Top-Down Effects of Motivation on Perception
Anna van Vree, 10542940
University of Amsterdam
January 7th, 2018
Abstract
The top-down effects of motivation on perception were examined in this research with two
experiments. The first experiment was a replication of Raymond and O’Brien (2009), who found such top-down effects. Participants associated punishment or reward, with a low or high probability, with faces in a value learning task. The recognition of these and novel faces were tested in an attentional blink task. In contrast with the original study, reward-associated and high probability-associated faces were not recognized better and the associated value did not have an effect on the AB. The second experiment excluded the influence of memory, and the participants were asked to localize the value-associated faces in an attentional blink task. This yielded the same results as the
recognition AB task. There were no top-down effects of motivation on perception found in either experiment. It is unclear whether these results prove the non-existence of top-down effects or simply show methodological shortcomings. However, it does show that top-down effects are hard to find.
1 Introduction
How objective is our view of the world? Is the way we perceive the things around us formed by our biases, preconceived notions, beliefs and desires? Do we see what we want to see, what we expect to see? When looking at the influences of these higher-level cognitive functions on
perception, we are faced with a great amount of studies that are often contradictory. The line between interpretation and perception becomes blurry, making it harder to pin down higher-level effects on the earliest stages of perception. In other fields of cognition, like memory, higher-level influences are less ambiguous, accepted as mainstream science decades ago. A classic study by Festinger and Carlsmith (1959) shows something becomes more fun in retrospect when we have received less money for it. We did it for free, so it had to have been fun. We try to make sense of the world, and by doing so, we change and reshape memories to make them fit into established
concepts, beliefs and schemas. When our experiences are contradictory, we experience cognitive dissonance, and change the story so it makes sense in our head. A classic experiment on schemas in memory (Barlett, 1932) shows that when recalling a story, it becomes more and more like a standard fairytale as time goes on. We alter our memories based on what we have seen many times before. But this is not only something that happens in memories. When interpreting an image, it is
influenced by our history, beliefs and motivation. We pay attention to things we find important, things that confirm our biases, things that can benefit us. The images, sounds and feelings that eventually reach our consciousness are carefully selected, influenced by higher-level cognition like motivations, emotions and beliefs (Kane & Engle, 2002). But is this the case with perception? The visual information of the world around us must be translated into an image that we can observe, consciously or unconsciously, so we can act, react or reflect. Do higher level cognitive functions have influence on this visual process, working top-down to penetrate perception?
Some studies have found results that suggest that this is indeed the case. Milders, Sahraie, Logan, and Donnellon (2006) examined this idea by using the attentional blink (AB) paradigm, where a visual target within a rapid serial visual presentation is not or barely seen if another target is presented 400-700 milliseconds before it. They found that AB can be overcome when the second target was a fearful face. The same was true for faces that weren’t fearful, but were manipulated to be associated with fear. Anderson, Siegel, Bliss-Moreau and Barrett (2011) looked at the effects of negative associations on perception processing. They used the binocular rivalry paradigm, where two incompatible images are being shown, one to each eye. Only one image has perceptual dominance at a given time. The participants had to read gossip about different faces, thereby creating negative, positive and neutral associations. The researchers found that the faces related to negative gossip had
2 comparatively more perceptual dominance. Both these studies suggest a penetration of high-level cognition on perception.
The current study focuses specifically on how perception is influenced by motivation, which is an example of high level cognition. There have been results found confirming this influence, measuring perception with binocular rivalry. Wilbertz, van Slooten and Sterzer (2014) used binocular rivalry with two circular gratings in different colors. Participants took part in a learning task that associated one color with either winning or losing money. The color associated with gain had more perceptual dominance than the neutral color, and the color associated with loss had less perceptual dominance than the neutral color. Balcetis, Dunning and Granot (2012) used a similar design but the competing images were either letters or numbers, where one of the two categories was learned to be linked to either gain or loss. They found that the category linked to gain was more perceptually dominant than the neutral category, and the category linked to loss was less dominant than the neutral category. In addition, they found that when the participants played for a disliked other, these effects did not occur, suggesting that motivations and not just learned behavior are at play here. However, there are some problems with binocular rivalry when studying motivation and early perception. Primarily there is the subjective nature of the measurement of perceptual dominance within binocular rivalry. The participants must indicate themselves which of the two images they perceived, thus making the experiment more vulnerable to errors and biases. In addition, binocular rivalry can be, consciously or unconsciously, influenced when you’re dealing with motivation and loss/gain games. There is some control in choosing which of the two images you’ll see. When a participant notices this, they can choose to see the image that will gain them the most. The experiment is then not so much about perception as about attention and (unconscious) control.
