Consciousness and Cognition 27 (2014) 42–52
Contents lists available at ScienceDirect
Consciousness and Cognition
journal homepage: www.elsevier.com/locate/concog
Semantic priming revealed by mouse movement trajectories
Kunchen Xiao ⇑, Takashi Yamauchi
Department of Psychology, Texas A&M University, Mail Stop: 4235, College Station, TX 77843-4235, USA
a r t i c l e
i n f o
Article history:
Received 23 July 2013
Available online 4 May 2014
Keywords:
Semantic priming
Congruency effects
Mouse movement
a b s t r a c t
Congruency effects are taken as evidence that semantic information can be processed automatically. However, these effects are often weak, and the straightforward association
between primes and targets can exaggerate congruency effects. To address these problems,
a mouse movement method is applied to scrutinize congruency effects. In one experiment,
participants judged whether two numbers were the same (‘‘3n3’’) or different (‘‘3n5’’), preceded by briefly presented pictures with either positive or negative connotations. Participants indicated their responses by clicking a ‘‘Same’’ or ‘‘Different’’ button on the
computer screen, while their cursor trajectories were recorded for each trial. The trajectory
data revealed greater deviation to unselected buttons in incongruent trials (e.g., ‘‘3n5’’ preceded by a green traffic light picture). This effect was influenced by the type of responses
but not by prime durations. We suggest that the mouse movement method can complement the reaction time to study masked semantic priming.
Ó 2014 Elsevier Inc. All rights reserved.
1. Introduction
Semantic priming has been studied for decades, and congruency effects are often taken as evidence that complex semantic processing occurs automatically. For example, in a semantic priming study, participants were instructed to judge whether
two numbers were the same or different (e.g., ‘‘3 and 5’’ or ‘‘3 and 3’’), preceded by two masked priming letters (e.g., ‘‘A and
a’’ or ‘‘A and g’’). Trials where the information of primes and targets conflicted were called incongruent trials (e.g., ‘‘A g’’
prime and ‘‘3 3’’ target); those trials that did not conflict were called congruent trials (e.g., ‘‘A a’’ prime and ‘‘3 3’’ target).
It was found that the reaction time in incongruent trials was longer than that in congruent trials, indicating that a ‘‘same/
different’’ relationship processed subliminally for priming letters could influence the ‘‘same/different’’ judgment for numbers
(Opstal, Gevers, Osman, & Verguts, 2010).
However, semantic congruency effects found in many other studies are often weak and difficult to replicate, especially for
masked priming (Van den Bussche, Van den Noortgate, & Reynvoet, 2009). In addition, a straight forward mapping between
primes and targets undermines the reliability of claims that priming occurred at a semantic level, because participants could
apply the same type of judgment to both the target number pairs and priming letter pairs. For example, in a picture priming
study (Kiesel, Kunde, Pohl, Berner, & Hoffmann, 2009), participants judged whether a piece of chess was a configuration of
check or non-check, preceded by a prime, which was also a configuration of check or non-check. In this case, a learned stimulus–response mapping could facilitate the processing of primes, therefore the congruency effects are exaggerated since the
semantic relation between primes and targets was straightforward. It is found that congruency effects are modulated by the
⇑ Corresponding author.
E-mail addresses: [email protected] (K. Xiao), [email protected] (T. Yamauchi).
http://dx.doi.org/10.1016/j.concog.2014.04.004
1053-8100/Ó 2014 Elsevier Inc. All rights reserved.
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
43
Fig. 1. Illustration of the area under the curve (AUC). In this example, a participant judges whether the numbers 3 and 5 are the same or different. The curve
represents a hypothetical trajectory of a cursor from the onset position (‘‘START’’ button) to the ending position (clicking the ‘‘Different’’ button). The
straight line represents the ‘‘ideal path’’ between the onset and ending position. The AUC is defined by the area circumscribed by the ideal path and the
actual trajectory curve that exceeds the ideal path toward the unselected option (shaded area), and is measured by pixels.
semantic relatedness between primes and targets, and are more robust when the prime-target relatedness is high (Ortells,
Marí-Beffa, & Plaza-Ayllón, 2012). To mitigate these problems, the current study introduced two new procedures.
