W. J. (Wouter) de Winter, BSc.
Student number: 6095984
Brain and Cognition, Psychology department, University of Amsterdam (UvA) September, 2015.
Supervised by Prof. Dr. J.G.W. (Jeroen) Raaijmakers
UNIVERSITEIT VAN AMSTERDAM
The interaction between subliminal and
supraliminal items on long-term
repetition priming
Master thesis
That subliminal information can have short-term priming effects is a recognized finding. However, varying results have been found on the long-term effects of subliminal information. In this study, we explored the interaction between supraliminal and subliminal items, with varied repetitions, on long-term recognition as measured on perceptual identification and lexical decision after an interval of ten minutes. Test order was manipulated between groups, with subliminal items being presented before or after supraliminal items during perceptual identification. When subliminal items were tested prior to supraliminal items, the long-term repetition effect was absent. No effect of subliminal priming took place when items were only presented once. We conclude that the long-term subliminal priming effect observed was highly dependent on contextual information provided by supraliminal information. The relevance of these findings is discussed within the framework of the SAM-REM theory and we suggest a revision of some of the current memory models.
1
Introduction
In popular psychology, the public has long been intrigued by the possibility that subliminal (i.e. non-conscious) information can influence our thoughts and behaviors. The most notorious case of subliminal priming was a study on subliminal advertising. In 1957, James Vicary, a market researcher, claimed to have increased the sales of Coca Cola in a movie theatre by subliminally flashing messages on the screen. During the film, suggestive messages were flashed such as: ‘buy Coca Cola’ or ‘eat more popcorn’ – which supposedly led to an increase in consumptions. However, no one has been able to replicate his findings and the study was put off as a publicity hoax. Even so, the concerns about the use of subliminal advertising grew so large that in 1974 the Federal Communications Committee (FCC) decided to pass a bill against it. Furthermore, American consumers spend roughly $50 million dollars each year on self-help tapes, supposedly employing subliminal messages to cure ailments and substance abuse. Science has yet to provide us evidence in favor of the use of these tapes (Lowery, et al., 2007; Pratkanis & Aronson, 2001).
Yet, quite a number of studies have shown that subliminal perception can indeed influence immediately following behavior. Strahan, Spencer & Zanna (2002) demonstrated that subliminally priming thirst could influence people’s drinking behavior. However, this was only the case when people were already thirsty to begin with: the priming had no effect on people who were already satiated. A similar study assessed whether subliminal priming of a brand name of a drink can affect people’s choices for the primed brand, and whether this effect is moderated by individuals’ feelings of thirst (Karremans, Stroebe & Claus, 2006). In accordance with the results of Strahan, Spencer & Zanna (2002), only an interaction effect was found between cognitive state (i.e. thirsty) and subliminal priming: priming had a positive effect on brand choice, but only when people were thirsty. These studies suggest that subliminal priming has to be goal relevant to affect behavior.
However, some methodological questions can be raised on the studies of Strahan, Spencer and Zanna (2002) and Karremans, Stroebe and Claus (2006). First, both studies use relatively short priming durations, respectively 16ms and 23ms. The choice for these durations was probably made due to the researchers being overly concerned in ensuring that the priming was subliminal and not supraliminal. This might have led to too short presentation durations, which ultimately diminished the
2
subliminal priming effects. Other studies have shown that items can be presented subliminally with durations up to 50ms (Shah & Kruglanski, 2002; Neville et al., submitted).
Secondly, in the study of Strahan and colleagues (2002), primes were presented at one of eight random locations on screen. So while participants had to focus on the center, the primes were actually presented at a different location on the screen. This combined with the very brief presentation time (16ms) could make it very hard to perceive the prime, either consciously or non-consciously. If anything it is rather striking that an effect whatsoever was found.
