (1) memory encoding, the process under which information comes into the memory system from different sensory inputs and is changed into a visual, acoustic, or semantic form to be placed into memory; (2) memory store, a hypothetical storage in which the encoded information is held; and (3) memory retrieval, the process of recovering the previously encoded information about an item or an event, also referred to as the memory trace (Tulving, 2000). One of the most influential memory models that reflect this multistage information-processing perspective is the Atkinson & Shiffrin model (1968), also known as the multistore model. This model conceptualizes memory as a sequence of three mental stores, from sensory memory to short-term memory (STM) and long-term memory (LTM), which are presumed to differ mainly in their encoding and retention characteristics. According to the Atkinson-Shiffrin model, information from the environment is accepted and processed by sensory registers and entered into STM. The sensory memory is considered to be pre-attentive in that stimuli can be entered into the sensory stores regardless of the subject’s conscious attention to that source (Atkinson & Shiffrin, 1971). It lasts only very briefly and then is forgotten—for example, 50 milliseconds for visual sensory memory (Sperling, 1960) or four seconds for auditory sensory memory (Cowan, Lichty, & Grove, 1990)—unless it is attended to and passed on for more processing. The STM is a place or a process in which small amounts of information can be temporarily stored. STM is of limited capacity, holding at most only a few items—for example, five to nine items (Miller, 1956) or seven to nine digits (Atkinson & Shifrrin, 1971)—usually for a short time. The information (memory trace) that remains temporarily in STM is eventually displaced or lost from the store when new information enters STM and replaces the earlier items. Once the memory trace is lost from STM, it cannot be recovered from STM. During the period in which information resides in STM, however, it can be transferred into LTM, a more permanent and capacious memory store, in which the copied information is presumed to be not forgotten at all or forgotten at a much slower rate compared to the information in STM (Shiffrin & Atkinson, 1969). The transfer of information from temporary stores to more durable stores is dependent on how the information is attended to and processed: An individual has to pay attention to an environmental stimulus for successful encoding, and the more rehearsal—an overt or covert repetition of information—that the individual engages in, the greater the likelihood that the information will be copied into LTM (Atkinson & Shiffrin, 1968, 1971).

Memory as a Level of Processing: An Alternative Perspective on Multistage Memory Models

The multistage or multistore model of memory has been very influential and is still in use more than 40 years after it was developed. However, such an approach is not without its problems. One of the most frequent critiques of the model is relevant to its key premise: one memory store is distinguishable from another. While the distinctive coding principles and retention capacities of different memory stores have been well documented in the literature, a wide range of variations in such characteristics has also been reported, even when a particular memory store was examined (Brown & Craik, 2000). For example, it is known that information is usually coded acoustically in STM, whereas the main coding system in LTM appears to be semantic (Atkinson & Shiffrin, 1968; Baddeley, 1966). However, subsequent studies have showed that the main coding system in STM can be acoustic, visual, or articulatory (Eagle & Ortof, 1967; Levy, 1971; Parkinson, Parks, & Kroll, 1971). Relatedly, several scholars argued that the coding principle of STM reported in the literature can primarily be the function of the study material examined (e.g., letters or numbers with little semantic content) rather than a coding principle unique to STM (Craik & Tulving, 1975). Substantial variations have also been reported in the amount of information that can be held in STM or the durability of the memory trace (e.g., Baddeley, Hatter, Scott, & Snashall, 1970; Craik & Masani, 1969; Shiffrin & Atkinson, 1969) depending on the type of study material and task demands.

The flexible nature of STM, demonstrated by the wide variations in estimated coding and retention capacities, made the idea of distinctive memory stores less meaningful, leading to various alternative perspectives on memory. One of the alternative ideas was that memory structure should be considered in terms of the depth of processing or level of elaboration rather than separate stores with distinctive features (Craik & Lockhart, 1972). This level-of-processing approach is similar to the multistore model in that it conceptualizes memory as stages based on the information processing perspective. It differs, however, in that the stages of memory are viewed as a continuum anchored to the products of a varied degree of perceptual processing. The later stages are hypothesized to be the result of greater degrees of semantic or cognitive processing, such as pattern recognition and the extraction of meaning, whereas the preliminary stages are hypothesized to be the result of shallow processing, such as the analysis of physical or sensory features (Craik & Tulving, 1975). Thus, under this model, the durability of memory trace is a function of depth of processing, where depth is defined as “greater degrees of semantic involvement” (Craik & Lockhart, 1972, p. 675). The level-of-processing idea is well supported in the literature. Craik and Tulving (1975) conducted a series of experiments in which subjects’ memory performance was tested after they were exposed to study material (words) under various orienting tasks designed to manipulate the level of processing: Shallow or intermediate levels of encoding were prompted by questions about the word’s physical characteristics (e.g., “Is the word in capital letters?”) or rhymes (e.g., “Does the word rhyme with WEIGHT?”), while deep levels were induced by asking questions about an appropriate category or sentence frame for the word (e.g., “Is the word a type of fish?”). Findings of the series of experiments demonstrated that deeper processing was associated with higher levels of performance on the subsequent memory test (recall or recognition). Consistent with Craik and Tulving’s (1975) findings, other studies also showed that memory performance was significantly better in semantically operationalized conditions than in non-semantically defined conditions, regardless of potential influencing factors, including the type of sentence examined, the scanning time per word, the presence of rewards, or even the amount of rehearsal the items received (Craik & Watkins, 1973; Hyde & Jenkins, 1973; Rosenberg & Schiller, 1971). These findings suggest that memory performance in a retrieval condition is a function of the qualitative (semantic versus non-semantic) nature of the task and the mental activity, rather than other factors that are known to influence the capacity of a particular memory store or the transfer of information from one store to another.

Despite the contribution of the level-of-processing model that emphasizes the importance of semantic cognitive efforts in encoding and retrieving information, its somewhat general conclusion—“meaningful events are well remembered” (Craik & Tulving, 1975, p. 270)—may generate additional inquiries. For instance, does the positive association between the two indicate that any well-retrieved events or items are the products of deep processing that involves cognitive elaboration? Isn’t it possible that there are other types of well-remembered events that might have little things to do our semantic mental activity? Another influential perspective on memory provides satisfactory answers to these types of questions.

