A Critical Analysis of the Unconscious Working Memory Paradigm
A Replication Study
Joris Huese
Abstract
There have been recent findings suggesting that working memory and consciousness might be dissociative cognitive processes. These results contradict the main view that these processes are heavily associated. This study replicated studies which used the unconscious working memory paradigm, an experimental paradigm used to show a possible dissociation between visual short term memory and conscious awareness. This paradigm uses subliminal distracters to impair performance on a visual short term memory task. Here, the validity of these previously found results was assessed. The possibility that unconscious working memory effects might be due to perceptual sensitivity for the distracter was assessed in light of the scientific discussion concerning perceptual threshold methods. A circular grating mask condition was added to explore perceptual sensitivity differences between mask efficiency in this paradigm. Furthermore an exploratory analysis was performed to study unwanted noise in the data generated by trial element randomisation. No visual short term memory
impairment effects were found, which inhibited this study in concluding anything about the validity of unconscious working memory effects. Participants were significantly perceptually sensitive to the black mask, and not to the circular grating mask. This result has implications for the perceptual threshold method debate. Significant systematic unwanted noise generated by trial randomisation was found. This consisted of a response bias for right button responses and a visual priming effect of distracter. These results show that this paradigm needs to be simplified to reduce the large number of trial sub-types which are normally ignored in the general statistical analysis.
Introduction
Working memory (WM) is heavily associated with human consciousness. All active components of WM are conscious: input, rehearsal, visuospatial operations, recall and report (Baars &Franklin, 2003, P. 170). Additionally, WM has crucial roles in leading theories of human consciousness. The global neuronal workspace theory by Dehaene (2006), which is based on the global workspace theory by Baars (1997), sees WM as the structure that consciousness uses for its functions. Models of phenomenal consciousness by Block and Lamme show that WM is crucial for but not synonymous to conscious experience (Block, 2005; Lamme, 2003). Lamme (2003) states for example that access consciousness relies on WM but phenomenal consciousness relies more on iconic memory, where recurrent
processing in visual cortex areas is sufficient for conscious experience.
Contrasting this idea of WM and human consciousness being intertwined mechanisms are recent masking studies which study the effect of subliminal distracters on visual short-term memory (VSTM), the visual component of WM. The study by Soto & Silvanto (2012) is one of the most clear examples showing a possible unconscious working memory
interference effect. Here a discrimination task was used to assess subliminal stimuli effects on explicit tasks. This study found that subliminal distracters impair performance on a
discrimination task where a supraliminal memory cue and probe were used. These findings, replicated in studies by Bona, Cattaneo, Vecchi, Soto, & Silvanto (2013) and Soto & Silvanto (2014), contradict the mainstream view that WM and conscious awareness are not
dissociable. Based on these findings, Soto & Silvanto (2014) claim that WM processes can be engaged without awareness and that multiple factors in the perception awareness process may lead to a dissociation to WM contents and subjective conscious experience.
To analyse the validity of the claims made by Soto & Silvanto (2012;2014) this study takes a closer look at the experimental design on trial level. The most basic design of the unconscious working memory paradigm is used in Soto & Silvanto (2012). A discrimination paradigm was used where VSTM performance was assessed. On each of 288 trials a memory cue grating was presented for 200 ms. This grating was tilted either to left or right, with random orientations differentiating between 30, 40 or 50 degrees. This memory cue was followed by a maintenance period after which in 66% trials, there was a distracter presented for 13ms which was masked by a black mask for 80ms. In the other 33% trials only the mask was shown. The distracter shown in 66% of trials was either congruent – same orientation as the memory cue – or incongruent – same amount of tilted degreed but opposite side
orientation. After the mask followed the second 1000ms part of the maintenance period, which was followed by a 300ms presented memory probe. This memory probe was tilted 10 degrees either to the left or to the right compared to the memory cue. After the probe,
participants had to indicate the orientation of the memory probe compared to the memory cue by pressing the left or right arrow key. Following this choice the participants had to indicate whether they consciously perceived a distracter, this was done by a 4 - point Perceptual Awareness Scale (PAS) adapted from Overgaard et al. (2010) and Sandberg et al. (2010).