First, we applied a mouse movement method to reveal semantic processing. The merit of the mouse movement method is
that it records dynamic temporal-spatial information about participants’ responses, in addition to the response time data
(Freeman & Ambady, 2010). Early works (Aglioti, DeSouza, & Goodale, 1995) show that hand movement can reflect one’s
visual perception without awareness: participants automatically adjusted their fingers when pointing to a vibrating object,
but were unaware of adjusting their hand motion. Recent studies suggest that the temporal-spatial pattern of hand movement can reveal hidden cognitive states (Song & Nakayama, 2009). By analyzing the temporal-spatial features of mouse cursor trajectories, more details and further insights can be gained to understand congruency effects. However, only few studies
have applied the mouse movement method to study priming. A notable exception is one by Friedman and Finkbeiner (2010),
which found that the repetition priming and semantic priming could be distinguished by different cursor trajectory patterns.
It is interesting to investigate whether the mouse movement method can complement reaction time techniques as an effective tool to measure semantic priming effects.
Recent studies shows that the trajectory of a cursor in trials with conflicting information was attracted to an unselected
option; for example, participants were instructed to judge whether a face belonged to a white or black man, and the cursor
trajectories were attracted to the option ‘‘white’’ when an atypical black face was presented, and vice versa (Freeman,
Ambady, Rule, & Johnson, 2008). The magnitude of attraction is measured by the area under the curve (AUC), which is calculated as the geometric area circumscribed by the straight line from the onset position to the ending position and the actual
trajectory that veers toward the unselected option. In the current experiment, participants judge whether two numbers are
the same or different, by clicking one of the two buttons (i.e., ‘‘Same’’ or ‘‘Different’’) on the top of the computer screen
(Fig. 1).
Second, to make the relation between primes and targets more abstract, we replace letter primes (e.g., ‘‘A a’’) with symbolic pictures with either positive (e.g., ‘‘go’’) or negative (e.g., ‘‘no go’’) connotations but still use the number pairs as targets
(Fig. 2). Participants make same/different judgments for number pairs (e.g., ‘‘3n5’’), whereas the priming pictures are not
directly linked to semantic meanings as ‘‘same’’ or ‘‘different’’. By replacing the primes, we enhance the complexity of
prime-target associations. It is expected that the congruent/incongruent relationships between primes and targets should
influence the number judgment if the complex semantic relationship linking primes and targets is processed. That is, when
responding to incongruent trials (e.g., positive primes followed by 3n5 or 5n3, or negative primes followed by 3n3 or 5n5), the
cursor trajectories should show a greater attraction (AUC) towards the unselected option, as compared to congruent trials.
In psychophysics studies that investigated perceptual learning (Schoups, Vogels, & Orban, 1995), the same/different judgments of stimuli were shown to respectively correspond to ‘‘yes’’ or ‘‘no’’ responses. It is well known that positive (‘‘yes’’)
responses take shorter reaction time than negative (‘‘no’’) responses (Sternberg, 1966). A similar trend is present in same/
different judgment tasks: ‘‘same’’ responses result in shorter RT than ‘‘different’’ responses (Ratcliff, 1985). According to
Proctor’s Unified Theory (1981), such reaction time differences indicate that ‘‘same’’ and positive judgments employ an analogous processing mechanism, as distinct from the process underlying ‘‘different’’ or negative judgments. Following this reasoning, we assume that our priming pictures with positive connotations (e.g., a green traffic light) are congruent with ‘‘same’’
responses (e.g., ‘‘3n3’’) while primes with negative connotations (e.g., a red traffic light) correspond to ‘‘different’’ responses
(e.g., ‘‘3n5’’). Accordingly, we predict that positive primes facilitate ‘‘same’’ responses and negative primes facilitate ‘‘different’’ responses; in contrast, negative primes impede ‘‘same’’ responses and positive primes impede ‘‘different’’ responses.
Translating these prime-target relationships into mouse trajectories, we expect that larger AUCs arise in incongruent trials, compared to congruent ones. If such an indirect prime-target association still yields priming effects, it can be suggested
that relatively complex relationships between primes and targets can be rapidly processed.
44
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
Fig. 2. Examples of (a) congruent and (b) incongruent trials. In the current experiment, participants were asked to judge whether the target numbers were
the same or different, preceded by a briefly presented picture with either positive (e.g., a green light) or negative (e.g., a red light) connotations. For example,
trials where the prime was a green light and the target was ‘‘3n3’’, or the prime was a red light and the target was ‘‘3n5’’, were considered as congruent trials.
These prime-target relationships were swapped in incongruent trials. (For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)
2. Materials and methods
The experiment consisted of two phases, a number judgment phase and a following awareness test phase. In the number
judgment phase, participants judged whether two numbers were same or different, preceded by a briefly presented picture
(Fig. 3a). The trials in the awareness test phase were identical to those in the number judgment task, but participants were
explicitly instructed to identify the prime and choose the correct prime from two options (Fig. 3b).