Memory as Multiple Systems: Threefold Classification of Memory

The multi-system model of memory (Tulving, 1972, 1983, 2002) was proposed to provide an overarching scheme that can coordinate the diverse perspectives and debates on memory. The fundamental assumption of the model is that there is more than one kind of memory: Memory is considered a system that consists of multiple interrelated but distinctive subsystems that are associated with different neural substrates. The three broad subsystems of memory, which differ in their behavioral or cognitive operation characteristics, are procedural memory, semantic memory, and episodic memory (Tulving, 1983, 1985a); later evidence for a fourth category of memory—priming—has also emerged (Tulving & Schacter, 1990).

Procedural memory involves the ability to acquire, retain, and utilize perceptual, cognitive, and motor skills, which enables organisms to behave appropriately in response to the environment (Anderson, 1976; Tulving, 1983). While the mode of the acquisition of knowledge, such as stimuli–response patterns and skillful performance, for procedural memory is behavior, covert cognitive activity is thought to be sufficient for the acquisition of knowledge for semantic and episodic memory (Tulving, 1985a). Semantic memory is characterized by its ability to mentally construct and manipulate the world independently of overt behavior. It involves the acquisition and use of factual knowledge regarding abstract words, symbols, concepts, and their relationships, although the context of obtaining the knowledge might be forgotten (Tulving, 1972; Tulving, 1985b; Tulving & Schacter, 1990). Episodic memory is distinguished from semantic memory by its additional ability to mentally “travel back in time” (Tulving, 1985a, p. 387), which enables people to obtain and retain knowledge about a specific event that occurred in a particular place at a particular time (Tulving, 1972, 1985b).

Each memory system also differs in the way it represents acquired information and in the type of conscious awareness that underlies its operations (Tulving, 1985a; Squire, 2004). Procedural memory is considered to be prescriptive (non-declarative), as it is expressed through performance (e.g., learned stimuli-response patterns) rather than through the conscious recollection of information about particular past experience (Dretske, 1982; Squire & Zola-Morgan, 1983). According to Tulving (1985a, 1985b), the type of awareness underlying procedural memory is anoetic (non-knowing) consciousness, which is associated with an ability to perceptually register and behaviorally respond to the aspects of the present environment. Thus, even very simple organisms can possess anoetic consciousness. Due to this anoetic, non-declarative nature, it is considered to be inappropriate or even impossible to focus on the retrieval of memory traces or to evaluate the correctness of the retrieved information when studying procedural memory (Craik, Routh, & Broadbent, 1983; Tulving, 1985a). By contrast, the way of representing acquired information in semantic or episodic memory is representational (declarative), as these memory systems provide a way to model the external world through the conscious recollection of facts and events (Squire, 2004). Specifically, semantic memory is characterized by noetic (knowing) consciousness, which allows an individual to be introspectively aware of and to operate cognitively on objects and events, even in their absence (Tulving, 1985b). Episodic memory is characterized by autonoetic (self-knowing) consciousness, which involves the additional capacity of an individual to be aware of his or her own existence in subjective time, mentally re-experiencing or extending an event from the past through the present and into the future (Tulving, 1983).

While evidence for the existence of multiple memory systems has been accumulated, another category of memory called priming—which is not procedural, semantic, or episodic memory—has been identified (Tulving & Schacter, 1990). Priming is defined as “a hypothetical process that underlies an observed improvement in identification performance” (Tulving, 2000, p. 39). Priming is distinct from procedural memory in that the identification performance is attributable to previous experience (i.e., retrieving stored information) but is similar in that it enhances perceptual skills without explicit or conscious recollection (Sherry & Schacter, 1987). In other words, when a subject uses information about previously acquired knowledge or past experience in an identification task, he or she is not aware of doing so. Priming is also similar to semantic memory in that it involves cognitive representations of the world and manifests itself cognitively rather than behaviorally, but it differs in that identifying the object could be the result of implicit retrieval (Tulving & Schacter, 1990). Two major categories of priming include perceptual and conceptual priming, which can be tested based on the clues used in object identification tasks (Tulving, 2000). In a test designed to challenge various perceptual or sensory systems, the subject is asked to identify an object (study cue) on the basis of its appearance. Prototypical tasks for the perceptual priming test include word identification, word stem completion, word fragment completion, fragmented picture identification, and sensory conditioning. In contrast, in a conceptual priming test designed to elicit responses that are meaningfully related to study cues, the subject is asked to identify an object on the basis of its meaning. Examples of this type of test include word association, category-instance generation, category verification, person or trait attributions, and object categorization (see Toth, 2000, for an exhaustive review of verbal and nonverbal tests associated with priming and procedural memory).

From Dichotomies to Multipart Systems

Broadly, evidence that demonstrates the existence of multiple memory systems can be found in two types of experimental study results: one showing that an individual with a type of impaired memory can normally perform other types of memory tasks (Cohen & Squire, 1980; Graf, Squire, & Mandler, 1984; Hamann & Squire, 1997; Jacoby & Witherspoon, 1982; Stark & Squire, 2000; Schacter & Tulving, 1982; Tulving, 1985b; Tulving & Schacter, 1990) and the other demonstrating the functional independence of different memory systems (e.g., a dissociation between the results of different memory tests that assess distinctive memory systems) in normal subjects (Graf & Mandler, 1984; Tulving, Schacter, & Stark, 1982; also see Schacter & Graf, 1986, for an excellent review of experimental research concerning the dissociations between implicit and explicit memory). For example, in describing episodic memory (autonoetic consciousness) as a distinctive category from procedural or semantic memory (anoetic or noetic consciousness), Tulving (1985b) wrote:

This young man, here referred to as N.N., suffered a closed head injury a few years ago as a result of a traffic accident. … He knows many things about the world, he is aware of this knowledge, and he can express it relatively flexibly. In this sense he is not greatly different from a normal adult. But he seems to have no capability of experiencing extended subjective time … he cannot remember any particular episodes from his life, nor can he imagine anything that he is likely to do on a subsequent occasion.