The conclusions of the studies using the unconscious working memory paradigm rest on the assumption that the distracter is indeed not consciously perceived by participants. In order to show a dissociation between WM and awareness participants should show null sensitivity on a direct task of conscious awareness while showing sensitivity on an indirect task which indicates unconscious processing (Snodgrass, Bernat & Shivrin, 2004).
In case of the unconscious working memory paradigm this means that there should be perceptual null sensitivity of the distracter indicated by the PAS while observing performance differences on the discrimination task based on distracter present/absent conditions.
Assessing null sensitivity on the direct task using a only subjective threshold method like the PAS is problematic. Firstly, when using the PAS to construct a subjective threshold where researchers can conclude that a stimulus was not consciously perceived, the assumption that the different answers on the PAS directly reflect different perceptual states has to be made (Schmidt, 2015). This means that when a participant answers 1 on the PAS a researcher can conclude that the presented stimulus was not consciously perceived and when a 4 is pressed there can be concluded that the stimulus was consciously perceived. This direct approach disregards the inevitable existence of measurement noise, individual differences in response criteria and guessing behaviour of participants (Snodgrass, Bernat & Shivrin, 2004).
Secondly, an additional problem with using the PAS to construct a subjective threshold is that it encourages the use of the not-seen-judgements-only procedure (Schmidt, 2015), which means looking only at the trials where participants indicated that they did not consciously perceive a stimulus. Restriction of the sample in this manner creates a false sense of
confidence in the used experimental design, the use of only trials where participants indicated they did not perceive a stimulus does say anything about the general visibility of the stimulus.
Looking at the studies by Soto & Silvanto (2012;2014) illustrates the two mentioned problems. In both studies the PAS is used to construct a subjective threshold of conscious perception. Firstly it is assumed that the manipulation of the distracter was successful because the majority of responses to trials where a distracter was presented are coupled to a 1 press on the PAS. Secondly they use the not-seen-judgements-only procedure to see if there is an effect. Since by using these methods Soto & Silvanto (2012;2014) cannot know if null sensitivity for the distracter has been achieved, they violate the main condition for proving a dissociation between WM and awareness (Snodgrass, Bernat & Shivrin, 2004).
Signal detection theory (SDT) takes care of the problems. By taking all response information on the direct task and condition information on the indirect task, you can
calculate the sensitivity to a stimulus based on the proportion of hit/miss rate and false alarm/correct rejection rate. To assess null sensitivity for a stimulus a measure called d’ can be used (Stanislaw & Todorov, 1999). A d’ of 0 means that there is no sensitivity to a stimulus, where the higher the d’ the more sensitivity to the stimulus there is. Under the assumptions of SDT, d’ is unaffected by response bias. Additionally, SDT uses all response data to assess the sensitivity to a stimulus which eliminates the need for post-hoc sample reduction which may comprise the integrity of research findings.
In their highly similar study to Soto & Silvanto (2012), bona et al. (2013)
acknowledge that SDT should be used to assess null sensitivity on the distracter awareness. They find however that there is no null sensitivity for the distracter, a mean d’ of 1.53 was found which significantly differs from the needed d’ of 0. This is problematic seeing the similarity of experimental design between bona et al. (2013) and Soto & Silvanto
(2012;2014) indicating that there is a high probability that the distracters in these studies were also not objectively invisible.