The attentional blink paradigm can provide an operationalization with cleaner results, for when used in a task it is often performance based, therefore giving an objective view of the participant’s perception. Furthermore, when a performance-based AB task is used it excludes the possibility of the participant manipulating the experiment. Raymond and O’Brien (2009) researched the effect of motivation on the recognition of faces using the AB task. The participants first did a value learning task, where two faces were presented, and one had to be chosen to maximize the monetary winnings. To manipulate the valence of the faces, for some pairs the choice resulted in either a monetary gain or nothing, while for others it resulted in monetary loss or nothing. With the control pairs either choice resulted in nothing. Within each pair, one of the faces had an 80% chance to result in either reward or punishment, while the other had a 20% chance, thus manipulating the motivational salience of the faces. When the faces were all associated with different values by the participant, an attentional blink task was performed. The first target (T1) was either circles or
3 rectangles, the second target (T2) was either a face that was learned in the value learning task or a novel face. To manipulate the amount of attention that can be given to T2, and thereby the strength of the AB, half of the trials contained a short lag (200 milliseconds) and the other half a long lag (800 milliseconds) between T1 and T2. When analyzing the results of the long lag AB, it was found that faces with a high probability, thus with a high motivational salience, were recognized better than faces with a low motivational salience, regardless of valence. When comparing the short lag and long lag conditions, the valence did matter. For faces with negative or neutral valence, the recognition lessened dramatically in the short lag condition. The recognition of faces that were associated with positive valence was similar in the short and long lag. Thus, the recognition of a stimulus with positive valence seems unrelated to attention, or is at least capable to overcome the AB.
Although the results mentioned above suggest penetration of high-level cognition on perception, there have been some mixed findings. Firestone and Scholl (2016) published an article criticizing the general idea of top-down penetration on perception, and attacked a large number of studies. Firestone and Scholl believe there are no top-down effects, and explain the fact that some studies have found these results as faulty research, naming six pitfalls that are supposed to overturn each study finding penetration. Criticism can be leveled against the recognition based AB task used by Raymond and O’Brien, namely that the task is based mainly on memory, which is one of the pitfalls identified by Firestone and Scholl. It can be argued that when searching your memory to determine if the face is familiar or novel, higher level cognitive functions are already involved. As discussed before, we already know that memory is influenced heavily by higher level cognition, making these findings considerable less interesting and groundbreaking. For example, in the Raymond and O’Brien study, the faces related to high motivational salience may simply be better remembered, leading to a more accurate recognition than faces with low motivational salience. In addition, the positive valence stimuli may have overcome the AB because they were better memorized and thus had stronger connections. To minimize the interference of memory in the findings of top-down penetration, one can look at a localization task. With localization there is only detection at play instead of recognition, which should not involve memory. Rutherford ea. (2010) found that using value-laden faces as spatial cues impaired the spatial target, thus showing that motivation has an influence on visual orienting. This promises that localization tasks can be used within the attentional blink paradigm to study top-down effects on perception.
A difficulty in dealing with top-down effects on perception is the fact that it is a process, thus having multiple stages varying in penetrability and level of cognition. There is a certain ambiguity when talking about perception and visual processing. Where exactly in the perception process are we looking for top down influences? What are the precise stages of visual process? There is still very
4 little known about the perception process, and it is even harder to pinpoint where in the brain and at what times things happen. This makes it difficult to move towards a shared and consistent theory. The fact that some studies have found results that seem to imply top-down penetration while others have not may very well be because the attempted penetration took place in different stages of the perception process. Recognition of a face and detection of a face are two things that could be completely separated in the perception process, with detection occurring earlier than recognition. However, at this stage it is unknown if this is the case and if so, how far removed recognition and detection are and what happens before, after and in between the two within perception processing.