(p. 4)

Quintessential evidence on priming—which emerged as another category of memory that does not fit into the conventional dichotomies and trichotomies of memory—can also be found in studies showing that amnesic patients are able to demonstrate intact priming while performing poorly on conventional recognition tasks associated with other types of memory (e.g., Hamann & Squire, 1997; Jacoby & Witherspoon, 1982; Stark & Squire, 2000; Tulving & Schacter, 1990; Warrington & Weiskrantz, 1974). Studies with normal subjects reached similar conclusions, demonstrating the independence of priming from other types of explicit memory (Graf & Mandler, 1984; Tulving, Schacter, & Stark, 1982).

It is necessary to note that these types of findings were initially used to support a dichotomized conceptualization of memory, such as explicit and implicit memory (Graf & Schacter, 1985; Schacter & Graf, 1986) or declarative and procedural memory (Cohen & Squire, 1980). In fact, such dichotomies can be easily found from the early literature—including a distinction between explicit and implicit recognition (McDougall, 1923) and between memory with and without record (Bruner, 1969). A main postulation that underlies this perspective is that there is only one memory but there are different operating processes or different ways of assessing it. For instance, a conventional feature that distinguishes implicit (or indirect) memory from explicit (or direct) memory is simply instructions. In direct tests, subjects are asked explicitly to try to recollect a previous event or previously acquired knowledge, while in indirect tests, subjects are asked to answer general knowledge without reference to specific past experience (Bower, 2000). An issue of this approach is that even when an individual is not aware of the nature of the relation between a present identification task and previous experience due to the absence of explicit instruction (indicating the implicit nature of memory), the subject can be fully aware that he or she tries to remember past experience or previously acquired knowledge (indicating the explicit nature of memory). Given that the classification of memory would vary according to the different definitions of implicit memory—unconsciousness about the purpose of the present memory task versus unconsciousness about the retrieving behavior itself—there has been considerable disagreement in the categorization of memory into the explicit–implicit dichotomy (Toth, Reingold, & Jacoby, 1994; Tulving, 2000). Thus, under the multi-system memory model, the types of presentation style (declarative versus non-declarative) and underlying consciousness (explicit versus implicit) are considered as properties of distinctive memory systems rather than umbrella categories that theorize human memory as a unitary or dichotomized construct (Tulving, 1985a).

More fundamental issues with such unitary or dichotomized perspective can appear if the various theoretical perspectives are discussed within a neuropsychological/biological framework. Memory researchers suggest that the idea of multiple memory systems evolved because they serve distinct and functionally incompatible purposes (Sherry & Schacter, 1987; Tulving, 2002; Tulving & Schacter, 1990). For instance, certain types of memory (e.g., semantic and episodic memory) do require the involvement of prefrontal and medial temporal brain, whereas other types of memory (e.g., procedural memory associated with simple conditioning) can be obtained, stored, and retrieved in the absence of it (Gabrieli, 1998; Shimamura & Squire, 1987; Squire, 1987). The use of modern neuroimaging techniques, such as positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), made it possible to accumulate such neuroanatomical information about the structure and mechanism of human memory. Although our current knowledge about the neural substrates of different forms of memory is imperfect, the number of different brain structures identified, along with the wide variety of memory tasks examined in relation to different forms of memory, made it increasingly clear that there are no universal principles of memory (for exhaustive reviews of brain imaging studies of different forms of memory, see Cabeza & Nyberg, 2000; Passingham, 1997; Schacter & Buckner, 1998).

Limited Capacity Model of Mediated Communication Message Processing

An important theoretical framework that focuses on the information-processing mechanism underlying mediated communication effects is the limited capacity model of motivated mediated message processing (LC4MP; Lang, 2000, 2006). The development of the model relies heavily on one of the important theoretical assumptions about human cognition and memory previously reviewed: an individual as a processer of information with limited cognitive capacity (Broadbent, 1958; Kahneman, 1973; Miller, 1956; Shiffrin & Atkinson, 1969). LC4MP posits a multistage structure of information processing in which an environmental stimulus is entered into a perceptual system (encoding), selectively attended and transformed to a more durable stage of memory (storage), and retrieved when an appropriate retrieval cue is activated (retrieval). The three sub-processes occur constantly, continuously, and simultaneously in the model, but people are assumed to have only a limited amount of cognitive resources to expend on the processes (Lang, 2000). Accordingly, an important question to health communication researchers using the LC4MP framework may come down to how a mediated health message is perceived, encoded, stored, and retrieved. In this question, encoding is considered a process where an audience member is exposed to a health message (i.e., a process where information in the message is captured by a person’s sensory receptors) and parts of the registered information are selected and transformed into the person’s short-term (working) memory. Given that information in a short-term memory store is the product of our mental activity, the encoded information is not necessarily an exact reproduction or accurate reflection of the original message that the person was exposed to. Moreover, due to our limited cognitive capacity, only a fraction of the encoded information about the message is expected to be stored in long-term memory, and not all of the stored information would become readily available for retrieval (Lang, 2006). According to LC4MP, what determines the process of storage is the strength of association formed between newly entered information and previously stored information. In other words, the more links a new piece of information has to existing memory trace, the more it is likely to be stored (Klimesch, 1994; Lang, 2000). Thus, under this framework, retrieval is thought of as “the process of searching the associative memory network for a specific piece of information and reactivating it in working memory” (Lang, 2000, p. 50).

Another important area of literature that provides a critical basis for the development of LC4MP focuses on fundamental human motivation systems: approach (or appetitive) and avoidance (or aversive) motivations (for the quintessential works in the literature, see Bradley, Codispoti, Cuthbert, & Lang, 2001; Cacioppo & Berntson, 1994; Cacioppo, Gardner, & Berntson 1999). The two motivational systems are considered to be underlying forces that direct an organism’s physiological, cognitive, emotional, and behavioral reactions to environmental stimuli. In applying this framework, LC4MP conceptualizes media content as one type of complex environmental stimulus that possesses properties that have the potential to elicit diverse patterns of approach–avoidance responses. Connecting these information-processing and approach–avoidance perspectives in cognitive psychology to the tradition of understanding communication effects as a process (Berger & Chaffee, 1987; Berlo, 1960; Schramm, 1961), LC4MP significantly extended the area of research in which media, especially televised messages, are understood as environmental stimuli consisting of various features that can produce different patterns of information-processing outcomes (e.g., Lang, Geiger, Strickwerda, & Sumner, 1993; Reeves, Lang, Thorson, & Rothschild, 1989; Reeves, Newhagen, Maibach, Basil, & Kurz, 1991; Thorson, Reeves, & Schleuder, 1986). The medium, structure, content, and goal of a message are the main variables of LC4MP, and they are theorized to interact with each other and with the characteristics of the message recipients to produce communication effects (Lang, 2006).