Taking these points in to account, we could hypothesize that while goal relevant subliminal priming (vastly) increases the priming effect, it does not mean that subliminal priming cannot occur with non-goal related stimuli. This idea is supported by Aarts, Custers and Marien (2008), where people were primed to display forceful action. Their results show that subliminally priming exertion prepares people to supply forceful action, but when these stimuli were accompanied by subliminal reward signals it motivated people to spend extra effort. While this type of experiment was different compared to the studies of Strahan, Spencer and Zanna (2002) and Karremans, Stroebe and Claus (2006), its results can be generalized: a) subliminal priming can exist without goal (reward) related primes, b) including goal related primes increases the priming effect.
Another important methodological issue is the use of proper soft- and hardware. Since the presentation times during subliminal priming are so brief, the experimental equipment must be properly fine-tuned. If presentation times are longer than intended the primes could be consciously processed, nullifying the intention of subliminal priming. This is especially an issue in situations where a subliminal item has several repetitions. An example of such a situation is discussed by Zeelenberg, Plomp and Raaijmakers (2003). These researchers attempted to replicate the findings of a false memory effect following extremely short presentations as reported by Seamon, Luo and Gallo (1998). In the original study lists of words were presented for 20ms that were semantically related to a non-presented word, the critical lure. The hypothesis was that subjects would be subliminally primed to report the critical lure on the following recognition task. Their results showed that a false memory effect was created, even when subjects could not discriminate studied list items from distractors. In several experiments these results could not be replicated, even when presentation times were doubled
3
to 40ms (Zeelenberg, Plomp & Raaijmakers, 2003). The only logical explanation is that Seamon, Luo & Gallo (1998) used equipment which was not suited to accurately present an item for 20ms. Zeelenberg, Plomp and Raaijmakers (2003) suggest that presentation times were probably much longer, enabling conscious perception and thus leading to the ill-founded original conclusions.
As mentioned above, an issue in the field of research on subliminal priming is that a large number of studies use different (and often erroneous) methods. Usually, the measured effect is quite small so to control for confounds experiments must be performed in a well-constructed manner. A simple but controlled way to measure subliminal priming in an experimental environment can be by presenting an item (i.e. a word in a lexical decision task) for a very brief period of time (30ms – 40ms) preceded and immediately followed by a mask for 500ms. The subliminal priming effect can then be measured through a task relevant to the priming method (i.e. perceptual identification) (Neville et al., submitted).
There is sufficient evidence that subliminal information can have short-term priming effects (van den Bussche et al., 2009; Kiefer & Martens, 2010, in Neville et al., submitted). However, there has been some debate whether subliminal information also has term effects. We speak of long-term priming when behavior that does not immediately follow subliminal priming, is influenced.
The notion of long-term subliminal priming goes against the now quite common idea that storage in long-term memory requires attentive processing, and that without attentive processing information cannot be stored. Because masked or subliminal information is perceived non-consciously, it might enter working memory, but should not be able to enter long-term storage. However, research suggests that subliminal presentations can lead to storage of information in long-term memory. It must be noted that this has only been concluded when using implicit memory tasks. As of now, no evidence for subliminal long-term priming effects has been found when using an explicit or direct memory task. In another study it has been carefully suggested that subliminal presented stimuli may only show an effect on indirect tests of long-term memory (for a review see Raaijmakers & Neville, 2015).
Recent studies suggest that in the absence of supraliminal information, subliminal information may not have enough power to induce long-term priming effects. Neville and colleagues (submitted)
4
suggested that priming effects can be explained in terms of retrieval of information from long-term memory. Especially the SAM-REM theory could provide an appropriate framework as it has been shown to account for repetition priming effects in both lexical decision and perceptual identification tasks (Wagenmakers et al., 2004; Schooler et al., 2001, in Raaijmakers & Neville, 2015). According to the SAM-REM theory, memory traces are divided into semantic and episodic traces, each containing a mixture of three types of information: content features, context features and inter-item features. For implicit memory, priming effects are due to an update of an items’ pre-existing lexical-semantic trace with the features currently present in short-term memory. Long-term priming could be explained by the increase in matching features for those items that have been previously presented compared to novel ones.