This study builds upon Raymond and O’Brien’s research, replicating it and expanding on it, excluding pitfalls and learning more about different stages of the perception process. In stage one the same value-learning task as in the Raymond and O’Brien study is used, where the faces all have their own valence (rewarded, punished or neutral) and motivational salience (80% or 20%
probability). In stage two the participants either do a recognition or localization task. Recognition in this study is operationalized identically to Raymond and O’Brien’s second task. T1 is a mosaic of either circles or rectangles, T2 either a face that was learned at the first task or a novel face. Half of the trials contain a short lag, the other half a long lag. We expand on Raymond and O’Brien’s
research by having a localization task, which is a modified AB task. This task will exclude the influence of memory on the perception process, thus giving a cleaner view of top-down penetration of
perception. Additionally, it will show an earlier point in the perception process in comparison to recognition, partly clarifying the hierarchy of the perception process. In the localization task three stimuli are presented to the participant. In the middle T1 appears, either being a rectangle of circle mosaic. T2 is a value positive, value negative, neutral or novel face, and appears either as the left or the right stimulus. The participants will indicate whether they saw a circles or rectangles, and whether they saw the face at the right or the left side. Half of the trials will contain a short lag and the others a long lag between T1 and T2.
We expect to find similar results as Raymond and O’Brien in the recognition experiment. The faces with higher motivational salience will be recognized more in the short lag as well as the long lag trials. The negative valence and neutral conditions’ recognition will lower from the long lag to the short lag, due to the AB. The positive valence condition will overcome the AB and have the same recognition rate in shot lag and long lag trials. We expect these results not only because it is a replication, but because recognition is heavily influenced by memory and it is established that memory is in turn heavily influenced by higher-level cognition like motivation. In the localization experiment, we expect to find a higher accuracy for faces with high motivational salience, and a similar accuracy for reward-associated, punishment-associated and neutral faces in the long lag
5 trials. If motivation does not penetrate the earliest stages of perception processing, we expect that all value conditions will fall victim to the AB and suffer in accuracy in the short lag condition in comparison to the long lag condition. If motivation penetrates the earliest stages of perception processing, we expect that only the negative valence condition and neutral valence condition will suffer from the AB and show lowered accuracy in the short lag in comparison to the long lag,
whereas there will be no difference between the short and long lag conditions in the accuracy of the positive valence condition. Thus, motivation created by faces with positive valence-association overcomes the attentional blink, showing top-down effects on perception.
Method
Perception was measured by the AB task, particularly by looking at the difference of performance when presenting a short lag or a long lag. In experiment 1 the AB performance measuring perception was based on recognition, in experiment 2 it was based on localization. Motivation was operationalized by associating values to faces, which were taught in the value learning task. Value of the faces was manipulated on two levels, valence and motivational salience, where valence was divided in three different conditions: reward-associated, punishment-associated and neutral, and salience was either high or low. With each face taking on both a valence and a salience characteristic, and the salience for neutral faces being irrelevant, there were five (3 x 2 – 1) types of faces used in the AB recognition task. In the AB localization task there was a sixth type: the novel one that did not appear in the value learning task.
Participants
There were 51 participants (37 females, 13 males, 1 other), their age ranging from 18 to 62 with a mean of 25.24 (SD = 9.82). There were 25 participants performing the AB recognition task and 26 participants performing the AB recognition task.
Materials Apparatus
The tasks were presented on an Optiplex 9010 computer running Matlab, with a 61-cm monitor (100-Hz refresh, 1920 x 1080 resolution).
Stimuli
In Figure 1 an example is shown of the faces used in the value learning task. There were 22 faces in gray-scale, all pulled from an existing dataset. For value learning 12 of 22 faces were used in
6 six different learning pairs, three pairs consisting of males and the other three consisting of females. The stimuli used for the first target (T1) in the AB tasks was a mosaic in green-scale of circles or rectangles (see Figure 2). The original 12 faces used in the value learning task in addition to 10 novel faces in the recognition AB task, and 4 novel faces in the localization AB task were used for the second target (T2). There were 154 mask stimuli that were created by scrambling the original 22 faces by cutting up each face into sets of squares and randomly rearranging them. There were 7 different scramble methods for each face, depending on the number of squares taken horizontally (between 3 and 5) and vertically (between 5 and 8).
