Draft of: Merikle, P. M., & Reingold, E. M. (1991). Comparing direct (explicit) and indirect (implicit) measures to study unconscious memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 224-233.
Comparing Direct (Explicit) and Indirect (Implicit) Measures to Study Unconscious Memory
Philip M. Merikle and Eyal M. Reingold
Comparisons of the relative sensitivity of direct and indirect tasks can provide definitive evidence for unconscious memory when the direct and indirect tasks are matched on all characteristics except instructions. To demonstrate unconscious memory in normal adults, subjects first viewed pairs of words and named one cued word in each pair. During the subsequent assessment of memory, new words and either previously cued words (Experiment 1) or previously uncued words (Experiments 2A and 2B) were presented against a background mask. The subjects judged whether a word was old or new (direct task) or whether the contrast between a word and the mask was high or low (indirect task). For cued words, the direct task was more sensitive than the indirect task. However, for uncued words, the indirect task was initially more sensitive than the direct task, even though the direct task exhibited hypermnesia so that it became more sensitive than the indirect task across trial blocks. The greater initial sensitivity of the indirect task implicates unconscious processes underlying memory for the uncued words. These results indicate that unconscious processes in normal adults can be revealed through comparisons of comparable direct and indirect measures.
Comparing Direct (Explicit) and Indirect (Implicit) Measures to Study Unconscious Memory
Questions regarding the relationship between consciousness and memory are attracting increasing attention (e.g., Jacoby, Woloshyn & Kelley, 1989; Kelley & Jacoby, 1990; Kihlstrom, 1987; Schacter, 1989). This increased interest begins to rectify what Tulving (1985) described as the "benign neglect of consciousness" (p. 1) in the study of human memory. However, in spite of the growing number of studies directed at demonstrating memory without awareness, definitive studies distinguishing conscious from unconscious memories in normal adults are rare. As noted by Schacter (1989), even though studies of amnesic patients provide compelling evidence for memory in the absence of conscious recollection, the evidence from studies with normal adults is not nearly as clear cut. A major reason it has been so difficult to distinguish conscious and unconscious memories in normal adults is that there is no general agreement as to what constitutes an adequate measure of conscious experience (Erdelyi, 1986; Reingold & Merikle, 1990). For this reason, we previously proposed an alternative approach to the study of memory and consciousness that is based on comparisons of the relative sensitivity of different types of tasks (Reingold & Merikle, 1988; 1990). In the present paper, we demonstrate that this alternative approach provides compelling evidence for memory without conscious awareness in normal adults.
A basic distinction among tasks used to study memory and consciousness concerns whether a given task provides a direct or an indirect measure of memory (Johnson & Hasher, 1987; Reingold & Merikle, 1990; Richardson-Klavehn & Bjork, 1988). Tasks in which subjects are explicitly instructed to discriminate old or previously presented stimuli from new stimuli that have not been presented within the experimental context are defined as direct measures of memory. In contrast, if task instructions do not make any reference to the old/new discrimination, then such tasks are defined as indirect measures of memory. Examples of direct memory tasks are stimulus recognition and recall. In these tasks, subjects are explicitly instructed either to make an old/new discrimination or to recall only stimuli that were presented previously within the experimental context. Examples of indirect tasks are word-stem completion (e.g., Graf & Mandler, 1984) and perceptual identification (e.g., Jacoby & Dallas, 1981). These tasks provide indirect measures of memory in the sense that no explicit reference is made in the task instructions to the distinction between old and new stimuli.
Any classification of memory tasks based on the distinction between direct and indirect measures is identical to a classification of tasks based on distinction between explicit and implicit measures of memory (e.g., Graf & Schacter, 1985; Schacter, 1987). We prefer the distinction between direct and indirect measures, however, because the explicit/implicit terminology is potentially very misleading (see Dunn & Kirsner, 1989; Reingold & Merikle, 1990; Richardson-Klavehn & Bjork, 1988). Specifically, the terms implicit and explicit have been used to classify both memory processes and memory tasks. This dual meaning of the terms has blurred the distinction between theoretical constructs and empirical measures. On the other hand, the direct/indirect distinction is based solely on task instructions and does not require any speculation as to either underlying memory processes or possible phenomenal experiences (e.g., awareness, intentionality) correlated with the different types of tasks. It is precisely for this reason that we and others ( Hintzman, 1990; Reingold & Merikle, 1990; Richardson-Klavehn & Bjork, 1988) have suggested that the terms direct and indirect rather than the terms explicit and implicit are better labels for these two classes of memory tasks.
To date, many dissociations between direct and indirect measures of memory have been reported (see Richardson-Klavehn & Bjork, 1988; Schacter, 1987; for reviews). However, there is no general consensus as to the best interpretation of many of these reported dissociations (cf. Dunn & Kirsner, 1989). This is the case because direct and indirect tasks are often quite dissimilar. Thus, the observed dissociations may simply reflect task differences. In general, dissociations between measures are usually more informative the more similar the tasks, just as associations between measures are usually more impressive and informative the more dissimilar the tasks. For this reason, a useful strategy when comparing direct and indirect tasks is to match the tasks on all characteristics except task instructions. This strategy constrains the number of possible interpretations of any observed dissociations.