Thus, an explanation for the lack of long-term priming effects in the absence of supraliminal information is that the supraliminal items are encoded on the lexical-semantic trace as context features that refer to the original study or presentation context of the subliminal items. These context features will help to activate subliminal items presented in the same context. Without the presence of supraliminal items, there are not enough matching features for priming to occur.
The goal of the current research is to set up a controlled experiment to investigate the interaction between supraliminal items and subliminal items, with varied repetitions, on long-term recognition as measured in standard behavioral memory tasks.
The experiment will consist of two phases, split by a ten minute interval. During the first phase participants are presented with both supraliminal and subliminal words. This will be done through a primed masked lexical decision task. In the second phase, participants will perform a perceptual identification task. During this task all the earlier presented words and a new list of novel words will be briefly presented (40ms) and participants will have to write down the word they thought they saw. The order of the words presented in the second phase will be manipulated to create two separate conditions. In the first condition all the supraliminal words, combined with half of the novel words, will be presented first, followed by the subliminal words and the other half of the novel words. In the second condition this will be the other way around: the subliminal words, mixed with half of the novel words will be presented first, followed by a mix of the supraliminal – and the rest of the novel
5
words. In the first condition, because the supraliminal words are presented prior to the subliminal words, there should be a sufficient match of context features and we expect that subliminal items are identified more frequently than control (the novel items). However, in the second condition, due to the supraliminal words being presented after the subliminal words there will not be a match in context features – there will not be a long-term priming effect and subliminal items will be identified equal to novel items. Finally, by alternating between one or three repetitions during the primed masked lexical decision task, we will explore the role of a repetition effect on perceptual identification.
Method
Participants
60 people participated in this study, consisting of 50 females and 10 males. Based on previous research 60 participants (30 per condition) were needed to ensure sufficient statistical power (Neville et al., submitted). Participants aged between 18 and 24 were recruited through the psychology student credit system of the University of Amsterdam. The average age of the participants was 20.25. All participants were native Dutch speakers with normal or corrected-to-normal vision.
Design
Because we examined the effects of both supraliminal as subliminal items, with varying order and repetitions, on perceptual identification, this lead to a 3 x 2 x 6 three-way mixed ANOVA with within-subjects factors of word type (subliminal, supraliminal, novel) and repetitions (one repetition, three repetitions) and a between-subjects factor of group (1 to 6) nested under position (subliminal first, supraliminal first). This study consisted of two blocks: a learning block and an experimental block. Data was collected through a computer-controlled task, and performance was measured in terms of the number of correct responses (correct identifications). RTs of correct responses were measured during the learning block. Participants with more than 25% errors during the first task were removed from analysis.
6
Materials and apparatusThree wordlists consisting of 30 words each, were selected from CELEX database, representing a subliminal presented wordlist (list A), supraliminal presented wordlist (list B) and a novel wordlist (list C) (Neville et al., submitted). Word length varied between four and six letters (M=4.62, SD=0.71) and word frequency between 3 and 630 occurrences per million (M=37.46, SD=91.71). In addition to the three wordlists, 30 pronounceable non-words were created for the lexical decision task. A Latin design was usedto ensure that each list was used in each condition (Winer, Brown & Michels, 1971). All lists had an equal number of words with the same length, and mean word frequency for each list was: list A) M=30.37, SD=41.15, list B) M=39.48, SD=115.43, list C) M=42.6, SD=104.31. The masks used in this study consisted of non-numeric and non-alphabetical signs.
The apparatus in this study was the same as in the experiments by Neville et al. (submitted) and van den IJssel (2014). Stimuli were presented on a CRT monitor at a frequency of 100 Hz and a resolution of 1024x768 via Presentation software (Version 0.70, www.neurobs.com). Font size was 25 points in Arial. The effectiveness of the used masking procedure in the primed masked lexical decision has been measured and validated by Neville et al. (submitted).