Defining Memory for Health Communication Research: The Concept and Measures of Encoded Exposure

Although LC4MP emphasizes that an encoded message is neither an exact copy of the original message nor a direct predictor of subsequent processes, the bottom line is information in a mediated message must be encoded in order to become part of a person’s long-term memory. For example, in a campaign evaluation situation where the message is disseminated by mass media, if a person does not encode any aspects (e.g., contents or design features) of the message, it would be unreasonable to examine how the information in the message is processed and associated with changes in the person’s beliefs or behavior. The notion of encoded exposure is derived from this idea, providing communication researchers with an important tool to examine memory as one of the first stages of communication effects associated with a mediated campaign message. Encoded exposure refers to a state in a memory process, in which a piece of environmental information is transferred into a person’s long-term memory and thus becomes available for retrieval (Southwell, Barmada, Hornik, & Maklan, 2002). Thus, in a campaign evaluation situation, encoded exposure can be thought of as “minimal memory trace” of exposure to a campaign material (Southwell et al., 2002, p. 446), which is at least retrievable by an audience member through a conscious effort to recollect his or her past engagement with any particular unit of the message (Southwell, 2005). While examining other types or stages of memory previously discussed—for example, non-declarative (implicit) memory or declarative memory assessed under an encoding phase—is certainly important for communication research, the concept of encoded exposure offers a critical point at which to establish the exposure-outcome link for the purpose of campaign design and evaluation.

Based on the conceptualization of encoded exposure as a retrievable phase of the memory process, health communication researchers have tended to measure the concept through the use of self-reported questionnaire items that assess people’s ability to remember elements of the information in question at some point after they have seen it (Southwell & Langteau, 2008). The typical methods to assess such a memory type include recall and recognition tasks, which have been widely used since the pioneering work of Ebbinghaus (1964; originally published 1885) and other scholars (e.g., Tulving, 1962) on learning and memory using systematic experiments. While studies showed that both recall and recognition are valid measures of established memory and the results of the two tasks co-vary (Singh & Rothschild, 1983; Stapel, 1998; Zinkhan, Locander, & Leigh, 1986), there are theoretical and methodological reasons that the two tasks should be distinguished.

Unaided, or free recall, tasks involve the study of one’s ability to offer detail about particular content when asked an open-ended question at some point after an initial opportunity to engage the content (Southwell et al., 2002). In a situation in which the effect of a televised antidrug campaign is examined, for example, a participant might be asked to recall having seen a particular antidrug advertisement in the last three months and to provide details about the ad or be asked to describe anything about advertisements he or she has seen on television recently. In a conventional free recall task, following the presentation of a set of discrete experiences (e.g., words, pictures, actions), subjects are asked to recall as many of those items as possible in any order (Bower, 2000). One frequent finding about the performance of recall, commonly measured by the number (or proportion) of items recalled, is related to the sequential position of an item’s presentation: Items presented at the beginning (i.e., primary effects) and the end (i.e., recency effects) of the list are typically recalled earlier and more often than items presented in the middle of the list (Atkinson & Shiffrin, 1971; Tulving, 1962). Given that most free recall tests used in a typical outcome evaluation situation can be thought of as a delayed recall test, rather than a controlled recall test conducted right after the list of study materials is presented, the systematic effect of message-presentation timing might not be a major concern. Another notable finding that might be more relevant to the context of health communication research is that subjects’ recall performance tends to increase when more specific associative cues are provided (Tulving & Psotka, 1971; Tulving, 2000). This type of recall task is known as aided or cued recall, a memory task in which a relevant cue, such as a category name or a product, is provided to assist in the retrieval of the object of interest (Leigh, Zinkhan, & Swaminathan, 2006). In an evaluation research situation, for instance, an aided recall test might begin, “Have you recently seen an antidrug ad that shows a group of teens playing Russian roulette?” and ask the respondent to provide either a yes or no answer. Aided recall is regarded as an intermediate form between a recall and a recognition task; however, theoretically, it is more similar to unaided recall than recognition, as it requires the mental activity to retrieve missing elements from a target item experienced earlier as opposed to simply requiring a familiarity assessment or an awareness of having experienced that stimulus previously (Bagozzi & Silk, 1983; Mandler, 1980).

In conventional recognition memory experiments, subjects are asked to determine whether or not a test item was explicitly presented on a list studied beforehand (Bower, 2000). The test item can be thought of as a copy cue or a replica of the previously presented material, whereas in aided recall tests, only a relevant cue is presented (Tulving, 1983). For example, questionnaire items, such as “Have you ever seen or heard this antidrug ad” or “In the past month, how many times have you seen or heard this ad” when shown either that exact advertisement or edited copy of that advertisement are examples of recognition questions commonly used in campaign-evaluation research. Thus, compared to a recall test, which requires a relatively high degree of information salience and accessibility, a recognition memory test typically involves a more basic cognitive ability to respond to close-ended questions regarding one’s past engagement with specific content (Singh, Rothschild, & Churchill, 1988; Southwell et al., 2002). While a recall test may provide a more stringent measure of memory, its burdensome nature that requires more cognitive efforts may increase the possibility of under-reporting actual levels of ad exposure (Krugman, 1977; Niederdeppe, 2005). Based on these theoretical and practical rationales, communication scholars have suggested that compared to recall-based tasks, recognition-based tasks are more appropriate and more efficient indicators of encoded exposure that relies on minimal memory trace (du Plessis, 1994; Lang, 2006; Southwell et al., 2002; Southwell, 2005).