2.1. Participants
A total of 388 undergraduate students from Texas A&M University were recruited for this study. Six of them were volunteers assigned to an initial pilot study testing our experimental procedures. The remaining 382 participants, who participated in the experiment for course credits, were randomly assigned to four conditions of prime duration: 20 ms (n = 94),
50 ms (n = 95), 100 ms (n = 97), and 150 ms (n = 96).
2.2. Materials
Prime stimuli were three pairs of symbolic pictures (Fig. 4). Each pair consisted of one with positive connotations and the
other with negative connotations. Four number pairs were used as target stimuli; two of them required ‘‘same’’ responses
(i.e., ‘‘3n3’’ or ‘‘5n5’’), and the other two required ‘‘different’’ responses (i.e., ‘‘3n5’’ or ‘‘5n3’’).
2.3. Apparatus and procedure
A 180 Hz monitor was used to present the stimuli. In each trial, a fixation cross was presented at the center of the screen
for 300 ms. Then, a pre-mask was presented for 100 ms, followed by a priming picture, then a post-mask for 100 ms. Finally,
the target was presented until participants responded. The prime (Fig. 4) was presented either for 20, 50, 100, or 150 ms,
depending on the condition to which participants were assigned. The software developed by Freeman and Ambady
(2010) was employed to present the stimuli and collect the mouse movement data.
The participants were instructed to judge whether the two numbers were the same or different, using a mouse to click the
‘‘Same’’ or ‘‘Different’’ button on the top of the screen (Fig. 3a) quickly and accurately, while ignoring any pictures flashed
prior to the numbers.
In the mouse-movement analysis, our response time measure was defined as the duration of time from the onset of a
target until participants clicked the ‘‘Same’’ or ‘‘Different’’ button. The AUC of a mouse movement trajectory in each trial
is calculated as the geometric area between a straight line from the onset position to the ending position and the actual trajectory that exceeds the straight line toward the unselected option (Fig. 2). Smaller AUCs indicate that the trials are easier to
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
45
Fig. 3. Illustrations of (a) the number judgment task and (b) the awareness task. (a) In the number judgment task, participants judged if two numbers were
the same or different preceded by a briefly presented prime (e.g., a check). (b) In the awareness phase, participants were asked to identify the preceding
prime and click the correct picture in the two options.
Pair
No.
Prime type
Positive Negative
Mask
Pair 1
Pair 2
Pair 3
Fig. 4. Pictures used as primes and masks in this experiment. Each picture is associated with either positive or negative connotations.
respond to, while larger AUCs more difficult (Freeman et al., 2008). To draw the trajectory, the position of the cursor is
recorded as one data point every 13–16 ms (Fig. 5), then all the data points are normalized into 101 steps for each trial using
a linear interpolation method.(see Fig. 6).
After the number judgment phase, participants carried out an awareness test. The trials in the awareness test were identical to those in the number judgment task. However, participants were informed about the prime and asked to identify it by
clicking one of the two options (Fig. 3b). There were 96 trials in the awareness test. The d0 measure obtained from the awareness test was applied to examine the extent to which the visibility of the primes influenced the magnitude of the priming
effects. Specifically, selecting the option that was presented as a prime was regarded as ‘hit’ and incorrectly selecting the
same option that was not actually presented was regarded as ‘false alarm’. (Macmillan & Creelman, 1991).
46
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
Fig. 5. Two examples of cursor trajectories for (a) an incongruent trial and (b) a congruent trial. The figure (a) shows a trajectory with a large attraction
towards the unselected option (large AUC), while the figure (b) shows a trajectory with a small AUC. The pixels on the screen were rescaled with the left
edge of the screen valued as 1.0, the right edge valued as +1.0, the lower edge valued as zero, and the upper edge valued as +1.5.
2.4. Design
The experiment was conducted with a 4 (prime duration: 20 ms, 50 ms, 100 ms, 150 ms; between-subjects) 2 (prime
type: positive, negative; within-subjects) 2 (target type: same, different; within-subjects) factorial design. For actual data
analysis, we applied two sets of ANOVA. In one set, we collapsed prime type and target type as one factor of congruency
(congruent, incongruent) and employed 4 (prime duration: 20 ms, 50 ms, 100 ms, 150 ms; between-subjects) 2 (congruency: congruent, incongruent; within-subjects) ANOVA. In the other set, we applied two 4 (prime duration: 20 ms, 50 ms,
100 ms, 150 ms; between-subjects) 2 (congruency: congruent, incongruent; within-subjects) ANOVAs separately for the
trials that required ‘‘same’’ responses and ‘‘different’’ responses. This extra analysis was necessary as different processing
mechanisms are assumed to mediate positive (or ‘‘same’’) and negative (or ‘‘different’’) judgments.