Reingold and Merikle (1988) suggested a number of methodological guidelines for evaluating the comparability of direct and indirect tasks. For present purposes, we describe the three most common differences between the direct and indirect tasks used in studies of memory that have revealed dissociations.
1) Retrieval or test cues: Direct memory tasks such as recognition and recall, and indirect memory measures such as word-stem completion and perceptual identification often present subjects with very different test or retrieval cues (e.g., TRA__ in a stem completion task versus TRAVEL in a recognition task). Thus, in any study in which the retrieval or test cues are not physically identical across tasks, dissociations between measures may simply reflect the fact that different test cues were used with each task (see Schacter, Bowers & Booker, 1989 for a related discussion).
2) Memory sensitivity versus response bias: Performance levels on direct and indirect measures may also be differentially affected by response bias. For example, while a recognition task permits independent assessment of memory sensitivity and response bias, sensitivity and bias cannot be distinguished in indirect tasks such as perceptual identification and word-stem completion. Often, differences in bias across tasks are accompanied by differences in retrieval cues. However, even if identical retrieval cues are used, the influence of response bias may not be equivalent across direct and indirect measures. For example, Graf and Mandler (1984) reported a dissociation between a direct cued recall task and an indirect stem completion task under conditions in which identical test cues were used across tasks. However, while subjects were required to provide a response to every cue in the stem completion task, they were not required to do so in the cued recall task. Thus, the reported dissociation may have simply reflected a difference in response bias across the two tasks.
3) Response metric: Memory measures such as stimulus recognition, recall, word-stem completion, and perceptual identification represent very different response metrics. The functions relating performance levels across tasks are unknown. Thus, observed dissociations between direct and indirect memory tasks may simply reflect differences in measurement scales and not reflect differences in underlying processes.
Matching direct and indirect memory tasks on these three task characteristics constrains the number of possible interpretations of any observed dissociation between tasks. Furthermore, Reingold and Merikle (1988) suggested that comparisons of the relative sensitivity of comparable direct and indirect measures have the potential to provide definitive evidence for unconscious perception and memory. This alternative approach to the study of unconscious processes is based on the following minimal a priori working assumption: "The sensitivity of a direct discrimination is assumed to be greater than or equal to the sensitivity of a comparable indirect discrimination to conscious, task relevant information." (p. 556, Reingold & Merikle, 1988). In terms of the present context, the rationale underlying this assumption is that conscious information relevant to an old/new discrimination, if it exists, should be used equally or more efficiently when subjects are instructed to make an old/new discrimination (i.e., a direct task) than when subjects are not so instructed (i.e., an indirect task). Conversely, it is difficult to imagine circumstances in which conscious information relevant to an old/new discrimination would enhance performance more when subjects were not instructed to make the old/new discrimination, than when subjects were explicitly instructed to make the old/new discrimination.
Although any a priori assumption can be criticized, the working assumption proposed by Reingold and Merikle (1988) has two distinct advantages relative to the assumptions underlying many other approaches to the study of unconscious processes. First, the assumption is stated explicitly and thus open to evaluation. Second, the assumption is much more minimal than many assumptions which relate direct and indirect tasks to conscious and unconscious processes. In fact, with the proposed working assumption, much of the controversy over the measurement of awareness is bypassed because no constraints are placed on how direct and indirect tasks may be related to conscious and unconscious processes. Rather, given the proposed assumption, both direct and indirect tasks may reflect conscious, unconscious or both conscious and unconscious processes. Thus, even in the absence of a valid measure of awareness (see Erdelyi, 1986, Reingold & Merikle, 1990; Schacter, 1987), the proposed assumption provides a basis for studies of unconscious processes.
The important consequence of this very minimal working assumption is that unconscious memory processes are implicated whenever an indirect measure shows greater sensitivity than a comparable direct measure to an old/new discrimination. This is the case because the assumption rules out the possibility that superior performance on the indirect task is attributable to conscious task relevant information. Therefore, by default, whenever an indirect measure indicates greater sensitivity than a comparable direct measure, it must reflect a greater sensitivity of the indirect measure to unconscious, task relevant information. It is important to emphasize that this interpretation is warranted only if the direct and the indirect measures are truly comparable except for the presence or absence of a reference to the discrimination of interest in the instructions given to subjects. Otherwise, the greater sensitivity of an indirect measure may reflect a methodological artifact rather than an unconscious process.
Empirical support for the possible value of comparing the relative sensitivity of comparable direct and indirect measures in the study of unconscious processes comes from studies of unconscious memory. The results of a number of studies indicate that, at least under certain conditions, indirect measures do in fact exhibit greater sensitivity to an old/new discrimination than comparable direct measures. In one important study, Kunst-Wilson and Zajonc (1980) initially showed subjects ten irregular geometric shapes, with each shape being presented five times for a very brief, 1-msec duration. Following these initial exposures, the subjects were shown 10 pairs of shapes, one old and one new, and they were instructed either to indicate which member of each pair had been presented previously (a direct measure) or to choose the shape they preferred (an indirect measure). With the indirect measure based on preference, the subjects chose the old stimulus in 60% of the pairs. However, with the direct recognition test, old stimuli were selected in only 48% of the pairs, which approximated the chance level of performance. These basic results have been replicated by other investigators (e.g., Bonnano & Stillings, 1986; Seamon, Marsh, & Brody, 1984). In addition, Mandler, Nakamura and Van Zandt (1987) used a similar procedure and found that a variety of tasks requiring indirect discriminations (preference, brightness, darkness) were more sensitive to the old/new discrimination than comparable direct measures of stimulus recognition. According to the logic underlying comparisons between comparable direct and indirect measures, these findings constitute strong evidence for unconscious memory. However, with one exception (Bornstein, Leone, & Galley, 1987), the pattern of results originally reported by Kunst-Wilson and Zajonc (1980) has only been demonstrated with a very specific stimulus set, namely irregular polygons selected from those of Vanderplas and Garvin (1959). Clearly, to establish the generality of these findings, it is necessary to compare the relative sensitivity of comparable direct and indirect measures of memory across other experimental manipulations and sets of stimulus materials.