While it is important to recognize the reasons for using popular measures to understand memory in the context of communication research, it is equally imperative to raise questions about their validity. In particular, in a typical aided recall or recognition test in which subjects are asked to assess each question item as either “yes” or “no,” there is a chance that the yes–no decision is made by guessing or by responding with social desirability rather than by correctly identifying items that were presented previously. One way to address this concern is to examine the incorrect identification of items that were not previously presented (i.e., false alarm or false recognition) in addition to examining the correct identification rate (Lockhart, 2000). For instance, a correct identification rate of 50% for true items reported and a corresponding false recognition rate of 10% for bogus items would have a different validity implication compared to a correct identification rate of 50% and a corresponding false recognition rate of 50% (for exemplary work on false memories and this type of balanced test, see Roediger & McDermott, 1995). In a study that examined the validity of recognition for measuring exposure to a national antidrug campaign, Southwell and colleagues (2002) utilized this type of diagnostic test to compare the average recognition level for advertisements that actually aired with the average recognition level for bogus advertisements that did not actually air. By comparing the two rates and showing that the respondents were significantly more likely to recognize actual ads than the false ads, the study provided stronger evidence for the utility and validity of recognition tests to assess minimal memory trace concerning mediated campaign content (also see Brown, Bauman, & Padgett, 1990, for an earlier discussion about the validity of campaign exposure measures).

Another way that has been used to address false reporting in a campaign evaluation situation is to include a follow-up question that asks the respondents who replied affirmatively to the initial aided ad recall question to provide additional details about the ad. This procedure is referred to as a confirmed recall or a confirmed awareness test (Farrelly et al., 2002; Niederdeppe, 2005; Sly, Trapido, & Ray, 2002). In this memory test, only those respondents with reported ad awareness who are able to provide additional details about the ad are considered to have confirmed ad recall (Niederdeppe, 2005). The assumption behind using confirmed recall is that this task can reduce the chance of false reporting and thus is a more valid measure of ad exposure compared to an aided ad recall test. Using evaluation survey data for a statewide antismoking campaign, Niederdeppe (2005) examined the validity of the two recall measures by comparing the association between each measure and ad-specific gross ratings points (GRPs) and a campaign-targeted belief for the test of convergent and nomological validity, respectively. The main results of the study—both measures were positively associated with the cumulative GRPs and were predictors of the campaign-targeted belief—suggest that while both measures are valid indicators of ad exposure, confirmed ad recall is not a more valid measure than unaided recall. Unfortunately, the use of these types of validation procedures has been rare, even when a binary recognition or recognition test was utilized to assess the memory of mediated campaign content. While more studies are needed to confirm those findings, the results from the studies by Southwell and colleagues (2002) and Niederdeppe (2005) provide valuable evidence for the validity of popular memory measures that have been used in communication and health research.

Predicting Memory for Health Communication Research

As previously discussed, memory has been defined in various ways, and how one defines memory determines the method of studying memory—including differential types of study materials, the ways the materials are presented, and instructions and cues given to the subjects. Scholars suggest that when a subject fails to retrieve information under a particular memory test, the forgetting-remembering conclusion should be limited to the type of memory that failed (Lockhart, 2000; Mäntylä & Nilsson, 1988; Tulving, 1983). In other words, the validity of memory measures should be assessed based on what types, stages, or subsystems of memory are examined rather than an assumption that a particular memory measure is generally superior to others. Another important consensus that has emerged from the cognitive psychology of memory is that memory is constrained by a host of internal and external factors other than the task itself.

First, it is clear that one’s memory depends on the nature of the information to be processed. In traditional memory experiments, for example, the characteristics of the study material (e.g., numbers, nonsense syllables, words, or sentences with varied physical characteristics or presentation styles) affect subjects’ memory performance. As emphasized in information-processing models of memory, an environmental stimulus must be entered into one’s perceptual system to be further processed. In addition, in order for the elements of a stimulus to be attended and transformed to more durable stages of memory, some cognitive resources within the limited capacity should be allocated to them. In the case of a health campaign evaluation, campaign materials, such as a media campaign message, constitute the environmental stimuli. Based on the proposition that novel or signal stimuli that represent changes or unexpected occurrences in the environment are more likely to lead to the automatic resource allocation process, LC4MP suggests that the physical characteristics of a mediated campaign message, in combination with the message recipients’ motivation and relevance, are important factors influencing the likelihood of the media message being encoded, stored, and ultimately recollected at the retrieval study phase (Lang, 2000). This idea has greatly influenced an important area of health communication research concerning how to design a campaign message to ensure that its important aspects are encoded. In particular, a substantial body of research has reported that the structural features of a health message (e.g., cuts, edits, sound, camera techniques, and voice changes)—together with the message recipients’ individual differences—influence viewers’ information-processing patterns and subsequent outcomes (e.g., Lang, Chung, Lee, Schwartz, & Shin, 2005; Leshner, Vultee, Bolls, & Moore, 2010; Wang, Solloway, Tchernev, & Barker, 2012).