This study will replicate the basic paradigm designed by Soto & silvanto (2012). To assess if the results found by Soto and Silvanto (2012;2014) and Bona et al. (2013) are not attributed to artefacts of response bias, other measurement noise or non-null sensitivity to the distracter created by merely using a subjective threshold method, an objective threshold condition will be added where a circular grating mask is used to reach a d’ of 0. Based on the liberal methods Soto & Silvanto (2012; 2014) used to claim null sensitivity on distracter awareness and the found non-null sensitivity in Bona et al. (2013), the hypothesis is that there will be a directly proportional relationship between d’ and the impairment of performance on the discrimination task. Where the lower the perceptual sensitivity to the distracter, the lower the performance impairment. Since the methods used by Soto & Silvanto resulted in conclusions that could be attributed to weakly conscious effects created by the
sensitivity to the distracter created by the sole use of subjective thresholds to assess perceptual sensitivity (Snodgrass, Bernat & Shivrin, 2004).
In addition to the replication analysis and the effects of the circular grating mask condition this study will explore possible unwanted noise effects in the unconscious working memory paradigm. More specifically, a response bias effect of left versus right presses and a possible priming effect of the distracter. The distracter is hypothesized to impair VSTM performance, visual priming studies suggest that the distracter might not work as intended in certain trials (Schacter & Buckner, 1998). Namely, in the trials where the direction of the distracter grating is opposite to that of the memory cue and where the distracter orientation is congruent with the correct response to the memory probe. This can only occur in the
conditions where the distracter is present and has a 40 degrees orientation difference.
Methods
Participants
33 first year students of the University of Amsterdam completed the experiment. Only participants who performed adequately enough to reach a mean performance accuracy of .67 on the no distracter trials were included in the data analysis which resulted in 26 participants aged M = 20.84, SD = 1.84. Four extra participants were excluded for the visual awareness analysis for the black mask condition, given that they reported to have never seen a distracter. All participants confirmed being adequately informed by providing a written consent form before starting the experiment and received a Educational Credit for their participation.
Materials
An adaption to the experiment of Soto & Silvanto (2012) was used that is highly similar to the adaption that Bona et al. (2013) used in their study. See Figure 1 for a detailed explanation on trial basis.
Each participant completed a total of 480 trials, with forced breaks of at least 20 seconds after intervals of 80 trials. Memory cue orientation was randomised between -50 to 50 degrees in steps of 10 degrees. Memory probe offset was randomised either +10 degrees or -10 degrees compared to the memory cue. Distracter was present in 320 (66%) of trials and absent in 120 (33%). Distracter trials consisted evenly distributed of either a congruent
Figure 1. A schematic overview of 1 trial in both masking conditions. A memory cue is presented for 300ms followed by a black mask for 83ms. A 1500ms rest period is followed by a distracter in 66% of total trials. A circular grating mask or a black mask is presented for 83ms depending on mask condition. This is again followed by a 1500ms resting period and a 500ms fixation point rest period. After which a memory probe is presented which has a -10 or +10 degrees orientation compared to the memory cue. This is followed by a forced ‘Left or Right’ choice indicating the relative orientation of the memory probe. Finally a PAS was used to assess subjective conscious awareness of participants, adapted from Overgaard et al. (2010) and Sandberg et al. (2010).
Results
Distracter visibility responses and perceptual distracter sensitivity
This study first assessed the perceptual sensitivity to the distracter in both mask conditions. Figure 2A and 2B show the proportion PAS responses and corresponding d’ values per mask condition. One-Sample T tests were used to assess perceptual sensitivity to the distracter on both mask conditions. The mean d’ value of the black mask condition (0.6) was significantly different from 0 t(25) = 4.375, p < 0.01. Which shows non-null sensitivity to the distracter in the black mask condition. The mean d’ value of the circular grating mask condition (-0.026) was not significantly different from 0 t(25) = -0.375, p = 0.71. Which shows null sensitivity to the distracter on the circular grating mask condition. Finally a Paired-Sample T test was used to assess if using a circular grating mask significantly lowered the mean d’ value by comparing mean d’ scores of both mask conditions, which it expectedly did, t(25) = 4.647, p < 0.01.