A study reported by Eich (1984) represents a potentially useful experimental approach. In this study, a version of the dichotic listening paradigm was used. During the study phase, subjects shadowed a prose passage presented on one auditory channel, and concurrently, on the non-shadowed channel, pairs of words were presented. Each word pair consisted of a homophone (e.g., FARE/FAIR) and a context word (e.g., TAXI) which biased interpretation of the homophone toward the less common meaning . In the test phase of the study, subjects were presented auditorily with old and new homophones and they were required either to make an old/new recognition judgement (direct instructions) or to spell the homophone (indirect instructions). The results indicated that the recognition task was completely insensitive to the old/new dimension. However, the homophone spelling task was sensitive to the old/new dimension in that the subjects were biased toward the less common spellings of the homophones.
Eich's study is completely consistent with the approach advocated by Reingold and Merikle (1988) for comparing the relative sensitivity of direct and indirect measures of memory to establish unconscious processes. The direct recognition task and the indirect spelling bias task were comparable except for the instructions given to the subjects. Thus, according to the logic of Reingold and Merikle's approach, the dissociation between recognition memory and spelling bias provides evidence that the non-shadowed words were remembered unconsciously. It is important to emphasize that this conclusion does not require any assumption concerning whether or not the non-shadowed words were truly unattended at the time of encoding. Eich (1984) assumed that the non-shadowed words were perceived without attention, but this difficult, if not impossible, to prove assumption concerning initial encoding is not required to demonstrate unconscious memory. Rather, according to the logic underlying comparisons of comparable direct and indirect measures, Eich's results demonstrate unconscious memory because the indirect homophone spelling task exhibited greater sensitivity than the direct recognition task at the time of retrieval.
This analysis of Eich's results in terms of the logic proposed by Reingold and Merikle (1988) suggests that selective attention may represent a useful experimental manipulation for documenting unconscious memory. Accordingly, in the present experiments, we explored the effects that allocation of visual attention may have on the relative sensitivity of a direct measure of old/new recognition memory and a comparable indirect measure. Each experiment consisted of two phases. During the study phase, subjects were presented with pairs of words and they were required to name one cued word in each pair. During the test phase, new words and either the previously cued words (Experiment 1) or the previously uncued words (Experiments 2A and 2B) were presented against a background mask. The subjects were required to judge whether a word was old or new (direct task) or whether the contrast between a word and the mask was high or low (indirect task).
The primary purpose of Experiment 1 was to determine if the indirect contrast measure was a sensitive measure of memory. To evaluate this measure under optimal conditions, memory for the previously cued or attended words was assessed. During the test phase of the experiment, the cued words from the study phase were presented against a background mask which degraded the visual quality of the letters. For each word, the subjects were required to judge whether the contrast between the word and the mask was high or low. The subjects were told that high contrast meant that the word and the background were relatively distinct from each other, while low contrast meant that the word and the background tended to blend together.
The rationale underlying the contrast measure as a possible sensitive indirect measure of memory is based on recent findings indicating that memory influences subjective perceptual experience. Jacoby, Allan, Collins and Larwill (1988) reported that background white noise was perceived as less intense when the noise was judged in the context of previously heard sentences or words than when the noise was judged in the context of new materials that had not been presented previously to the subjects. Another way to describe these results is to state that the perceived contrast between sentences or words and the background noise was greater when the subjects had previously heard the sentences or words in the context of the experiment. If this change in subjective perceptual experience due to memory for the words or sentences reported by Jacoby et al. reflects a fairly general phenomenon, then it is reasonable to expect that memory for the cued words in the present experiment should influence subjects' subjective experience of the perceptual contrast between the words and the background mask.
Finally, possible changes in memory over time were evaluated. Erdelyi (1986) has discussed previously how different measures of memory may be differentially affected by either forgetting or hypermnesia (i.e., an improvement in memory over time). Consequently, the relative sensitivity of comparable direct and indirect measures may change over time following initial presentation. For this reason, sensitivity across trial blocks was assessed for both the contrast measure and a comparable measure of recognition memory.
Subjects. The 96 subjects were undergraduate students at the University of Waterloo. Each subject had normal or corrected to normal vision and was paid $5 for participation. Forty-eight subjects were assigned on the basis of an alternating sequence to the direct and indirect tasks when they arrived at the laboratory to participate in the experiment.