An important concept relevant to the structural features that may elicit greater message processing is message sensation value (MSV), defined as a message’s potential to produce sensory, affective, and arousal responses (Palmgreen et al., 1991). The concept was developed based on the tradition of the activation model of information exposure (AMIE), which posits that exposure to a message is determined based on the combination of the sensation-enhancing features of the message and an individual’s tendency to seek sensation (Donohew, Lorch, & Palmgreen, 1998; Harrington, Palmgreen, & Donohew, 2014). Within this framework, a message with high sensation value—a message with novel, unusual, complex, intense, arousing, graphic, or fast-paced attributes—is thought to be more effective than a message with low sensation value in drawing attention and eliciting a greater level of message processing, particularly among viewers with a high sensation-seeking tendency (Palmgreen, Donohew, Lorch, Hoyle, & Stephenson, 2001; Palmgreen, Lorch, Stephenson, Hoyle, & Donohew, 2007). There have been some concerns about the conceptualization and operationalization of MSV, as researchers have understood the concept in two ways—either as an intrinsic message feature or as a message recipient’s response to such a message feature. Later work (Harrington et al., 2003; Morgan, Palmgreen, Stephenson, Hoyle, & Lorch, 2003; Palmgreen, Stephenson, Everett, Baseheart, & Francies, 2002) further clarified these two closely connected but distinctive concepts by refining the former as a content-analytic message feature consisting of video, audio, and content dimensions (MSV as such) and the latter as perceived message sensation value (PMSV). While conflicting predictions and mixed findings regarding the role of MSV are still prevalent, scholars have suggested that, in general, messages with high sensation value produce greater message processing and more favorable evaluations compared to messages with low sensation value (e.g., Harrington, Palmgreen, & Donohew, 2014; Niederdeppe, Davis, Farrelly, & Yarsevich, 2007; Stephenson & Palmgreen, 2001; for the theoretical discussions and empirical findings regarding the distracting effect of MSV, see Kang, Cappella, & Fishbein, 2006). Notably, a recent study that conducted a qualitative synthesis and a meta-analysis of 38 studies that examined the effect of MSV reported that MSV had a positive effect across various outcomes including memory (operationalized as recall, recognition, or attention), message evaluations (e.g., perceived message effectiveness, ad liking), and attitude or behavioral intention, which supports the idea of MSV as a facilitator of message processing and persuasion (Kim, Ahn, Zhou, & Morgan, 2016).

We also know that individual differences between people affect memory performance and can modify the effects of stimulus characteristics. For instance, given that different memory stages or systems appear to draw upon different resources in the brain, the ability to remember an object changes considerably with age (Squire, 1987; Sherry & Schacter, 1987; Tulving, 2002). Neuropsychological evidence suggests that various changes in brain structure and function, such as decreased brain metabolism, reduced blood flow, and altered neurochemical systems, occur with increasing age (Anderson, Craik, & Naveh-Benjamin, 1998; Madden et al., 1997). In particular, the frontal areas known to control the networks that become activated or available during encoding and retrieval are especially sensitive to age increase (Moscovitch & Winocur, 1995; West, 1996). One perspective explaining memory decrements associated with the neurological changes involves a reduction in attention capacity. Attentional capacity, which has provided an important theoretical basis for the development of LC4MP, refers to the limited pool of processing resources available for allocation concerning any given cognitive task (Craik, Routh, & Broadbent, 1983; Kahneman, 1973). According to the attentional-capacity view, attentional resources—those needed to engage in cognitively demanding tasks—decrease as people age, and such deficiencies in processing resources negatively influence the ability to encode and retrieve information (Craik & Byrd, 1982; Salthouse, 1988). Evaluation studies have documented such age-related differences in the self-reported memory of health-campaign advertisements; for example, older adults have reported substantially lower recognition or recall rates compared to younger adults or youth (e.g., Hillsdon, Cavill, Nanchahal, Diamond, & White, 2001; Southwell et al., 2002). The age-related collapse in attentional capacity is also known to influence the accuracy of memory. While it is generally difficult to eliminate the potential of false reporting in examining human memory (Roediger & McDermott, 1995; Roediger, Watson, McDermott, & Gallo, 2001), memory studies with various contexts have consistently reported that older adults are more susceptible to false memories than are healthy young adults (e.g., Balota et al., 1999; Jacoby & Rhodes, 2006; Norman & Schacter, 1997; St. Jacques, Montgomery, & Schacter, 2015; Southwell & Langteau, 2008). Notably, while it is widely accepted that memory performance declines in older adults, the literature on memory and cognition also establishes that not all aspects of memory are equally impaired (Balota, Dolan, & Duchek, 2000; Tulving & Schacter, 1990). For instance, the research addressing episodic and semantic memory indicates that older adults, relative to younger adults, have a much larger disruption in episodic memory tasks than in semantic memory tasks (Craik & Byrd, 1982; Tulving, 1972). Similarly, it has been reported that older adults have difficulty with declarative (explicit), long-term memory tasks, such as recalling and recognizing items presented earlier, but their priming effects do not differ from those of young adults (Light, Singh, & Capps, 1986; Parkin & Streete, 1988).

Just as stimulus characteristics and individual cognitive ability can apparently influence the encoding and retrieval of information, so too can the environmental context in which people are exposed to and engage in campaign information. In the context of a mediated communication campaign, for example, it is important to provide the physical presence of the campaign information in the audience member’s immediate environment (Southwell et al., 2002). The underlying assumption here is that the information encoding process is fundamentally contingent on the environmental presence of information. For example, the widespread reliance on media weights for a particular campaign advertisement—usually indicated by the form of GRPs or average opportunity to see (AOTS)—in a campaign planning and evaluation situation is a good demonstration of the idea that environmental availability should be related to campaign exposure and remembrance. While such a physical opportunity does not necessarily guarantee any meaningful engagement with important campaign information presented in media (Clarke & Kline, 1974; Salmon, 1986; Schumann & Thorson, 1990; Speck & Elliott, 1997), it seems reasonable to assume that people should be more likely to remember ads that aired frequently compared to ads that aired less often (Niederdeppe, 2005). Studies supported this assumption showing that the environmental prevalence of a nationwide (Southwell et al., 2002) and a statewide health campaign (Niederdeppe, 2005)—measured by GRP density and cumulative GRPs, respectively—bear a strong positive relationship to various memory measures including aided and confirmed recall. These findings highlight the importance of generating widespread availability for campaign messages as a simple but crucial step in campaign planning.