Figure 2 (A). Proportion of PAS responses for both distracter present and distracter absent trials. Black mask condition. Calculated d’ = 0.6. Error bars indicate +- 1 SEM. (B). Proportion of PAS responses for both distracter present as distracter absent trials. Circular grating mask condition. Calculated d’ < 0. Error bars indicate +- 1 SEM.
Main analysis
For the main analysis this study replicated the main analysis by Bona et al. (2013). For the black mask condition distracter present trials were split by congruent (0 degrees) and incongruent (40 degrees) trials. Additionally trials were split by subjective visibility, where PAS responses of 1 were assumed to be non-visible and PAS responses of 2-4 were assumed to be visible trials. Figure 3A shows the mean VSTM performance per condition for the black mask trials. A factorial repeated-measures ANOVA was used to assess the effects distracter orientation and visibility on VSTM with distracter congruency and distracter visibility as indexed by the PAS as factors. This ANOVA included only distracter-present trials. The ANOVA found no effect of distracter congruency F(1,20) = 0,504, p = 0,486. Additionally, there was no effect of distracter visibility F(1,20) = 0,917, p = 0,350 and no interaction F(1,20) = 0,212 p = 0,650.
Effects of distracter on VSTM on the black mask condition were assessed by comparing mean VSTM performance on the no distracter trials and the incongruent (40 degree) distracter trials using a Paired-Sample T test. There was no effect of distracter presence in the black mask condition t(25) = -0.117, p = 0,908.
Effects of distracter on VSTM on the circular grating mask condition were assessed in similar fashion. Figure 3B shows the mean VSTM performance per condition on the circular grating mask trials. The Paired-Sample T test showed no significant difference on VSTM performance in the no distracter trials compared to incongruent distracter trails, t(25) = -1.059, p = 0,299. Indicating, no effects of distracter presence.
Figure 3 (A). Mean (n = 21) VSTM performance in the black mask condition as a function of distracter type. Distracter awareness was included as a condition on distracter present trials. Error bars indicate +- 1 SEM. There was no effect of distracter type, distracter awareness or interaction effects on VSTM performance. There was no effect of distracter presence on VSTM performance. (B). Mean (n = 26) VSTM performance in the circular grating mask condition as function of distracter type. Error bars indicate +-1 SEM. There was no effect of distracter presence on VSTM performance.
Correlation
This study also looked at the possible correlation between VSTM performance impairment and d’ values in both mask conditions. VSTM performance impairment was calculated by subtracting incongruent distracter trials from no distracter trials. This value was then correlated to d’ values. Figure 4A and 4B show the correlations for both the black and the circular grating mask conditions. No significant correlations between VSTM impairment and d’ value were found in both the black mask condition, r(24) = -0.238, p > 0.05, and the circular grating mask condition, r(24) = 0.255, p > 0.05.
Figure 4 (A). Correlation between d’ value and VSTM performance difference between the incongruent distracter and no distracter condition for each participant. No significant correlation was found. Mask = black mask. (B). Correlation between d’ value and VSTM performance difference between the incongruent distracter and no distracter condition for each participant. No significant correlation was found. Mask = circular grating mask.
Exploratory Analysis of unwanted effects caused by trial randomisation
The Unconscious Working Memory paradigm used in this study has 3 distracter type conditions (no, congruent, incongruent) and 2 mask type conditions (black, circular grating). These are however not the only randomised elements when constructing each of the 480 trials. Memory cue, distracter orientation and memory probe orientation are randomised on each trial. This can lead to unwanted effects.
In this study we explored possible unwanted effects caused by trial randomisation. Incongruent distracter trials were chosen as main point of interest. Firstly because if participants show an impairment of VSTM, previous studies show that this effect is exclusively seen in Incongruent distracter trials (Soto & Silvanto, 2012; 2014; Bona et al. 2013). Secondly because trial randomisation could possibly have the largest unwanted effects here. For example a 40 degree incongruent distracter orientated to the right compared to a -20 degree left orientated memory cue crosses the vertical line which makes it a +20 degree right
orientated stimulus. Now, if in the randomisation process this distracter would be orientated to the left compared to the memory cue, it would be a -60 degree left orientated stimulus.