Figure 1. Face pair used in the value Figure 2. Mosaic of circles used as T1 in AB
learning task task
Procedure
Value Learning Task
All 51 participants took part in the value learning task, consisting of 360 trials, with short breaks after 120 and 240 trials. In each trial the participant was presented with two faces, one on the left and one on the right side, with a fixation cross in the middle (see Figure 1). The participant chose one of the two faces, and was informed on the screen if they had won money, lost money or if nothing had happened. They were also informed of the total of what they had earned so far. The faces were divided up in six pairs, with two pairs being reward-related, two being punishment-related and two being neutral. The location of a face being either left or right was randomized each trial. Within the punishment-related and reward-related pairs, one face had a .80 probability for reward or punishment and the other had a .20 probability. This resulted in faces having a different valence (either being reward-related, punishment-related or neutral) and a different motivational salience (either having a high or a low probability). Association to valence and salience for each face was counterbalanced between subjects.
7 Attentional Blink Task
Experiment 1: Recognition. After the value learning task, 25 participants took part in the AB task. In
the first experiment the AB task was recognition-based. First there were five practice trials to make sure the participants could accurately identify T1. There were 440 legitimate trials, with short breaks after 110, 220 and 330 trials. In each trial, stimuli appeared in rapid succession, each stimulus being visible for 100 milliseconds. T1 was the first stimulus and appeared as a green mosaic of either circles or rectangles. Afterwards a mask stimulus, a scrambled face, was shown. Twelve of the faces shown as T2 were learned in the value learning task, thus being a value laden faces. The other ten faces shown as T2 were novel faces that the participants had not learned before. T2 was shown after either a short or a long lag. With a short lag T2 appeared 200 milliseconds after T1, with a long lag T2 appeared 800 milliseconds after T1. Following T2 a mask stimulus appeared. After each trial the participant had to indicate, without a time limit, if they had seen circles or rectangles. They did this with their left hand, using key 1 for circles and key 2 for rectangles. If they answered correctly, the fixation cross turned green, and if they answered incorrectly, it turned red. After that they had to indicate whether the face was old or new. They did this with their right hand, using the left arrow if the face was old and the right arrow if the face was new. There was no feedback for T2.
8
Experiment 2: Localization. In the second experiment AB task was based on localization. Again, there
were five practice trials, followed by 384 legitimate trials with breaks after trial 128 and trial 256. A trial consisted of a rapid serial visual presentation. Every item, or frame, appearing in the serial visual presentation, appeared for 90 milliseconds, with 24 frames in total. A frame was made up of three images next to each other, all of them being either distractors, T1 or T2. The distractors were scrambled faces, as seen in the top image of Figure 3. T1 was a green mosaic of either circles or rectangles. T2 was a face that were either learned in the value learning task or a novel face that had not appeared in the value learning task. The first frames consisted of scrambled faces for all three images. In frame eight, nine, ten, eleven or twelve, T1 appeared as the middle image. After either two frames (short lag) or eight frames (long lag) T2 appeared, either as the left or the right image. At the end of each trial the participant indicated whether they saw circles or rectangles with their left hand, pressing the 1 key if they saw circles and the 2 key if they saw rectangles. If they answered correctly the fixation cross turned green, if they answered incorrectly it turned red. After that they indicated with their right hand whether they saw the face on the left side by pressing the left arrow, or the right side by pressing the right arrow. Again, there was no feedback for T2.
Likeability Rating and Interview
After the AB-task the participant rated all 22 faces on likeability. They indicated this on a scale of one to seven, ranging from unlikeable to likeable, using the keyboard numbers. There was no time pressure. Lastly the participants were asked to answer, on paper, the following three questions (originally in Dutch): 1. What did you think was the purpose of the experiment? 2. What was your strategy to maximize your gain in task 1? 3. What was your strategy to be as accurate as possible in task 2?
Results Value Leaning Task
All 51 participants took part in the value learning task and all data was included in the analysis. Performance was measured in three components: performance on the reward condition, which was the mean of accurate picks within the reward pairs, performance on the punishment condition, which was the averaged accuracy within the punishment pairs, and performance on the neutral condition, where a random face was chosen, making performance around chance. Time was measured in six bins, with each bin containing 60 trials.
9 Figure 4. Performance on value learning task for the different face pairs per bin
To analyze the increase of performance over time, a Factorial Repeated Measures ANOVA with the factors bin (one through six) and pair value type (reward, punishment or neutral) was used.