Procedure
During the first task, a primed masked lexical decision task, instructions were provided both on the screen and repeated orally by the experimenter. Participants were warned that the task would be difficult and required full attention. On each trial subjects were presented with the following sequence of events: first a fixation cross appeared at the center of the screen for 1000ms, then a mask was presented for 300ms, followed by a first word (subliminal prime) briefly presented for 40ms and immediately followed by a backwards mask (300ms). Afterwards a second word or non-word (supraliminal target) was presented for 1000ms or until a response was made, see Figure 1. The subliminal primes and the supraliminal targets were not presented in fixed pairs, but instead were repaired each trial. Using a Latin design, lists were counterbalanced across conditions between participants, meaning that participant 1 received list A subliminally and list B supraliminal, whereas
7
participant 2 had list B presented subliminally and list A supraliminal, and so on for list C. Half of the primes and targets were presented once, while the other half were presented three times for the repetition effect. Because half of the trials during the first task also consisted of non-words, this lead to a total of 120 trials. After a 10 minute break, where they were asked to solve a Sudoku puzzle, participants completed the test phase of the experiment involving a perceptual identification task on both novel and old items (the subliminal primes and supraliminal targets from the original list).
During the test phase all of the previous primes and targets of the study phase were presented for identification, plus 30 novel items, giving a total of 90 trials. One half of the subjects had to identify the subliminal primes first (plus novel words) and supraliminal targets later (plus novel words) and the other half of the participants vice versa. The items were presented one at a time, randomized for each subject separately.
Each perceptual identification trial consisted of the following sequence of events: a fixation cross was presented for 1000ms, followed by a mask for 300ms. Then the target word was presented for 40ms, followed again by a mask for 300ms. Participants were instructed to type the word they thought had just been presented, see Figure 1. After the initial response a correction screen was presented so that participants could correct any misspelling errors. Only correctly spelled words will be counted as correct responses.
Figure Error! Unknown switch argument. Schematic depiction of the structure of the experiments of Neville et al.,
8
At the end of the experiment participants were interviewed and debriefed.
Analysis
Our first hypothesis stated that subliminal primes, presented after the supraliminal targets, would be identified more frequently than the novel items (i.e., there will be a repetition priming effect). The null hypothesis here was that the subliminal primes, presented after the supraliminal targets, would not be identified more frequently than the novel items. The second hypothesis stated that subliminal primes, presented before the supraliminal targets, would not be identified more frequently than the novel items, the alternate hypothesis being that the subliminal primes would be identified at a higher level. A 3 x 2 x 6 three-way mixed ANOVA was used to determine the evidence for the hypothesis that the percentage of correct identifications is above control.
Results
Of the 60 participants, four were excluded from analysis. All four participants failed to achieve the requisite of at least 75% correct on the LD task.
Lexical Decision task
The average reaction times for one repetition and three repetitions on the LD task were calculated for each participant, resulting in two separate scores. For the PI task, the sum of correct responses was calculated for each possible combination, leading to six separate scores for each participant: (1) subliminal 1 reps, (2) subliminal 3 reps, (3) supraliminal 1 reps, (4) supraliminal 3 reps, (5) novel 1 reps, (6) novel 3 reps. Initially there was only one novel condition, but this was divided into a one repetition and three repetition group to fit the statistical model. This was done by using the same counterbalancing scheme as for the other word types.
A one-way ANOVA was performed to check if there we no difference between the mean correct scores of the groups on the LD task. Homogeneity of variances can be assumed (F < 1), and no significant difference between mean scores was found, F(5, 50) = 1.56, p = .188.
9
A 2 x 6 repeated measures ANOVA was used to analyze the RTs of the LD task, with a within-subjects factor of repetitions (one, three) and a between subjects factor of group (1 to 6). The condition of sphericity was met and equality of error variances can be assumed.