Figure 5 illustrates 12 possible incongruent distracter trial configurations. It is clear from this illustration that vertical crossing of distracter stimuli creates a subset of
qualitatively different trials that are not accounted for in the general statistical analysis of the unconscious working memory paradigm.
Figure 5. This tree plot illustrates 12 different trials subtypes within the incongruent
distracter condition of the unconscious working memory paradigm. L/R Cue = Left or right memory cue orientation. L/R Distr = distracter orientation compared to memory cue.
Cross/Not Cross Vert = vertical crossing of distracter. L/R Probe = memory probe orientation compared to memory cue.
The first exploratory analysis concerns a possible response bias toward left or right button presses. Figure 6A shows the mean proportion of Left/Right responses. With no response bias an evenly distributed response ratio is expected here since memory probe orientations are randomised. There was however a significant difference in mean response button presses. A Paired-Samples T test showed that participants had significantly more right button presses than left button presses t(25) = -3.469, p < 0.01. Effect size and post-hoc power calculations were performed for this response bias effect. The Effect Size (ES) in this analysis was -0.680, considered to be medium using Cohen's (1992) criteria. With an alpha = .05 and sample size of 26 the achieved power was calculated to be 0.915 (GPower 3.1)
Subsequently the effect of this response bias was studied. Figure 6B shows the mean VSTM performance per memory probe orientation. The right button response bias has a clear effect on VSTM performance. Paired-Samples T test for the incongruent distracter condition showed participants had significantly better VSTM performance in the right memory probe condition compared to the left memory probe condition t(25) = -3.348, p < 0,01. Effect size and post-hoc power calculations were performed for this response bias effect. The Effect Size (ES) in this analysis was -0.657, considered to be medium using Cohen's (1992) criteria. With an alpha = .05 and sample size of 26 the achieved power was calculated to be 0.896 (GPower 3.1)
Figure 6 (A). Bar graph showing the mean proportion of button presses of all participants. Participants had a tendency to press the right button more compared to the left button. This was a significant difference. (B). Mean VSTM accuracy per memory probe orientation compared to the memory cue. This corresponds with the button that needed to be pressed. Participants performed significantly better when a right press was needed. This is most likely due to their tendency to press the right button.
The final analysis explored a possible priming effect of the distracter in the incongruent distracter condition. Here absolute distracter orientation was compared to memory probe orientation, such that when the final orientation of the distracter was right of vertical (> 0 degrees offset) it was approached as a right stimulus and when it was left of vertical (< 0 degrees offset) it was approached as a left stimulus. These absolute distracter orientations were then compared to memory probe orientations related to the memory cue (i.e. the required button press response). When the absolute orientation of the distracter (left or right compared to vertical) was the same as the required button press (memory probe orientation compared to memory cue) this was seen as congruent, when absolute distracter orientation was not the same as the required button press it was seen as incongruent. Since this analysis explores unwanted effects caused by randomisation, only trials where distracter orientation randomisation could cause qualitatively different trial types are taken into
distracter orientation randomisation could result in vertical crossing of the distracter stimulus. Figure 7 shows the mean VSTM performance for congruent absolute distracter orientation trials compared to Incongruent absolute distracter orientation trials. Participants performed significantly better on congruent absolute distracter orientation trials t(25) = 2.932, p < 0.01. This indicates a priming effect of absolute distracter orientation on required button press in trials where incongruent distracters are present. Effect size and post-hoc power calculations were performed for this priming effect. The ES in this analysis was 0.570, considered to be medium using Cohen's (1992) criteria. With an alpha = .05 and sample size of 26 the achieved power was calculated to be 0.804 (GPower 3.1)
Figure 7. This bar graph shows the mean VSTM accuracy on trials where the absolute distracter orientation (left or right stimulus) was congruent or incongruent with the memory probe offset (required button press). This is within the 40 degree offset distracter condition. Participants performed significantly better when the absolute distracter orientation was congruent with the required button press.