Apparatus. All stimulus materials were displayed on a Zenith flat tension color monitor (Model ZCM-1490) that was driven by a Zenith microcomputer equipped with an 80286 processor. Subjects viewed the monitor while seated with their head position constrained by a chin rest adjusted to a comfortable height. The viewing distance was approximately 105 cm. During the study phase of the experiment, a microphone connected to a voice key was clipped to the chin rest. In addition, during both the study and the test phases of the experiment, a button box with three buttons was placed directly in front of the subjects. The center button was used to initiate each trial during both phases of the experiment, and the buttons on the left and right were used to indicate either "old/new" or "high/low contrast" decisions during the test phase of the experiment.
Procedure. Each subject was tested individually in a single session that lasted approximately 25 min. At the beginning of the session, subjects were informed that the experiment involved two different tests of word perception. The instructions implied that the two tests were unrelated, and no information concerning the test phase was given until after the study phase was completed.
Prior to the beginning of the study trials, all subjects were told that they would see two words on each trial and that their task was simply to read aloud the one word with arrows pointing towards it. The instructions emphasized that each cued word should be read into the microphone clipped to the chin rest and that both naming speed and naming accuracy were important. In addition, the voice key was demonstrated by showing subjects that it made a clearly audible click each time they spoke a word into the microphone.
The sequence of events on each of the 60 study trials began with the subjects viewing a fixation dot located in the center of the visual field. Subjects initiated a trial by pressing the center button on the button box, which led to the fixation dot being replaced by a 200-msec blank field. Following the blank field, a target display containing a pair of words was presented for approximately 500 msec. Each display also contained an arrow located at each end of one word. These arrows cued the subjects as to which word they should read aloud. Following each target display, the fixation dot reappeared after a brief untimed interval to indicate that the computer was ready for the next trial.
The test phase of the experiment began immediately after completion of the study trials. The microphone clipped to the chin rest was removed, and the subjects were told that they would now see a series of 96 words, with each word being presented against a mottled background. To illustrate the display conditions, a practice display was then presented. All subjects were instructed that their task on each trial was to read the word aloud and then to make a decision concerning the word. The subjects assigned to the direct task group were told to decide whether the word was "old" or "new." An "old" word was defined explicitly as one of the cued words presented during the study trials, and a "new" word was defined as any word not presented during the study trials. For the subjects assigned to the indirect task group, the required decision concerned whether the word was presented under "high" or "low" contrast conditions. High contrast conditions were defined as conditions in which the word appeared to "stand out" against the background, whereas low contrast conditions were defined as conditions in which the word appeared to "blend" into the background.
Before beginning the 96 test trials, all subjects in both groups were given eight practice trials. The sequence of events on each practice trial began with the presentation of the fixation dot. Once the start button was pressed, the fixation dot was replaced by a display containing a word presented against a background mask. Each word was presented until a decision was made. For the subjects assigned to the direct task group, the words "old" and "new" were presented four times each across the eight practice trials, and the subjects pressed the right button to indicate an "old" decision and the left button to indicate a "new' decision. For the subjects assigned to the indirect task group, the word "word" was presented on all eight practice trials, and the subjects in this group indicated their decisions on each trial by pressing the right button if the word had "low contrast" and the left button if the word had "high contrast" with the background. To ensure that the subjects were not confused as to which button to press during the test trials, the buttons on the right and left of the button box were labelled either "old" and "new" or "low contrast" and "high contrast," depending on the group to which a subject was assigned.
The sequence of events on each of the 96 test trials was the same as the event sequence on the practice trials. In addition, in each block of 16 trials, equal numbers of "old," "new," low contrast, high contrast, five-letter, and six-letter were presented. The subjects in both groups were told that the required decision was sometimes very difficult to make. It was emphasized that there were equal numbers of old and new words (direct task) or high and low contrast words (indirect task) and that for this reason, each of the two responses should be used approximately equally often. No feedback as to the correctness of responses was given during the test trials.
Stimulus Materials. A pool of words containing 384 five-letter and 384 six-letter nouns was selected from the Kucera and Francis (1967) norms. The frequency of the selected words ranged from 2 to 15/million.
For the study phase of the experiment, 60 pairs of words were selected randomly from the word pool. Thirty of these pairs consisted of two five-letter words, while the remaining 30 pairs consisted of two six-letter words. A different set of 60 pairs was selected for each subject. Each pair of words was presented with one word above and one word below a centrally located fixation dot. The vertical distance between the words was approximately 16 mm (0.9o).1 The words were presented in white, lower-case letters in EGA graphics mode against a black background, and the font was Borland Turbo C default font, double-sized. Each letter in each word when presented on the monitor was approximately 5 mm (0.3o) wide and 5 mm to 7 mm (0.4o) high, depending on the particular letter. The length of each word, considering the spacing between letters, was approximately 30 mm (1.6o) for the five letter words and 36 mm (2.0o) for the six letter words.
The order of presentation for the 60 pairs of words was randomized for each subject. The first six and the final six pairs presented to each subject were considered filler pairs and no words from these pairs were used in the test phase of the experiment. For the remaining 48 pairs, the cued word in each pair was presented equally often above and below the fixation dot and equal numbers of cued words contained five and six letters. In addition, the arrows used to cue subjects as to which words they should read aloud were approximately 7 mm (0.4o) long and located approximately 3 mm (0.2o) from each end of the cued words.