There was no significant effect of group (F < 1) on RT. This means that there was no significant difference between the RTs of the groups, suggesting that the different word lists provided to groups were of the same difficulty (word frequency). This is important as we do not want an effect of different words confounding the results of our other analyses.
There was a main effect of number of repetitions on RT, F(1, 50) = 87.13, p< .001. We can conclude that a repetition effect took place: items presented three times were responded to faster than an item only presented once, see Figure 2. These results confirm that our manipulation of the number of repetitions was a success.
Figure 2. The mean reaction times of 1 repetition and 3 repetition trials on LD for each group.
Perceptual Identification task
Data of the PI task were analyzed using a 3 x 2 x 6 three-way mixed ANOVA with within-subjects factors of word type (subliminal, supraliminal, novel) and repetitions (one repetition, three repetitions) and a between-subjects factor of group (1 to 6) nested under position (subliminal first, supraliminal
10
first). Mauchly’s test indicated that the assumption of sphericity had been violated for the within-subjects factor condition (χ²(2) = 7.76, p = .021), therefore degrees of freedom were corrected using Huynh-Feldt estimates of sphericity (ε = .872). An issue with this analysis is that all experimental conditions had significant results on Levene’s test, violating the assumption of equality of error variances. This means that no strict conclusions can be made of the current results.
There was a significant effect of group on identification, F(4, 50) = 4.042, p = .006. This effect tells us that regardless of the type of word presented and how many repetitions, groups differed in their scores on identification. As the factor group is nested under position, this result was not unexpected. However, there was no significant effect of position (F<1). The largest difference in mean scores was between groups 1 and 4, see Table 1.
Table 1 Percentage of identified correct across all trials on the PI task for each group.
The percentages in Table 1 show that group 4 scored much higher on the PI task compared to the other groups. Also, group 1 seems to score lower than the other groups. Due to counterbalancing these differences will not have a confounding effect on the other results.
There was no significant effect of word type, indicating that participants scored the same for each type of presented word (sub, supra, novel) regardless of group and the number of presentations. This does not follow our expectations, as supraliminal presented items were consciously viewed and presented several times during the study phase; therefore they should be perceived more often than novel items. The fact that subliminal items are not perceived more than control is in line with our hypotheses: subliminal presented words are only identified more than novel words when they are presented after the supraliminal targets. However, when looking at our planned simple contrasts of our word type factor there was a significant difference between supraliminal words and novel words, F(1,
Group Percentage Correct
1 3.0% 2 7.6% 3 9.2% 4 25.9% 5 18.0% 6 13.8%
11
57) = 4.67, p = .036. This shows that supraliminal presented words are indeed perceived more often than novel, no matter the number of repetitions or the group (or position) they are placed in.
No main effect of repetitions was found, suggesting that participants across all groups scored the same for all words presented either once or thrice. This would seem unexpected, as more repetitions should lead to a larger probability of identification. However, this result does not tell us much: to observe the true effect of number of repetitions on identification, we would have to look at the interaction between the number of repetitions and type of word.
Indeed, there was a significant interaction between number of repetitions and the type of word presented on identification, F(2, 100) = 47.416, p = .005. This effect means that the effect of number of repetitions on identification was different for the three types of words. A closer look at the interaction suggests that an increase in repetitions leads to a large increase in identification of subliminally primed words, while the identification of supraliminal and novel words stayed the same, see Figure 3. It must be noted that the novel items were divided into two sets according to the counterbalancing scheme to fit the statistical model. Figure 3. suggests that the novel items were also presented either once or three times, but this was not the case.
12
Figure 3.The interaction between word type and number of repetitions on mean correct scored on PI task.
Our planned contrasts confirm that an increase in repetitions leads to a large increase in identification of subliminally primed words showing a linear effect, F(1, 50) = 6.235, p = .016. No repetition effect was found on supraliminal words (F < 1). It was to be expected that subliminal presented words would profit greatly from the extra number of repetitions, simply because the chance that they are perceived non-consciously increases for each extra repetition, thereby leading to an increased chance of identification. Still, it is unclear why a second or third rehearsal has no influence on the chance of perception of supraliminal items. It could be possible that task difficulty leads to a ceiling effect on perception, but other experiments with the same task did show an effect of repetition (Neville, et al., submitted).