Another important environmental factor that can influence a person’s remembrance of mediated campaign information involves the opportunity to interact with others. It is not a new idea that mass-mediated campaign effects do not occur in a vacuum (Katz & Lazarsfeld, 1955; McCombs & Shaw, 1972; Rogers & Storey, 1987). More recently, communication researchers (e.g., Hornik & Yanovitzky, 2003; Southwell & Yzer, 2007, 2009) have re-emphasized the importance of interpersonal diffusion through communication networks as one of the routes of communication campaign effects. They noted that by generating discussion among family, friends, or other members in the social network, campaign messages and ideas can be disseminated not only to people directly exposed to the media campaign, but also to those not directly exposed to it. Empirical studies have demonstrated that interpersonal conversation relevant to the topic of a mass-mediated campaign—as an indicator of an interpersonal network within which people are either directly or indirectly engaged with the campaign information—can facilitate or dampen its impact on a person’s beliefs and attitudes related to the topic (e.g., David, Cappella, & Fishbein, 2006; Morton & Duck, 2001). Similarly, engagement in conversation relevant to the topic of a mediated health campaign can influence a person’s memory for the campaign information. For instance, Southwell (2005) found that adolescents who had conversations about drugs with others were more likely to remember the televised anti-drug advertisement examined in the study. The information-processing perspective of memory (e.g., Atkinson & Shiffrin, 1968, 1971) can provide a theoretical explanation for this facilitating role of interpersonal communication for memory. On a basic level, for example, interpersonal networks can provide an initial opportunity for a person to be exposed to and engage with campaign content (Southwell & Yzer, 2007). On a more advanced level, conversation about the campaign content with others can constitute a mechanism for the overt or covert repetition of information (i.e., rehearsal) that can help encoded information to be copied into a more durable memory store (long-term memory) and become available for retrieval (Southwell, 2005).

What may further complicate the prediction of memory for mediated health-campaign information is a situation in which the influence of one of the factors discussed is modified by another factor. Indeed, the need for using multilevel approaches to examine the unique and combined effects of factors at different levels on individual behaviors has been extensively emphasized in public health and health communication research (Kaplan, 1996, 2004; Slater, Snyder, & Hayes, 2006). For example, in public health research, the notion of social epidemiology, often called the socio-ecological model of public health, highlights that the effects of health interventions should be understood based on multiple levels of influence, including individual, interpersonal, institutional, community, and policy levels, rather than focusing on single-level determinants (Sallis, Owen, & Fisher, 2008; Smedley & Syme, 2000). Just like various individual behaviors, memory for health campaign messages is arguably also the product of a multilevel model of predictors. For example, while studies have reported a generally positive relationship between widespread availability of mediated campaign content and remembered campaign information (Niederdeppe, 2005; Southwell et al., 2002), the effect of the opportunities for exposure may vary as a function of another environmental factor or other message-level or individual-level factors. In a multilevel modeling study that examined the relationship between encoded exposure and media weight obtained for a televised anti-drug advertisement, Southwell (2005) found that the positive relationship between GRPs and encoded exposure increased in strength when the amount of drug conversations that occurred in adolescents’ social networks increased. Another study that examined people’s memory of a television news program designed to deliver scientific information also showed that the strength of the association between the media weight and memory for the program appeared to vanish among older adults compared to younger adults (Southwell & Langteau, 2008). While the effort has already begun, future memory research in health communication invites the investigation of not only factors that describe people or messages but also variables describing social contexts as well as potential interactions between those different levels (Southwell, 2005; Southwell & Torres, 2006).

A Different Perspective: Embodied Cognition and Memory

Despite their varied theoretical details, the memory models introduced thus far commonly focus on internal mental processes hypothesized to occur between the environmental stimulus (e.g., sensory input) and the response to it (e.g., behavioral output). This view is in line with the core idea of standard cognition theories that reemerged during the mid-20th century: the mind, an abstract information processor, plays a central role in understanding and predicting human knowledge and behavior. Major accounts of social cognition assume that sensory information obtained from the external world is stored and re-described in a wide variety of abstract, amodal symbols, such as feature lists, semantic networks, schemata, and propositions (Collins & Loftus, 1975; Newell & Simon, 1972). On this view, mental operations on abstract concepts constitute a central processing system required to perform various forms of cognition, while perceptual and motor systems are merely peripheral input and output devices (Block, 1995; Fodor, 1983).

The standard theories of cognition and memory, though they remain widespread and influential, have been challenged by a radically different view that grants the sensory and motor systems a central role in shaping the mind. Rejecting the exclusive emphasis on mental representations in cognitive processing, the alternative view underscores the critical role of embodied representations that consist of patterns of possible bodily interactions with the world (Harnad, 1990; Lakoff & Johnson, 1980; Searle, 1980). This approach—collectively referred to as theories of embodied cognition—has been extensively discussed in diverse branches of cognitive science and neuroscience under various labels, including situated cognition, grounded cognition, distributed cognition, and active externalism (Barsalou, 1993, 1999; Clark, 1997; Clark & Chalmers, 1998; Glenberg, 1997; Lakoff & Johnson, 1999; Rowlands, 1999, 2003). Although an extensive summary of the debate between the embodied and disembodied perspectives is not our focus, a brief discussion is worthwhile.

The term embodied has been used in multiple ways across the literature, but researchers in embodied cognition generally hypothesize that cognitive processes and knowledge are deeply rooted in bodily actions and sensory-motor experiences (Wilson, 2002). In particular, contemporary embodied cognition theories highlight the notion of simulation in supporting this hypothesis (Barsalou, 1999, 2003; Rubin, 2006). According to Barsalou (2003), simulation refers to the process of reenacting the modality-specific information that represents the perceptual, bodily, and introspective states acquired during experiences with the environment. Importantly, this reenactment process is thought to be used in offline situations, including inference, categorization, and memory, even in the absence of any input from the original stimulus. This mechanism, from a neuropsychological perspective, can be understood as a function of bimodal neural systems that are typically characterized by their automatic activation of motor programs during perceptual or conceptual tasks without an actual interaction with an object (for discussions of visuomotor neurons such as canonical and mirror neurons, see Gallese, 2005, 2007; Garbarini & Adenzato, 2004; Rizzolatti & Craighero, 2004). Under this approach, memory consists of embodied experiences: modality-specific properties of an experience captured by the brain, such as particular bodily states associated with perception or emotion (Barsalou, 2008; Niedenthal, Barsalou, Winkielman, Krauth-Gruber, & Ric, 2005). These embodied representations are thought to be used later, when knowledge is needed for cognitive processing. For example, in Glenberg’s (1997) theory of memory, memory is conceptualized based on different modes of using embodied experiences (“meshed sets of patterns of action,” in Glenberg’s terminology [1997, p. 3]) underlying conceptualization and knowledge-gain processes. Though the details vary, other researchers similarly emphasized the centrality of embodied experiences in conceptualizing memory (Conway, 2001; Rubin, 2006). For example, episodic memory can be viewed as the simulation of multimodal information (e.g., visual, auditory, spatial, and affective aspects), which is encoded based on the original experience when a person makes a conscious effort to recollect a past event (Barsalou, 2008).