Participants performed increasingly better across bins within the punishment and the reward pairs, F(3.77,176.35)=32.33, p<.001. As seen in Figure 4, the reward and punishment conditions
accelerated similarly, thus there were no significant results found for condition or condition and bin interaction. Because a random face was chosen as the correct answer for the neutral face pairs, the accuracy is expected to average around chance, in this case 48.1% (SE = 0.6%). In the final bin, performance was 76.1% (SE = 2.9%) for punishment and 75.8% (SE = 3.5%) reward face pairs.
Attentional Blink Recognition
Twenty-five participants completed the AB Recognition task. To analyze whether the attributed value of the faces had an influence on the AB experienced in the task and the T2 performance on the task, a Factorial Repeated Measures ANOVA was used. It tested the main and interaction effects of the factors lag (long and short), valence (reward and punishment) and salience (low probability and high probability) on T2 hit rate. There was a significant main effect for lag, F(1,23)=10.10, p=.004, with the short lag having a lower hit rate than the long lag. This means the attentional blink paradigm was effective. There were no significant results found for valence, F<1, p=n. s., which means there was no difference between reward-associated faces and punishment-associated faces for accurately recognizing the faces across lags. Additionally, there were no
10 significant results found for salience, F(1,23)=1.84, p=.189, meaning that faces that were associated with a high probability had the same hit rate as the faces associated with low probability. There were no interaction effects found, both F<1, p=n. s., which demonstrates that the strength of the
attentional blink was not influenced by the value of the stimuli. Additionally, a Factorial Repeated Measured ANOVA was used to examine whether faces with a particular motivational value were remembered better than others, and if value had an effect on the attentional blink. The effects of the factors lag (long lag and short lag) and motivational value (high probability reward, low probability reward, high probability punishment, low probability punishment and neutral) on the T2 hit-rate was analyzed. There was neither a main effect of value, F<1, p=n. s., nor an interaction effect with lag found, F<1, p=n. s., on recognition hit rate, as can be seen in Figure 5. This means there was no effect of value on performance or on attentional blink strength in the AB recognition task.
Figure 5: T2 hit-rate in AB recognition task for long lag and short lag per value, high punishment (-.80), low punishment (-.20), neutral (0), low reward (.20) and high reward (.80)
Attentional Blink Localization
Twenty-six participants completed the AB Recognition task. One participant performed below chance on T1 and was excluded from the analysis. To analyze the effect of value on localization performance and the interaction with the attentional blink, a Factorial Repeated Measures ANOVA was used, analyzing the effect of the factors valence (reward and punishment), salience (high probability and low probability) and lag (long lag and short lag) on T2 accuracy. The attentional blink was effective, with the long lag condition having a higher accuracy than the short lag condition, F(1,25)=36.60, p<.001. There was no significant result for valence, F(1,25)=1.59,
11 p=.219, meaning there was no distinction between reward-associated and punishment-associated faces when accurately localizing the face. There was no significant result for salience, F<1, p=n. s., thus no difference in T2 accuracy for high probability and low probability faces. There were no significant interaction results found between either two of the three factors, all F<1, p=n. s., indicating neither salience nor valence influenced the strength of the attentional blink. To further examine the effects of value, the faces were split in five categories and separately analyzed, examining the effect of the factors value (high probability reward, low probability reward, high probability punishment, low probability and neutral) and lag (long and short) on T2 accuracy. There was no significant main effect of value, F<1, p=n. s., and no significant interaction effect of value and lag found, F<1, p=n. s, as can be seen in Figure 6. This further stipulates that there was no influence of value-association on the T2 performance or on the attentional blink itself. There was an additional analysis done to examine if the familiarity of faces influenced the T2 accuracy and strength of the AB. Neutral faces were used to exclude influence of value. To analyze the effect of the factors familiarity (neutral and novel) and lag (long lag and short lag) on T2 accuracy, a Factorial Repeated Measures ANOVA was conducted. There were no significant results found, all F<1, p=n. s., suggesting that familiarity of a face had no influence on T2 accuracy or the strength of the attentional blink.