Discussion
The main goal of this study was to critically analyse the findings by Soto & Silvanto (2012) and similar findings by Bona et al. (2013) and Soto & Silvanto (2014). This was done by assessing if the effects found by these studies were not attributed to artefacts of perceptual non-null sensitivity for the distracter, response bias or other measurement noise. A highly similar experimental paradigm was used to replicate the study by Soto & Silvanto (2012), a circular grating mask condition was used in addition to SDT to assess null-sensitivity effects and an exploratory analysis was done to analyse response bias and unwanted noise effects caused by trial randomisation.
Perceptual sensitivity for distracter
The difference between subjective and objective threshold methods was assessed by adding a circular grating mask condition and comparing perceptual sensitivity for the distracter in both masking conditions. In the black mask condition, used in the studies by Soto & Silvanto (2012;2014) and Bona et al. (2013), participants were perceptually sensitive to the distracter. Under the assumptions of the objective threshold method any effects found in this condition cannot be used to conclude a dissociation between VSTM and consciousness (Snodgrass, Bernat & Shivrin, 2004). Under the assumptions of the subjective threshold method, as used in Soto & Silvanto (2012;2014) and Bona et al. (2013), this non-null sensitivity does not prevent such conclusions. The circular grating mask condition showed perceptual null-sensitivity for the distracter. This means that for both perceptual threshold methods a circular grating mask can be used to enable researchers to contribute any found effects to a possible dissociation between VSTM and consciousness (Snodgrass, Bernat & Shivrin, 2004).
Main effects
The effects found by Soto & Silvanto (2012;2014) and Bona et al. (2013) were not replicated in this study. No effects for distracter presence and distracter visibility on VSTM performance were found for the two mask conditions. Subsequently no correlation effects between d’ value and VSTM impairment were found for the two mask conditions. This does contradict the main hypothesis which stated that it was expected that there would be a directly proportional relationship between d’ value and VSTM impairment. However, since there was no significant VSTM impairment caused by distracter presence nothing can be concluded about a possible relationship between d’ value and VSTM impairment.
There are multiple possible explanations for the lack of VSTM impairment effects. Firstly, the dissociation of VSTM and consciousness awareness found in Soto & Silvanto
(2012;2014) and Bona et al. (2013) could actually be contributed to weakly conscious perceptual effects for the distracter (Snodgrass, Bernat & Shivrin, 2004). This is plausible seeing that the d’ value in the black mask condition, in which a replication of the unconscious working memory findings was expected, was significantly lower than the d’ value in Bona et al. (2013). This could mean that even though the black mask condition in this study showed a mean d’ value of 0.6 which is significantly different from 0, this sensitivity is still too low to achieve VSTM impairment effects. The large mean d’ value of 1.53 in Bona et al. (2013) indicates that this is a possibility. This would mean that VSTM impairment effects only occur when distracter stimuli are sufficiently strong, and/or mask is sufficiently inefficient, to achieve significant perceptual sensitivity. Secondly, the lack of effects could be attributed to researcher instruction. In this study participants were instructed to focus on the VSTM task and informed that they would probably rarely perceive a distracter. This instruction could derive participant attention from the distracter presence identification task. This could result in less attention to the paradigm during distracter presentation. Seeing how participants did
show significant perceptual sensitivity to the distracter in the black mask condition, this explanation is less plausible. Lastly, the lack of effects might be attributed to hardware differences. Soto & Silvanto (2012) used a screen with a 800x600 resolution where this study used a screen with 1920x1080 resolution. Also, a different program was used to run the experiment. Soto & Silvanto (2012) used Psychology Software Tools Inc.( Pittsburgh, USA; http:// www.pstnet.com/eprime.cfm) where this study used the software package
Psychtoolbox-3 (http://psychtoolbox.org/) which was ran on MATLAB R2014a. It is unclear how these hardware differences could influence VSTM impairment effects.