Number pairs consisting of same numbers were called positive targets (e.g., ‘‘3n3’’), and pairs of different numbers were
called negative targets (e.g., ‘‘3n5’’). Before the target, a masked picture with either positive or negative connotations was
briefly presented as the prime, which was respectively called the positive or negative prime. Trials that contained a positive
prime and positive target were called PP trials; Trials that contained a negative prime and negative target were called NN
trials; Trials that contained a positive prime and negative target were called PN trials; Trials that contained a negative prime
and positive target were called NP trials. PP and NN trials were classified as congruent trials, while PN and NP trials as incongruent trials. There were 240 trials for each participant (120 congruent trials and 120 incongruent trials). These trials were
further divided into two categories requiring either ‘‘same’’ or ‘‘different’’ responses (Table 1). The order of presenting trials
was random, and whether the ‘‘Same’’ button was on the left or right side of the screen was fixed for each participant
whereas randomly determined between participants.
3. Results
Eight participants did not complete the experiment. Another seven participants were excluded from our data analysis
because their overall accuracy for the number judgment task was below our criterion (i.e., 90% correct). Thus, we analyzed
the data from 367 participants. Only correct trials with response time less than 2000 ms were included in the analysis.1 Less
than 4% of the total trials were rejected.
Based on previous findings that the positive judgment is faster than negative judgment, we assume that positive (i.e.,
‘‘same’’)/negative (i.e., ‘‘different’’) number judgments are carried out by distinct cognitive processes (Proctor, 1981). For this
reason, we analyzed congruency effects with both overall data (collapsing ‘‘same’’ and ‘‘different’’ judgment trials) and separate data (separating ‘‘same’’ and ‘‘different’’ judgment trials).
In the result section, we focus on the mouse movement measure AUC. We start with a summary of overall congruency
effects, and congruency effects analyzed separately for positive response trials (i.e., trials required ‘‘same’’ responses) and
negative response trials (i.e., trials required ‘‘different’’ responses). Following this analysis, we examine the extent to which
the visibility of primes influences the congruency effects using regression analyses. To supplement the mouse movement
analysis, we include our response time analysis at the end.
1
We employed other response time criteria (i.e., 6000 ms, 4000 ms, 3000 ms, 2500 ms, and 1500 ms) and the results from these different criteria yielded
analogous results.
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
47
Fig. 6. Regression plots examining whether d0 can predict the magnitude of priming in each duration (i.e., 20 ms, 50 ms, 100 ms, 150 ms). The x-axis
represents d0 and the y-axis shows the magnitude of the priming (MoP), which is calculated by subtracting the performance (AUC) in congruent trials from
that in incongruent trials.
3.1. Congruency effects measured by AUCs
The mouse movement data showed significant congruency effects. Overall, the AUCs in incongruent trials was larger
(M = 0.73, SD = 0.39) than those in congruent trials (M = 0.67, SD = 0.35); F(1, 366) = 54.76, MSE = 0.61, p < .001, partial
48
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
Table 1
Number of trials in each condition.
Congruency
Combination
Prime
Target
Number of trials
Congruent trials (120)
PP (60)
Positive primes
Positive primes
Negative primes
Negative primes
3n3
5n5
3n5
5n3
30
30
30
30
Negative primes
Negative primes
Positive primes
Positive primes
3n3
5n5
3n5
5n3
30
30
30
30
NN (60)
Incongruent trials (120)
NP (60)
PN (60)
Note: There were 240 trials for each participant: 120 trials were congruent (60 PP and 60 NN trials) and 120 trials were incongruent (60 NP and 60 PN trials).
These trials were further divided into two categories with either the ‘‘same’’ (‘‘3n3’’ or ‘‘5n5’’ targets) or ‘‘different’’ (‘‘3n5’’ or ‘‘5n3’’ targets) response trials.
Table 2
AUCs (Area Under the Curve) for congruent and incongruent trials.
Durations
Incongruent
Congruent
T
p
Cohen’s d
20 ms
50 ms
100 ms
150 ms
N
88
96
91
92
0.75
0.69
0.77
0.70
0.70
0.63
0.71
0.64
3.47
4.92
3.34
3.39
0.001
<0.001
0.001
0.001
0.37
0.50
0.35
0.35
Overall
367
0.73
0.67
7.45
<0.001
0.39
g2 = 0.13. Such congruency effects were substantial in all durations (Table 2); t’s > 3.30, p’s < .002. There was no interaction
between prime duration and congruency; F < 1.0, suggesting that the duration of prime did not affect the congruency effects.