Figure 6: T2 accuracy in AB localization task for long lag and short lag per value, high punishment (-.80), low punishment (-.20), neutral (0), low reward (.20) and high reward (.80)
12 Control Analysis: Likeability Rating
To test whether faces that were associated with different values became more or less likable to the participants, two Factorial Repeated Measures ANOVA’s were used to examine the effect of value on the likeability attributed by the participants at the end of the experiment on a scale from 1-7. First the main and interaction effects of the factors valence (reward and punishment) and salience (high and low probability) on likeability rating were tested. There were no significant results for valence, F(1,49)=1.54, p=.221, nor for salience or for interaction of valence and salience, both F<1, p=n. s. There were similarly no effects found for the factor value (high reward, low reward, high punishment, low punishment and neutral) on likeability, F<1, p=n. s. The associated value of the faces had no influence on the likeability of the faces. Lastly, there was a Factorial Repeated Measures ANOVA conducted analyzing the effect of familiarity (novel and neutral) on likability, with no
significant results, F<1, p=n. s., suggesting that familiarity did not influence the likeability of the faces.
Control Analysis: Face Exemplars
AB Recognition Performance To make sure differences between the faces aside from
value-association did not have an effect on the results of the main analysis, the T2 hit rate per face in the AB recognition task was examined with an Factorial Repeated Measures ANOVA. There was a main effect found for faces, F(1,168.02)=4.08, p<.001, indicating that some faces were always recognized more or less, regardless of the associated value. There was no significant interaction effect with lag, F<1, p=n. s., indicating that no face was so recognizable it influenced the strength of the attentional blink. As seen in Figure 7A, there are two faces in particular that had a higher hit rate across lags compared to the other faces, namely face 4, as depicted in Figure 7B, averaging an 81.0% (SE=3,6%) hit rate, and face 8 (Figure 7C), averaging an 84,4% (SE=3,8%) hit rate. This suggest differences between faces that may have impacted the results found in the main analysis.
13 Figure 7. A. shows the averaged T2 AB Recognition performance per face across lags. Face 4 (B) and face 8 (C) had a particularly high hit rate
AB Localization Performance The same Factorial Repeated Measures ANOVA was conducted to analyze the T2 accuracy for each face in the AB localization task. There was no effect of face exemplars, F(15,375)=1.48, p=.110, nor an interaction effect of face exemplar and lag, F<1, p=n. s., indicating that all faces were localized equally well and had a similar reaction to the AB.
Likeability Rating To further test inherit differences between the faces, the likeability rating per face was examined to determine whether some faces were inherently more or less likeable. This was indeed the case, F(10.13,496.52)=6.85, p<.001. The average rating per face is shown in Figure 6. To determine whether the inherent likeability of the faces and the differences found in the AB
Recognition T2 performance were related, a Pearson Correlation was conducted. There was no significant correlation between likeability rating and T2 hit-rate across lags found, r=.069, p=.799, for the sixteen face exemplars. This suggests that the facial differences influencing the T2 performance were not related to likeability, and more likeable faces were not necessarily recognized better than faces with low likeability.
14
Figure 8. The average likeability rating per face across all participants, with a significant difference between faces
Control Analysis: Value Learning Performance
To examine if the performance on the value learning task had any influence on the AB task performance, it is important to first extract the main effect that was found in the experiment of Raymond and O’Brien and was hypothesized in this experiment. This is the finding that faces with a positive valence overcome the attentional blink, having the same T2 performance in the long and the short lag and thus having no lag effect. To calculate this, the lag effect (long lag – short lag) of the reward-associated faces was subtracted from the lag effect of the average of the punishment-associated faces and the neutral faces, thereby creating a lag effect difference variable. This was done for the AB recognition task and for the AB localization task and was correlated, per participant, with the amount of money they made in the value learning task. There were no significant results for AB recognition, r=-.161, p=.442, nor for AB localization, r=-.019, p=.927. There was similarly no significant correlation between the lag effect difference and the value learning performance for reward face pairs for AB recognition, r=-.127, p=.547, nor for AB localization, r=.010, p=.960. Lastly there was no correlation between the lag effect of rewarded faces and the value learning
performance for rewarded faces for AB recognition, r=.008, p=.970, nor for AB localization, r=.370, p=.063. This suggests that the performance on the value learning task did not have any influence on the strength of the attentional blink with reward-associated faces.