There was a significant interaction between word type and position, F(2, 100) = 17.44, p < .001. This means that the effect of position on identification was different for supraliminal words compared to subliminal words. A visual representation of the interaction effect is shown below; see Table 2 and Figure 4.
13
Figure 4. The mean scores
correct on identification of the interaction between word type and position.
Table 2. The percentage correct of identification on the PI task for each type of word in both positional conditions.
Subliminal Supraliminal Novel Supraliminal first 14.9% 9.9% 10.2%
Subliminal first 8.7% 19.2% 13.6%
Both supraliminal and subliminal words showed differences between identification, based on their position. When supraliminal items were tested first, subliminal words scored higher than novel and also higher than the supraliminal words. Pairwise comparisons showed that these differences are significant, p = .006 and p = .029 respectively. Supraliminal words are identified equal to novel when they were tested first. This is unexpected, as we would expect a priming effect of supraliminal items, regardless of the position they are placed in. When subliminal items were tested first, subliminal words scored lower than novel, p= .004. Again an unexpected result, as the subliminal presented words should at least be identified as often as the novel words. Supraliminal words, however, were identified more often than novel when subliminal items were tested first, p = .002. These results suggest the presence of a practice effect on participants’ performance. The trials which were tested later, scored higher than the trials tested first.
14
There was also a significant interaction between word type and group, F(8, 99) = 2.06, p = .048. This means that groups responded differently to the effect of word type on identification. This is in addition to the interaction between word type and position, as the factor group was nested in position. When looking at the groups separately we see patterns similar to the one depicted by the interaction between word type and position, see Figure 4 and Figure 5.
Figure 5.The interaction between the different groups and word type on identification.
Groups 1, 4 and 6 showed the same pattern as the subliminal first condition, whereas groups 2, 3 and 5 showed the same pattern as the supraliminal first condition displayed in Figure 4. However, when looking at the pairwise comparisons, no significant difference was found between the identification of the three word types in groups 1, 2 and 3. In groups 4 and 6 there was a significant difference between the identification of supraliminal words and novel words (p = .005 and p = .017 respectively), and in group 5 there was a significant difference between the identification of subliminal words and novel words (p = .005). These significant differences are in line with our expectations. Yet, group 4 also shows a significant negative difference between the identification of subliminal words and novel words. This is the same phenomenon that presented itself in the interaction between word type and condition and the pattern shown in Figure 5., just like in Figure 4., also suggests the presence of a practice effect.
These results support our hypotheses that (1) subliminal primes, presented after the supraliminal targets, would be identified more frequently than the novel items and that (2) subliminal primes, presented before the supraliminal targets, will not be identified more frequently than the novel items. This suggests that supraliminal items presented prior to subliminal items activate
15
context information, ensuring the memory threshold is met for subliminal long-term priming. Remarkably, it seems that this effect also works in the opposite direction; supraliminal items, tested after subliminal items are identified more frequently than novel items, however when they are tested before subliminal items they are not. This could mean that, although the subliminal items are not activated enough for identification, they do contribute to the activation of context information, leading to an enhanced frequency of the recognition of supraliminal items. It could also mean, as suggested before, that a training effect was present, which lead participants to improve in task performance over the course of time.
Finally, there was a significant three-way interaction between the type of word presented, number of repetitions and in which position it was placed, F(2, 100) = 3.08, p = .05. This interaction tells us whether the interaction between word type and position was the same for one and three repetitions. It seems that the number of repetitions has no effect on identification of all word types in the subliminal first condition. However, the number of repetitions does have an effect on identification in the supraliminal first condition, see Figure 6.
Figure 6. The interaction between word type and position on identification for each level of repetition.