For the test phase of the experiment, 48 "new" words were selected randomly from the word pool. A different set of "new" words was selected for each subject and each set consisted of 24 five-letter and 24 six-letter words. The "new" words were combined with 48 "old" words from the study phase to form a single set of 96 words. For each subject, a different random order of presentation was determined for this set of words, within the constraint that in each block of 16 trials, there were equal numbers of "old," "new," five-letter, and six-letter words.
All words used in the test phase of the experiment were presented against a rectangular background mask. This mask was presented at the center of the monitor and measured approximately 45 mm (2.5o) horizontal by 10 mm (0.6o) vertical. The density of the mask was varied at two levels; either 50% or 60% of the pixels within the rectangular area were selected randomly and displayed whenever a word was presented. Prior to the beginning of the experiment, four different 50% density masks and four different 60% density masks were determined for each subject. Each of these eight masks was used with one "old" and one "new" word in each block of 16 trials. Within this constraint, the selection of a mask on each trial was random.
Results and Discussion
For each subject, the 96 test trials were divided into three blocks of 32 trials, and proportions of hits and false alarms in each of block of trials were computed. For the contrast task, a hit was defined as an old word judged to have high contrast against the background mask, and a false alarm was defined as a new word judged to have high contrast against the mask. The mean proportions of hits and false alarms for each task are shown in Table 1.
In draft, Table 1 would appear here
To obtain a measure of memory for each task that was independent of response bias, the sensitivity of each task to the old/new dimension was expressed in terms of A'.2 This nonparametric measure of sensitivity can vary from 0.00 to 1.00, with A' = 0.50 indicating a complete absence of sensitivity (see Snodgrass & Corwin, 1988, for a detailed discussion of the A' measure). Figure 1 shows the mean A' values for each task across the three trial blocks.
In draft, Figure 1 would appear here
Inspection of Figure 1 indicates that both the recognition and the contrast tasks were sensitive measures of the old/new discrimination and that the sensitivity of both tasks decreased slightly across trial blocks. A 2 X 3 analysis of variance with task and trial block as factors revealed that the recognition task was superior to the contrast task, F (1,94) = 369.25, p < .001, but that the small decrease in performance across trial blocks was neither significant, F (2,188) = 1.56, nor did it interact with task, F (2,188) < 1.
To evaluate the sensitivity of the contrast task, performance in each trial block was compared with the chance level of performance. These comparisons revealed that the sensitivity of the contrast task significantly exceeded the chance level of performance both in Block 1 (.57), t (47) = 3.68, p < .001, and in Block 2 (.55), t (47) = 2.53, p < .025. However, the sensitivity of the contrast task in Block 3 (.53) only approached significance, t (47) = 1.85, p < .10.
The results of Experiment 1 demonstrate that the contrast task is a sensitive indirect measure of memory, at least during initial blocks of trials. Although the contrast measure was less sensitive than the direct recognition measure, the results indicate that judgements of perceptual contrast between a word and the background mask are influenced by memory for previously attended or cued words. Given these results, the contrast measure may also be a sensitive indirect measure of memory for the uncued words.
Experiments 2A and 2B
The major goal of Experiments 2A and 2B was to compare the relative sensitivity of the contrast and recognition tasks as measures of memory for the uncued words. According to the logic underlying comparisons of comparable direct and indirect measures, unconscious memory would be demonstrated if the indirect contrast measure is more sensitive than the direct recognition measure.
The two experiments were quite similar in that Experiment 2B was basically a replication of Experiment 2A with a) a different set of stimulus materials and b) a different density mask to degrade the visibility of the words in the contrast and recognition tasks. Given these methodological differences, successful replication across experiments would increase the generality of any findings demonstrating unconscious memory for the uncued words.
Experiment 2A. The general methodology was the same as used in Experiment 1 except that memory for the uncued rather than the cued words was tested. In addition, as a consequence of this changed procedure, the subjects assigned to the direct task group were instructed explicitly that an "old" word was one of the uncued words presented during the study trials and that a "new" word was any word not presented during the study trials.
Experiment 2B. This experiment was very similar to Experiment 2A. However, two aspects of the methodology were different. First, the word pool contained 160 five-letter and 160 six-letter low frequency nouns with a Kucera and Francis (1967) frequency of one/million. Second, the density of the eight background masks used during the test trials was always 45%. Even though mask density was constant, each mask had a different appearance, given the different random arrangements of pixels, and each word had a different appearance against each mask, depending upon how the letters in the word overlapped the locations of the selected pixels in the mask. In all other respects, the general methodology was the same in Experiments 2A and 2B.3
Subjects. The 96 subjects in each experiment were undergraduate students at the University of Waterloo. Each subject was paid $5 for participation and had normal or corrected to normal vision. In both experiments, 48 subjects were assigned on the basis of an alternating sequence to the direct and indirect groups when they arrived at the laboratory to participate in the experiment.
Results and Discussion
The data in Experiments 2A and 2B were analyzed in the same manner as the data in Experiment 1. Briefly, the 96 test trials for each subject were divided into three blocks of 32 trials, and proportions of hits and false alarms in each trial block were computed. These mean proportions are shown in Table 1. In addition, A' values were also computed, and the mean values for each experiment are shown in Figure 2.
In draft, Figure 2 would appear here
Inspection of Figure 2 indicates that the same basic pattern of results was found in Experiments 2A and 2B. In both experiments, the indirect contrast measure was more sensitive than the direct recognition measure in the first two block of trials. This greater sensitivity of the indirect measure reflects the influence of unconscious memory. On the other hand, no evidence for unconscious memory was found in the third block of trials in either experiment, as the results of both experiments indicate that the recognition measure was more sensitive than the contrast measure.