The typical evidence supporting this embodiment perspective can be found in studies demonstrating that the manipulation of bodily states can affect various types of social cognition, including attitude formation, language apprehension, and memory (e.g., Cacioppo, Priester, & Bernston, 1993; Chen & Bargh, 1999; Glenberg & Kaschak, 2002; Riskind, 1983). In particular, a number of studies have reported the effect of compatibility between motor actions and memory tasks, demonstrating that memory performance (e.g., word recognition, delayed recall, or retrieval time) is enhanced when motor behavior (e.g., body posture, arm and head movement, or facial expression) is congruent with the characteristics of the stimuli retrieved (Casasanto & Dijkstra, 2010; Riskind, 1983). For example, Riskind (1983) found that the retrieval time for pleasant memories was shorter than the retrieval time for unpleasant memories, when the bodily states adopted in the study were positive in valence (i.e., an erect posture and smiling). A similar compatibility effect found in other studies focused on the context-dependent nature of memory. For example, Dijkstra, Kaschak, and Zwaan (2007) reported shorter retrieval time for and better recall of autobiographical events when body positions during the retrieval of the past events were congruent with the body positions in the original events than when body positions were incongruent (“lying down in a recliner for the dentist memory” compared to “standing up with the hands on the hips for the dentist memory” [p. 4]). Building on this type of empirical evidence supporting the embodiment effect, health researchers may posit that the memory of health messages and the likelihood of adequately performing recommended behaviors could be improved by engaging a sensory-motor experience during a persuasion process. Although rare, this hypothesis has been tested in several studies (Gangi, Sherman, & White, 2011; Sherman, Gangi, & White, 2010). For instance, Gangi, Sherman, and White (2011) tested whether a motor manipulation could lead to an increased accessibility to the health information described in a persuasion video. The findings supported the main hypotheses, showing that participants who both touched a piece of dental floss and imagined themselves flossing while watching a video (relative to those who only imagined themselves flossing while watching the video) demonstrated better flossing skills with increased comfort and thoroughness.

All the findings described support the view that memory for a stimulus or event is stored in the perceptual and motor pathways established during online acquisition processing (Barsalou, Niedenthal, Barbey, & Ruppert, 2003; Niedenthal et al., 2005). Note that most standard memory theories assumed that mental operations underlying knowledge and memory are functionally separate from perceptual and motor systems. If memory is embodied, however, the traditional approach might lose much of its ground. Indeed, scholars who hold a strong embodiment view refuted the foundation and achievement of traditional memory research. Glenberg (1997), for example, rejected the validity of distinct memory types (e.g., short-term memory vs. long-term memory; Atkinson & Shiffrin, 1968; Baddeley, 1966, 1986) or memory systems (e.g., semantic vs. episodic memory systems; Tulving, 1972, 1983, 1985a), which is developed based on the idea that each stage or system has different encoding, retrieval, or enhancement features. In Glenberg’s view, such distinction simply reflects differences in the methods of assessment or the nature of the environmental information, and the diversity in theorization of memory arises from inappropriate use of tools relying on the manipulation of meaningless symbols (1997).

Certainly, such a view is not without some criticism. One typical criticism of the embodiment approach is that offline cognitive processing that is not embodied exists, such as planning for the future or mentally representing situations a person has never experienced purely based on linguistic information (Wilson, 2002). Contrary to earlier embodied cognition theories focusing on the role of actual bodily states in shaping cognition, modern embodied cognition theorists emphasize that cognition is grounded in multiple ways and, thus, does not always necessitate bodily states or simulations (Barsalou, 1999, 2003; Niedenthal et al., 2005). Despite this clarification, the widely cited study results that demonstrate the impact of bodily manipulations on attitude and memory (e.g., Cacioppo, Priester, & Bernston, 1993; Riskind, 1983) might not be sufficient to reject the disembodied view. One main reason for this argument is that, from the theorists of the disembodied perspective, the motor system becomes activated subsequent to, rather than prior to, the abstract mental representations (Mahon & Caramazza, 2008). Although evidence established in experimental settings already exists (e.g., Barsalou et al., 2003; Glenberg & Kaschak 2002), assessing the causal roles of simulations and embodiments requires much further research (Barsalou, 2008). In addition, according to the embodied cognition view, people build and simulate embodiments based on multiple modality-specific sources, including their experiential action, emotional states, and cognitive states. Although modern embodiment theorists elaborate that the object and nature of simulations could vary according to task goals (Barsalou, 2003; Niedenthal et al., 2005), the flexibility in using multimodal sources in simulations could generate further questions. For instance, if people (consciously or unconsciously) rely on the emotional and cognitive aspects of the experience and activate associated bodily states during the reenactment process, then can this be considered evidence that only supports the centrality of the sensory-motor system in shaping cognition? Another primary argument against the exclusive acceptance of the embodied cognition approach is that the essential evidence supporting embodiment effects on memory might actually provide an endorsement of existing perspectives from amodal memory theories (Crowder & Wenk, 1997). For example, the context-dependent nature of memory (aka a compatibility effect), which is demonstrated in Riskind (1983) and other studies, can be explained by the encoding specificity principle, which posits that retrieval can be enhanced when experimental conditions or cues are similar to the condition in which the study items were originally encoded (Thomson & Tulving, 1970; Tulving, 1983). That is, the embodiment view and the encoding specificity principle commonly emphasize that the sensory-motor environment in which the memory of an object or event is stored should be considered an integral part of memory and cognitive processing (Dijkstra, Kaschak, & Zwaan, 2007; Smith & Vela, 2001). Despite their fundamental differences, this overlap could be a good example of demonstrating the possibility of reconciling the two important frames for understanding human cognition and memory in a constructive way.

Once again, discussing whether and how the new paradigm will be fully accepted or integrated with traditional paradigms is beyond our scope. However, this increasingly important embodiment approach, as well as the classic memory-cognition models, offers exciting research directions to the study of health and risk behaviors that inherently possess substantial motor components.