These findings are in accordance with the interaction we found between number of repetitions and word type. Subliminal presented words that are only presented once are not identified more than
16
novel, regardless of the positional condition they are placed in. These results give an insight into the underlying neural mechanisms of the activation of context information and subliminal identification.
General Discussion
The current experiment investigated the interaction between supraliminal items and subliminal items, with varied repetitions, on long-term subliminal priming. The results support our hypotheses that (1) subliminal primes, presented after the supraliminal targets, are identified more frequently than the novel items and that (2) subliminal primes, presented before the supraliminal targets, are not identified more frequently than the novel items. These findings support the suggestions of Neville et al. (submitted) that in the absence of supraliminal information subliminal items may not have enough power to induce long-term priming effects. However, in the presence of supraliminal information subliminal items can induce long-term priming effects: when supraliminal items were tested first, subliminal items were perceived above chance level. This was only the case when the subliminal items were presented several times during the lexical decision task. No effect of subliminal long-term priming took place for items only presented once. Other studies have found similar repetition effects (Aalbrecht & Volberg, 2010; Neville et al., submitted).
According to the SAM-REM model, these findings can be explained due to the fact that the supraliminal items are encoded on the lexical-semantic trace as context features that refer to the original study of the subliminal items. Long-term priming occurs due to an increase in matching (context) features for those items that have been previously presented compared to novel ones. In the absence of the context features created by the supraliminal items, the lexical-semantic trace of the subliminal items does not have enough matching features to be distinguished from novel items. As for the repetition effect, it seems that several subliminal presentations are necessary to create a lexical-semantic trace strong enough for long-term priming.
These findings go against the now widely accepted memory models, which claim that attentive processing (rehearsal) is necessary for information to enter long-term storage. This study has given strong evidence that information, which is not consciously processed, can still enter long-term storage. We suggest that the current models are incomplete, missing another pathway through which
17
(subliminal) information can enter long-term storage. Future research will have to decide if this pathway goes straight from the sensory register, or passes through short-term memory on its way to long-term storage.
Remarkably, our hypotheses on subliminal items also held true for supraliminal items: supraliminal items were only perceived above chance if they were preceded by subliminal items. This finding is rather unexpected as the supraliminal items have received attentive processing and should be stored in long-term memory. An explanation for this finding could be that the task proved to be too difficult. Comparable with our results on subliminal items, the supraliminal words require subliminal contextual information to successfully be identified. Placing these results in to the SAM-REM model, we could suggest that subliminal information only presented once can be encoded as context features on supraliminal information, enough so that identification occurs when it otherwise would not. This idea is comparable to the findings of Seamon, Luo and Gallo (1998) where false memories were created off subliminal cues. In their study as in the current study, subliminal information which could not be told apart from distractors (novel items), can still manage to influence memory. Their study was heavily criticized after proving unable to replicate their findings, suggesting that the wrong equipment was used (Zeelenberg, Plomp & Raaijmakers, 2003). The results of the current study would suggest that maybe another look should be taken at the results of Seamon and colleagues (1998).
However, another possibility is that the current study suffered from a training effect. Due to the perceptual identification task being quite difficult, participants required more than just the practice trials to adapt to the experiment. This resulted in an improvement of performance over time, where items presented at the end would be recognized more often than items presented at the start of the task. Due to the way the current experiment is set up – with the manipulation of presentation order – this can have quite large consequences on our results. A training effect would explain why no effect of supraliminal items on long-term priming was found when these items were tested first. It would also explain why subliminal items, when tested first, were identified less often than the novel items.
This study does contain a few limitations. First off, our main analysis violated the assumption of equality of error variances. This was most likely caused by the fact that a few participants were excluded from analysis which lead to unequal group sizes. Because of this violation it is suggested that
18
all conclusions based on these statistical results must be taken lightly. However, when measuring correct responses, it is not uncommon to violate this assumption. Correct responses usually result in a skewed distribution, thus the normality assumption is easily violated (Glass & Hopkins, 1996).