The results from Experiments 2A and 2B were evaluated by an overall 2 X 2 X 3 analysis of variance, with experiment (Experiment 2A vs. Experiment 2B), task (Recognition vs. Contrast), and trial block (Block 1, Block 2 and Block 3) as factors. Neither the main effect of experiment nor any interactions involving experiment as a factor were significant sources of variance, all Fs < 1. Thus, the results of the overall analysis are consistent with the conclusion that Experiment 2B was a successful replication of Experiment 2A.
The only significant effect in the overall analysis was the interaction between task and trial block, F (2,376) = 7.65, p < .001. As suggested by Figure 3, which shows the combined result from both experiments, this interaction reflects the greater sensitivity of the contrast relative to the recognition task in Block 1, t (190) = 2.87, p < .01, and in Block 2, t (190) = 2.17, p < .05, coupled with the greater sensitivity of the recognition relative to the contrast task in Block 3, t (190) = 2.44, p < .05.
In draft, Figure 3 would appear here
Given the logic underlying the experiments, the most important results are those obtained in Blocks 1 and 2. Even though the absolute sensitivity of the indirect contrast measure was relatively modest, it exhibited greater sensitivity than the direct recognition measure in these blocks of trials. Furthermore, additional analyses of performance indicated that the mean A' for the contrast task was significantly greater than chance in both Block 1 (.539) and Block 2 (.536), both ts (95) > 2.57, p < .025, but that the mean A' for the recognition task did not differ from chance in either Block 1 (.485) or Block 2 (.491). This greater sensitivity of the indirect contrast measure relative to the direct recognition measure provides an unequivocal demonstration of unconscious memory for the uncued words.
An unexpected finding was the increased sensitivity of the recognition task across trial blocks. This hypermnesia was statistically significant, F (2,190) = 4.54, p < .05, as performance on the recognition task, which did not differ from chance in the first two blocks of trials, was significantly greater than chance (.542) in the third block of trials, t (95) = 3.00, p < .01. This improvement in recognition memory differs qualitatively from the significant decrease across trial blocks in the sensitivity of the contrast measure, F (2,190) = 3.19, p < .05. In addition, this hypermnesia in recognition memory for the uncued words also differs qualitatively from the small decrease in recognition memory for the cued words across trial blocks found in Experiment 1.
Even though hypermnesia in recognition memory was unexpected, it is not inconsistent with previous findings. For example, Mandler et al. (1987) conducted an experiment that was very similar to the original Kunst-Wilson and Zajonc (1980) study. They found that recognition memory for geometric forms increased from 46.7% to 52.5% accuracy across two blocks of 10 trials. In addition, a measure of preference indicated that the percentage of old forms selected decreased from 61.7% in the first block trials to 56.7% in the second block of trials. These changes in performance across trial blocks for the direct recognition task and the indirect preference task parallel the changes found in Experiments 2A and 2B for the recognition and contrast tasks.
Taken together, both the results reported by Mandler et al. and the present results suggest that strategies for doing the recognition task may change across trial blocks. During the early trial blocks, it is reasonable to assume that subjects follow the instructions and attempt to decide whether each word is old or new. However, if no conscious information relevant to the old/new decision is available, then subjects may base their decisions on conscious but irrelevant aspects of the task (e.g., the number of letters in a word). Such a strategy would lead to chance performance. Over time, subjects may abandon these ineffective strategies based on consciously available information. As an alternative, many subjects may adopt a more passive strategy and simply base their decisions on general impressions. If this occurred, then the recognition task would in fact become an indirect task. In effect, what is being suggested is that the hypermnesia reflects a release from the interference produced during early trials when irrelevant conscious information is used to guide performance. Once these ineffective strategies are abandoned, the recognition task becomes a sensitive indicator of unconscious influences. This speculative interpretation provides a reasonable account of the hypermnesia found in Experiments 2A and 2B, as well as by Mandler et al. (1987), and it suggests that strategy changes over trial blocks may be an important determinant of the relative sensitivity of different tasks.
According to the logic underlying the approach advocated by Reingold and Merikle (1988), unconscious memory is demonstrated whenever an indirect measure exhibits greater sensitivity than a comparable direct measure of memory. Given this logic, the greater sensitivity of the indirect contrast measure relative to the direct recognition measure in Experiments 2A and 2B necessarily implicates unconscious processes. Likewise, previous findings demonstrating greater sensitivity for indirect than for comparable direct measures of memory (e.g., Eich, 1984; Kunst-Wilson & Zajonc, 1980) also provide strong support for unconscious memory. Thus, both the present results and previous findings provide convincing demonstrations of memory without conscious awareness in normal adults. In addition, given that the results of Experiments 2A and 2B extend the generality of previous findings to another indirect task and another type of encoding manipulation, the present results suggest that greater relative sensitivity of indirect relative to direct measures is a phenomenon that may occur under a wide variety of experimental conditions.