15 Discussion
The goal of this paper was to determine whether high-level cognitive functions have top-down effects on perception, and more specifically if motivation penetrates (early) perception. We know that high-level cognition like motivation influences interpretation and memory, but research is inconclusive if this is true for perception. We replicated a study that had found these top-down results, and built upon it by creating an experiment excluding memory. Both experiments used value learning with faces and an attentional blink task. There were no top-down effects of motivation on perception found in this experiment. Based on the established top-down effects of motivation on memory and the earlier findings of Raymond and O’Brien, whose study we replicated, we
hypothesized that we would find similar penetration of motivation on the attentional blink
recognition task. As a main effect, we expected to find that the recognition hit-rate for punishment-associated stimuli and neutral stimuli would drop dramatically from the long lag to the short lag, whilst the recognition hit-rate for reward-associated stimuli would remain constant across lags, thereby overcoming the attentional blink. We did not find any effect of value on hit-rate, meaning that reward-associated, punishment-associated and neutral faces were recognized the same amount, and were similarly affected by the attentional blink. Additionally, Raymond and O’Brien found a general difference between the high probability and low probability stimuli in the long lag condition, with high probability faces having a higher recognition hit-rate regardless of valence. We
hypothesized that we would find similar results. This was not the case, with salience having no influence on hit-rate in either lag condition.
The localization experiment was added to exclude the influence of memory on top-down penetration and to examine top-down effects at an earlier stage of perception. If motivation indeed penetrated early perception, we expected to find an effect of value on the T2 accuracy in the attentional blink localization task. Specifically, we expected to find that accuracy for punishment-associated and neutral stimuli would drop from the long lag to the short lag, falling victim to the attentional blink. In contrast, the accuracy for reward-associated stimuli would remain the same across lags, overcoming the attentional blink. If motivation did not penetrate perception, we expected to find no effect of value on T2 accuracy, with reward-associated, punishment-associated and neutral stimuli all falling victim to the attentional blink in a similar manner. We found results that support the second hypothesis, with no effects of valence and salience on the T2 accuracy or on the strength of the attentional blink. However, it is difficult to conclude that this means that there are no top-down effects on perception, because we did not get the results in the recognition experiment we expected.
In short, although the AB recognition experiment was a replication, we found conflicting results, indicating that motivation did not penetrate (memory-aided) perception in our experiment.
16 Additionally, and logically following, we found no results suggesting top-down effects of motivation on early perception processing in the AB localization task. Although contrary to research discussed earlier that found top-down effects, our results are in line with other failed replicas that didn’t find any top-down effects of high-level cognitive functions on perception (Rabovsky, Stein & Rahman, 2016 and Stein, Grubb, Bertrand, Suh & Verosky, 2017). This could mean that there are in fact no top-down effects of motivation on perception, not even when the penetration is guided by memory, making Raymond and O’Brien’s results false positives. The fact that they, and other researchers previously mentioned found top-down results, could be explained by the pitfalls Firestone and Scholl (2016) laid out, like the demand and response bias and low-level differences. Additionally, Theeuwes, Reimann and Mortier (2006) propose that bottom-up priming is often mistaken for top-down effects. It could be that this was the case for the previously mentioned studies finding top-down results.
Our lack of results could also indicate that high level penetration on perception is possible but very rare, making it hard to replicate results. However, it is also possible that we found alternative results because the study was not an exact replica, and the difference of result is a difference of methodology. One such difference is the faces used in the experiment. Raymond and O’Brien used computer-generated faces to minimize the effect of low-level differences whereas this study used photographs of faces. In the analysis per face exemplar it became clear that some faces were recognized better than others, regardless of value. However, this effect was not found in the localization condition, suggesting that this difference in hit-rates is not related to low-level
differences, but to interesting or pleasant facial characteristics that were simply remembered better. This is supported by the debriefing questions the participants answered, where some indicated that they picked faces in the value learning task based on how kind they looked (eleven participants) or by their remarkable facial features (six participants). This could be avoided in further research by using computer-generated faces.