Secondly, a main effect of group was found. This suggests that there was some confounding factor influencing the performance of some of the groups, even though word frequency and word length were the same for all groups. A possible explanation for this is that some groups received a different reward. Groups 1, 2, 3 and 6 could only receive student credit as a reward for participation, whereas subjects for groups 4 and 5 could also receive a monetary reward. This difference in reward could lead to a difference in motivation and result in varying performance across groups.Indeed, when comparing performance between monetary and student credit reward groups using an independent samples t-test a significant difference was found, t(58) = -3.42, p< .001. Monetary reward groups performed much better than the groups only able to receive student credit. It must be noted, that due to counterbalancing the effect of group was minimized on our final results.
Due to the limitations of this study, the current experiment must be replicated to confirm our conclusions. We suggest having a larger number of practice trials on the perceptual identification task to control for a training effect. Furthermore, we suggest that studies in the future use only one type of reward for participation to prevent motivational confounds.
Altogether, this study suggests that masked or subliminal presentations do not lead to an effect of long-term priming. However, in the combination of supraliminal information updated on the lexical-semantic traces, subliminal priming can occur if the required number of repetitions is met. It also seems that supraliminal primes might be able to benefit in a similar fashion, through the encoding of subliminal information on their lexical-semantic traces. It seems that there may be another path leading to long-term storage, which does not require attentive processing. These findings suggest that the most common memory models currently used in psychology may need to be revised.
20
References
Albrecht, T., & Vorberg, D. (2010). Long-lasting effects of briefly flashed words and pseudo words in ultra-rapid serial visual presentation. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 36, 1339–1345.
Aarts, H., Custers, R. & Marien H. (2008). Preparing and motivating behavior outside of awareness.
Science, 319, DOI: 10.1126/science.1150432.
Glass, G. V., & Hopkins, K. D. (1996). Statistical methods in psychology and education. Allyn &
Bacon, Needham Heights, MA.
Karremans, J. C., Stroebe, W. & Claus, J. (2006). Beyond Vicary’s fantasies: The impact of subliminal priming and brand choice. Journal of Experimental Social Psychology, 42, 792- 798.
Lowery, B. S., Eisenberger, N. I., Hardin, C. D. & Sinclair, S. (2007). Long-term effects of subliminal priming on academic performance.Basic and Applied Social Psychology, 29:2, 151-157, DOI: 10.1080/01973530701331718.
Neville, D. A., van Maanen, L., van Gaal, S. & Raaijmakers, J. G. W. (submitted). Durable repetition priming for subliminal words, submitted.
Pratkanis, A. R., & Aronson, E. (2001). Age of propaganda: The everyday use and abuse of persuasion. New York: Henry Hold and Company LLC.
Raaijmakers, J.G.W., & Neville, D.A. (2015). Is conscious processing required for long-term memory? In J.G.W. Raaijmakers et al. (Eds.), Cognitive modeling in perception and memory: A Festschrift for Richard M. Shiffrin. (Pp. 162-172). New York: Psychology Press.
Strahan, E. J., Spencer, S. J. & Zanna, M. P. (2002). Subliminal priming and persuasion: striking while the iron is hot. Journal of Experimental Social Psychology, 38:6, 556-568.
Shah, J. Y. & Kruglanski, A. W. (2002). Priming against your will: how accessible alternatives affect goal pursuit.Journal of Experimental Social Psychology, 38, 368-383.
Van den IJssel, D. V. (2014). Subliminal priming. Master thesis. Brain and Cognition, University of Amsterdam.
21
Winer, B. J., Brown, D. R., & Michels, K. M. (1971). Statistical principles in experimental design (Vol. 2. Pp. 596). New York: McGraw-Hill.
Zeelenberg, R., Plomp, G., & Raaijmakers, J.G.W. (2003). Can false memories be created through nonconscious processes? Consciousness and Cognition, 12, 403-41