Exploratory analysis
A general exploratory analysis was performed to assess response bias and unwanted noise effects in the unconscious working memory paradigm. A significant response bias effect was found showing that participants had a strong tendency to respond with a right button press compared to a left button press. This also had a significant effect on VSTM performance in left memory probe versus right memory probe trials, where performance in right memory probe trials was significantly better. This systematic noise should be accounted for in the analysis of data generated by this experimental paradigm.
Additionally unwanted noise caused by trial randomisation was studied. Figure 5 showed the large amount of trial sub-types in the incongruent distracter condition created by randomisation of stimuli for each trial. It was shown that there is a significant priming effect of absolute distracter orientation on VSTM performance in trials where distracter orientation randomisation could cause qualitatively different trials. These qualitatively different trial sub-types are regarded as the same in the general statistical analysis for this paradigm. This means that there are not only significant unwanted noise effects caused by trial randomisation, these noise effects are ignored in the general statistical analysis for this paradigm.
General limitations
The main limitation in this study, and other studies using the unconscious working memory paradigm, are caused by the design of the paradigm itself. Even in the absence of a VSTM impairment effect, there is systematic noise in the generated data. This study has shown that this noise at least consists of response bias effect and a priming effect which both effect VSTM performance. Both of these effects are caused by orientation randomisation of stimuli on a trial level. It could be argued that averaging VSTM performance scores filters out this noise, and therefore there would be little reason to research effects on a sub-type trial level. This however is an assumption that is not always valid (Savitzky & Golay, 1968) and could additionally, have problematic consequences for future research. For example, it is important to understand the effects caused by the distracter stimulus. Since this study found no effects of distracter presence on VSTM performance researchers could, based on the general statistical analysis in this paradigm, conclude that the distracter had no effects on the VSTM performance of participants. The found priming effect shows otherwise, the absolute orientation of the distracter compared to the memory probe orientation significantly effects VSTM performance. This gives insight into the working mechanisms behind unconscious working memory, as it enables researchers to compare the neurocognitive pathways involved in subliminal visual priming and subliminal VSTM interference (Bar & Biederman; 1998).
Apart from the importance of looking into effects caused by qualitatively different trial sub-types and the fact that these effects are normally overlooked, it is problematic that the unconscious working memory paradigm has so many different trial sub-types. This means it is very difficult to be constantly aware of possible unwanted noise effects. This study suggests that future research using the unconscious working memory paradigm uses a simplified version of this paradigm. Where there are less possible stimulus orientations of memory cue and distracter, reducing trial sub-types.
Data was collected by four students of the University of Amsterdam. Since each participant was given instructions by one of these four students, there may be differences in paradigm instructions which could alter performance per participant. The general quality of given paradigm instructions by these four students is also questionable, seeing as 7
participants had to be excluded based on poor performance on the no distracter condition. This indicates that at least 7 participants had a poor understanding of the paradigm.
The final limitation of this study is the lack of replication of Soto & Silvanto (2012;2014) and Bona et al. (2013). This prevents any conclusions on the relationship between distracter perceptual sensitivity and VSTM performance. And therefore also any conclusions about the plausibility of the claims by Soto & Silvanto (2012;2014) and Bona et al. (2013) about the dissociation between WM and consciousness, which was one of the aims of this study.
Final remarks and conclusion
Unconscious working memory effects which indicate a dissociation between VSTM and consciousness could be attributed to weakly conscious effects (Snodgrass, Bernat & Shivrin, 2004). Conclusions concerning a possible dissociation rest on assumptions about subliminal distracter perception which can be categorised under subjective and objective method assumptions (Schmidt, 2015). It is crucial for future research that there is consensus regarding which method is sufficient to use when researching unconscious working memory. This study showed that using a circular grating mask is adequate for achieving perceptual null-sensitivity for the distracter which makes this paradigm valid for use under both objective and subjective threshold method assumptions, where the black mask is only valid for use under subjective threshold method assumptions. This study could not give further insight in the validity of assumptions made under subjective and objective threshold methods
as there was no replication of the VSTM impairment effects found by Soto & Silvanto (2012; 2014) and Bona et al. (2013).