The idea that indirect measures may be more sensitive than direct measures to unconscious influences is certainly not new. One of the earliest proponents of this view was Poetzl (1917) who claimed to demonstrate that unconsciously perceived information tends to emerge later in the content of subjects' dreams. The current widespread interest in comparing direct and indirect measures of memory is also based on the assumption, made either explicitly or implicitly, that indirect tasks are somehow more sensitive than direct tasks to unconscious influences. Consequently, whenever experimental results demonstrate a dissociation between a direct and an indirect task (e.g., Graf & Mandler, 1984; Kunst-Wilson & Zajonc, 1980), it is tempting to interpret such findings as evidence for unconscious processes.
The method for comparing the relative sensitivity of direct and indirect measures of memory illustrated by the present experiments extends and refines these earlier approaches to the study of unconscious processes. As noted previously, one problem complicating many comparisons of direct and indirect measures is that the measures are often quite dissimilar. Thus, interpretation of any observed dissociations is compromised by the fact that differences in performance across tasks may simply reflect task differences and not reflect differences in the underlying processes mediating performance. Matching tasks in terms of the methodological criteria proposed by Reingold and Merikle (1988) ensures comparability of direct and indirect measures and allows for a more straightforward interpretation of any findings indicating greater sensitivity of an indirect measure.
An important advantage of using comparable direct and indirect measures to study unconscious processes is that the methodological requirements are less stringent and the a priori assumptions more minimal than required by many other approaches. Often, when the sensitivity of direct and indirect tasks is compared, it is assumed that the direct measure must indicate null sensitivity if unconscious processes are to be demonstrated convincingly (e.g., Eich, 1984; Holender, 1986; Kunst-Wilson & Zajonc, 1980). Not only is true null sensitivity difficult to demonstrate, but null sensitivity is only meaningful if it is assumed that the direct measure provides an exhaustive index of all relevant conscious experience (Reingold & Merikle, 1988, 1990). If this exhaustiveness assumption is not made, then conscious processes may be involved even when a designated direct measure indicates null sensitivity. By using comparable direct and indirect measures, both the methodological problems associated with demonstrating null sensitivity and logical problems associated with justifying the exhaustiveness assumption are avoided. Rather, with comparable direct and indirect measures, unconscious processes are implicated whenever the indirect measure is more sensitive than the direct measure, even if the direct measure indicates considerable sensitivity.
When direct and indirect measures are comparable, it is also unnecessary to make any assumptions concerning consciousness at the time of initial encoding in order to study unconscious influences at the time of memory retrieval. In previous studies, unnecessary and difficult to justify assumptions have been made, even when the direct and indirect measures have been well matched. For example, both Eich (1984) and Kunst-Wilson and Zajonc (1980) assumed that the stimuli which were later remembered unconsciously were also perceived unconsciously. Eich assumed that the selective attention manipulation ensured that the unattended words were perceived without conscious awareness, and Kunst-Wilson and Zajonc assumed that brief exposures of the random polygons ensured that the subjects were never consciously aware of the stimuli. Without considerable converging evidence, these assumptions are difficult to justify. Fortunately, both assumptions are completely unnecessary. As noted by Reingold and Merikle (1988, 1990), with comparable direct and indirect measures, greater sensitivity of the indirect measure implicates unconscious memory processes, independent of any assumption concerning conscious awareness at the time of initial presentation.
At the empirical level, a consistent picture is beginning to emerge concerning some of the conditions under which it is possible to demonstrate unconscious processes. One empirical generalization is that indirect measures appear to exhibit greater sensitivity than comparable direct measures only when relatively few test trials are administered. Thus, in Experiments 2A and 2B, the indirect contrast measure was more sensitive than the direct recognition measure in the first two blocks of 32 test trials but not in the third block of 32 trials. Similarly, in previous studies involving comparable direct and indirect measures, successful demonstrations of greater relative sensitivity for the indirect measure have been based on as few as 6 test trials (Bornstein et al., 1987), as many as 64 test trials (Eich, 1984), or most frequently, 10 test trials for each task (e.g., Kunst-Wilson & Zajonc, 1980; Mandler et al., 1987; Seamon et al., 1984). A second empirical generalization is that indirect measures are more likely to exhibit greater sensitivity than direct measures when initial processing or encoding is limited. Thus, both brief exposures (e.g, Kunst-Wilson & Zajonc, 1980) and limited allocation of attentional resources, as demonstrated in Experiments 2A and 2B and in Eich's (1984) earlier study, are two conditions leading to greater relative sensitivity for indirect relative to comparable direct measures. Interestingly, Poetzl (1917) suggested that brief exposures and inattention were two equivalent empirical conditions for demonstrating unconscious processes.
The results of the present experiments demonstrate the value of comparing the relative sensitivity of comparable direct and indirect measures to study unconscious processes. When direct and indirect tasks are matched on all characteristics except instructions, greater sensitivity for the indirect measure necessarily implicates unconscious processes, given a minimal a priori assumption. The success of Experiments 2A and 2B suggests that characteristics of unconscious processes in normal adults can be revealed by studying other pairs of comparable direct and indirect tasks. Although we do not suggest that an approach based on comparisons of the relative sensitivity of comparable direct and indirect measures is the only way to investigate unconscious processes, we do suggest that this approach may provide more definitive results than many other approaches because fewer and more minimal a priori assumptions are required to interpret any observed dissociation between measures.
Bonnano, G. A., & Stillings, N. A. (1986). Preference, familiarity, and recognition after brief exposures to random geometric shapes. American Journal of Psychology, 99, 403-415.