Another explanation for the lack of results could be that while Raymond and O‘Brien used 600 trials in their value learning task, we used 360 trials. This could mean that the value associations were not learned enough, explaining why they did not have any influence on the AB task
performance. However, this seems unlikely for several reasons. First, the accuracy in the last bin of trials was similar to the accuracy in the last bin in Raymond and O’Brien’s study, suggesting that the associations were learned well enough. Additionally, because there was a difference in the
memorability of the faces, one would assume that the value learning task would be easier. This would explain why the associations were learned faster than in the Raymond and O’Brien task, but it would also suggest that the associations would be better remembered in the AB tasks. Lastly, if a shorter value learning task was indeed the problem, it would be assumed that a shorter value learning period causes lower association learning, which in turn causes the lack of effect of these
17 value associations on the AB task. This means that lower association learning correlates with the effect of the value associations on the AB task. However, we did not find any correlation between value learning performance and strength of the attentional blink in the reward condition. Therefore, it seems unlikely, although not impossible, that the lack of results in this study are explained by a shorter value learning period.
A shortcoming in the attentional blink localization task was the lack of a fixation cross or specific instruction that participants should keep looking at the middle image. Ten out of 25
participants in the localization AB task indicated that their strategy for localizing the face was to only look at one side, and if they did not see anything they would assume the face was on the other side. Because of this, they did not see half of the faces presented in the task. It is unlikely that this is the reason why there were no results in the AB localization task, seeing as the AB recognition task similarly yielded no results and did not have this issue. However, it could explain why the facial differences influencing the recognition task were not found in the localization task: because they were simply not seen. This problem could be avoided in further research by a fixation cross.
It remains unclear if high-level cognitive functions like motivation penetrate perception processing. Although our results suggest that there are no top-down effects, it is contradictory with earlier research, including the one we were replicating. It is possible that our lack of results is because of methodological short comings, like a difference in the memorability of faces. In either case, this study shows that top-down effects of motivation on perception are hard to find, and studies like Raymond and O’Brien’s are not easy to replicate. Whether that means that top-down effects are non-existent, possible but rare, or simply not found in this experiment, may be determined by further research.
References
Anderson, E., Siegel, E. H., Bliss-Moreau, E., & Barrett, L. F. (2011). The visual impact of gossip. Science, 332(6036), 1446-1448.
Balcetis, E., Dunning, D., & Granot, Y. (2012). Subjective value determines initial dominance in binocular rivalry. Journal of Experimental Social Psychology, 48(1), 122-129.
Bartlett, F. C. (1932). Remembering: An experimental and social study. Cambridge: Cambridge University.
Festinger, L., & Carlsmith, J. M. (1959). Cognitive consequences of forced compliance. The Journal of Abnormal and Social Psychology, 58(2), 203.
18 Firestone, C., & Scholl, B. J. (2016). Cognition does not affect perception: Evaluating the evidence for" top-down" effects. Behavioral and brain sciences, 39.
Kane, M. J., & Engle, R. W. (2002). The role of prefrontal cortex in working-memory capacity, executive attention, and general fluid intelligence: An individual-differences perspective.
Psychonomic bulletin & review, 9(4), 637-671.
Milders, M., Sahraie, A., Logan, S., & Donnellon, N. (2006). Awareness of faces is modulated by their emotional meaning. Emotion, 6(1), 10.
Rabovsky, M., Stein, T., & Rahman, R. A. (2016). Access to awareness for faces during continuous flash suppression is not modulated by affective knowledge. PloS one, 11(4).
Raymond, J. E., & O'Brien, J. L. (2009). Selective visual attention and motivation: The consequences of value learning in an attentional blink task. Psychological Science, 20(8), 981-988.
Rutherford, H. J., O'Brien, J. L., & Raymond, J. E. (2010). Value associations of irrelevant stimuli modify rapid visual orienting. Psychonomic Bulletin & Review, 17(4), 536-542.
Stein, T., Grubb, C., Bertrand, M., Suh, S. M., & Verosky, S. C. (2017). No impact of affective person knowledge on visual awareness: Evidence from binocular rivalry and continuous flash suppression. Emotion, 17(8), 1199.
Theeuwes, J., Reimann, B., & Mortier, K. (2006). Visual search for featural singletons: No top-down modulation, only bottom-up priming. Visual Cognition, 14(4-8), 466-489.
Wilbertz, G., van Slooten, J., & Sterzer, P. (2014). Reinforcement of perceptual inference: reward and punishment alter conscious visual perception during binocular rivalry. Frontiers in psychology, 5.