This study has shown that random generation of trial sub-types creates unwanted noise. This noise is generally ignored in studies using the unconscious working memory paradigm. The presence of this noise is problematic (1) as the assumption that averaging filters out this noise may not be valid, (2) the cognitive mechanisms behind the effects creating the noise may give important insight into the possible existence of unconscious working memory and (3) given the large amount of trial sub-types, it is difficult to have a clear overview of all possible sources of noise. This study suggests that the unconscious working memory paradigm gets simplified to make sure that any measured effects are indeed contributable to possible unconscious working memory effects.
To be able to validly show a possible existence of unconscious working memory, future research should aim to clearly state which perceptual threshold method assumptions are made and use a version of the unconscious working memory paradigm which does not violate these assumptions. Additionally a simplified version of the unconscious working memory paradigm should be used to reduce unwanted noise effects. In the case a non-simplified version of the unconscious working memory paradigm is used, unwanted noise effects should be explored and any conclusions based on found VSTM impairment effects should only be made after it is clear these effects are not attributed to unwanted noise caused by trial randomisation.
References
Baars, B. J. (1997). In the theatre of consciousness. Global workspace theory, a rigorous
scientific theory of consciousness. Journal of Consciousness Studies, 4(4), 292-309.
Baars, B. J., & Franklin, S. (2003). How conscious experience and working memory
interact. Trends in cognitive sciences, 7(4), 166-172.
Bar, M., & Biederman, I. (1998). Subliminal visual priming. Psychological Science, 9(6),
464-468.
Block, N. (2005). Two neural correlates of consciousness. Trends in cognitive sciences, 9(2),
46-52.
Bona, S., Cattaneo, Z., Vecchi, T., Soto, D., & Silvanto, J. (2013). Metacognition of visual
short-term memory: dissociation between objective and subjective components of
VSTM. Frontiers in psychology, 4, 62.
Cohen, J. (1992). A power primer. Psychological bulletin, 112(1), 155.
Dehaene, S., Changeux, J. P., Naccache, L., Sackur, J., & Sergent, C. (2006). Conscious, preconscious, and subliminal processing: a testable taxonomy. Trends in cognitive sciences, 10(5), 204-211.
Lamme, V. A. (2003). Why visual attention and awareness are different. Trends in cognitive sciences, 7(1), 12-18.
Overgaard, M., Timmermans, B., Sandberg, K., & Cleeremans, A. (2010). Optimizing
Sandberg, K., Timmermans, B., Overgaard, M., & Cleeremans, A. (2010). Measuring
consciousness: is one measure better than the other?. Consciousness and
cognition, 19(4), 1069-1078.
Savitzky, A., & Golay, M. J. (1964). Smoothing and differentiation of data by simplified least
squares procedures. Analytical chemistry, 36(8), 1627-1639.
Schacter, D. L., & Buckner, R. L. (1998). Priming and the brain. Neuron, 20(2), 185-195.
Schmidt, T. (2015). Invisible stimuli, implicit thresholds: Why invisibility judgments cannot
be interpreted in isolation. Advances in cognitive psychology, 11(2), 31.
Silvanto, J., & Soto, D. (2012). Causal evidence for subliminal percept-to-memory
interference in early visual cortex. Neuroimage, 59(1), 840-845.
Snodgrass, M., Bernat, E., & Shevrin, H. (2004). Unconscious perception: A model-based
approach to method and evidence. Perception & Psychophysics, 66(5), 846-867.
Soto, D., & Silvanto, J. (2014). Reappraising the relationship between working memory and
conscious awareness. Trends in Cognitive Sciences, 18(10), 520-525.
Stanislaw, H., & Todorov, N. (1999). Calculation of signal detection theory