Bornstein, R. F., Leone, D. R., & Galley, D. J. (1987). The generalizability of subliminal mere exposure effects: Influence of stimuli perceived without awareness on social behavior. Journal of Personality and Social Psychology, 53, 1070-1079.
Dunn, J. C., & Kirsner, K. (1989). Implicit memory: Task or process? In S. Lewandowsky, J. C. Dunn, & K. Kirsner (Eds.), Implicit memory: Theoretical issues (pp. 17-31). Hillsdale, NJ: Erlbaum.
Eich, E. (1984). Memory for unattended events: Remembering with and without awareness. Memory & Cognition, 12, 105-111.
Erdelyi, M. H. (1986). Experimental indeterminacies in the dissociation paradigm. Behavioral and Brain Sciences, 9, 30-31.
Graf, P., & Mandler, G. (1984). Activation makes words more accessible, but not necessarily more retrievable. Journal of Verbal Learning and Verbal Behavior, 23, 553-568.
Graf, P., & Schacter, D. (1985). Implicit and explicit memory for new associations in normal and amnesic subjects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 501-518.
Hintzman, D. (1990). Human learning and memory: Connections and dissociations. Annual Review of Psychology, 41, 109-139.
Holender, D. (1986). Semantic activation without conscious identification in dichotic listening, parafoveal vision, and visual masking: A survey and appraisal. Behavioral and Brain Sciences, 9, 1-23.
Jacoby, L. L., & Dallas, M. R. (1981). On the relationship between autobiographical memory and perceptual learning. Journal of Experimental Psychology: General, 110, 306-340.
Jacoby, L. L., Woloshyn, V., Kelley, C. (1989). Becoming famous without being recognized: Unconscious influences of memory produced by dividing attention. Journal of Experimental Psychology: General, 118, 115-125.
Jacoby, L. L., Allan, L. G., Collins, J. C., & Larwill, L. K. (1988). Memory influences subjective experience: Noise judgements. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 240-247.
Johnson, M. K., & Hasher, L. (1987). Human learning and memory. Annual Review of Psychology, 38, 631-668.
Kelley, C. M., & Jacoby, L. L. (1990). The construction of subjective experience: Memory attributions. Mind & Language, 5, 49-68.
Kihlstrom, J. F. (1987). The cognitive unconscious. Science, 237, 1445-1452.
Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, Rhode Island: Brown University Press.
Kunst-Wilson, W. R., & Zajonc, R. B. (1980). Affective discrimination of stimuli that cannot be recognized. Science, 207, 557-558.
Mandler, G., Nakamura, Y., Van Zandt, B. J. S. (1987). Nonspecific effects of exposure on stimuli that cannot be recognized. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 646-648..
Poetzl, O. (1917). The relationship between experimentally induced dream images and indirect vision. Translated by J. Wolff, D. Rapaport, and S. Annin. Psychological Issues, 1960, 3, Monograph 7, 41-120.
Reingold, E. M., & Merikle, P. M. (1988). Using direct and indirect measures to study perception without awareness. Perception & Psychophysics, 44, 563-575.
Reingold, E. M., & Merikle, P. M. (1990). On the inter-relatedness of theory and measurement in the study of unconscious processes. Mind & Language, 5, 9-28.
Richardson-Klavehn, A., & Bjork, R. A. (1988). Measures of memory. Annual Review of Psychology, 39, 475-543.
Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518.
Schacter, D. L. (1989). On the relation between memory and consciousness: Dissociable interactions and conscious experience. In H. L. Roediger, III & F. I. M. Craik (Eds.), Varieties of memory and consciousness: Essays in honour of Endel Tulving (pp. 355-389). Hillsdale, NJ: Erlbaum.
Schacter, D. L., Bowers, J., & Booker, J. (1989). Intention, awareness, and implicit memory: The retrieval intentionality criterion. In S. Lewandowsky, J. C. Dunn, & K. Kirsner (Eds.), Implicit memory: Theoretical issues (pp. 47-65). Hillsdale, NJ: Erlbaum.
Seamon, J. G., Marsh, R. L., & Brody, N. (1984). Critical importance of exposure duration for affective discrimination of stimuli that are not recognized. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 465-469.
Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recognition memory: Applications to dementia and amnesia. Journal of Experimental Psychology: General, 117, 34-50.
Tulving, E. (1985). How many memory systems are there? American Psychologist, 40, 385-398.
Vanderplas, J. M., & Garvin, E. A. (1959). The association value of random shapes. Journal of Experimental Psychology, 57, 147-154.
1 The visual angle subtended by each stimulus dimension is provided within parentheses following each linear measurement.
2 A'=0.5+[(H-FA)*(1+H-FA)]/[(4H*(1-FA)]; H=hits, FA=false alarms
3 One reviewer was concerned that the "old" and "new" items may have differed in some systematic way given that these item sets were selected randomly for each subject rather than being counterbalanced across conditions. This concern is unwarranted for two reasons. First, it is highly improbable that there was any bias in the random selection procedure that could lead to a systematic difference in the "old" and "new" sets of items assigned to each subject across the completely different populations of items used in Experiments 2A and 2B. Second, even if the item selection procedure had an undetected bias, comparisons of the relative sensitivity of the direct and indirect tasks would not be compromised since the same bias would have influenced item selection for both tasks.