Note that this absence of interaction was not due to a lack of statistical power. Our power analysis suggests that the probability of falsely approving a null effect (type II error) is 0.04.
The congruency effects were significant for both positive and negative response trials, and the effect size appeared larger
in positive response trials (Table 3), though the interaction between response type (i.e., positive response trials and negative
response trials) and congruency was not significant; F(1, 366) = 1.08, MSE = 0.02, p > .30, partial g2 < 0.01. For positive
response trials, robust congruency effects were found in all four durations (Table 3); while for negative response trials, congruency effects were weaker (Table 4). In addition, there was no interaction between duration and congruency, neither in
positive nor in negative response trials; F < 1. Note that we calculated the Cohen’s d using the pooled standard deviation following the procedure suggested by Cumming (2012, pp. 290–294)2; thus, the effect size measures were far more conservative.
3.1.1. Awareness analysis for AUCs
To investigate the influence of prime visibility on congruency effects, we analyzed the priming effects with respect to the
results from awareness tests. Specifically, the d0 measure was calculated to assess the extent to which participants could
identify primes and to which the visibility of primes influenced the magnitude of the congruency effects. Not surprisingly,
the longer the duration of primes was, the higher d0 was (Table 5).
A regression analysis was performed on the d0 as the independent variable, while the magnitude of priming served as the
dependent variable. The magnitude was calculated by subtracting the AUCs in congruent trials from those in incongruent
trials. If the intercept of the regression line is significantly larger than zero, it may be argued that the priming effect is significant when the visibility of the primes is nearly zero (Greenwald, Draine, & Abrams, 1996). Furthermore, if the slope of the
regression line is not different from zero, it can be suggested that the priming effect is relatively independent of the visibility
of the primes (Desender & Van den Bussche, 2012). Overall, the regression on d0 revealed significant intercept (b0 = 0.05,
t = 3.77, p < .001), and the slope of the regression line was not different from zero (b0 = 0.003, t = 0.46, p > .64), implying that
the visibility of primes had little influence on the congruency effects.
Finally, to examine whether the priming effects were influenced by response type (i.e., ‘‘Same’’ or ‘‘Different’’) or by location of the ‘‘Same’’ and ‘‘Different’’ buttons (e.g., ‘‘Same’’ button on the left or right side of the screen), a four-way ANOVA,
prime duration (20, 50, 100, 150 ms) congruency (congruent, incongruent) response type (same, different) button
location (left, right), was performed. This analysis showed that the congruency factor did not interact with any other factors,
F < 1.0. Taken together, these results are consistent with the idea that the congruency effects were robust and impervious to
external factors.
2
Specifically, the two paired groups, congruent and incongruent trials, are treated as independent samples, and the standard deviation to calculate d was
p
obtained from an average of the standard deviations of congruent trials (SDcongruent) and incongruent trials (SDincongruent) (i.e., SDaverage = (SDcongruent2 +
p
2
SDincongruent )/2). Thus, d values reported in this article are far conservative compared to those calculated directly from t scores (i.e., d = 2 t/ (N 2)).
49
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
Table 3
Comparison of AUC for positive response trials.
Duration
Incongruent
Congruent
t
p
20 ms
50 ms
100 ms
150 ms
N
88
96
91
92
0.75
0.67
0.74
0.71
0.68
0.60
0.66
0.64
3.81
3.74
3.78
3.64
<0.001
<0.001
<0.001
<0.001
Overall
367
0.71
0.64
6.37
<0.001
Table 4
Comparison of AUC for Negative Response Trials.
Duration
Incongruent
Congruent
t
p
20 ms
50 ms
100 ms
150 ms
N
88
96
91
92
0.74
0.70
0.79
0.67
0.71
0.64
0.76
0.63
1.58
2.99
1.34
1.82
0.119
0.004
0.182
0.074
Overall
367
0.74
0.69
4.61
<0.001
Table 5
Descriptive statistics of d0 from the awareness test.
Duration
N
Mean
Median
Minimum
Maximum
Std.
20 ms
50 ms
100 ms
150 ms
88
96
91
92
0.55
1.56
2.25
2.52
0.54
1.68
2.88
3.00
0.20
0.07
0.11
0.07
1.47
3.00
3.00
3.00
0.47
1.00
1.09
0.86
3.1.2. Awareness analysis for positive and negative responses in each duration
Our regression analyses applied in each duration, separately for positive and negative response trials, showed significant
priming for positive response trials in duration 20 ms, 50 ms and 100 ms, except for 150 ms. For negative response trials, a
significant intercept was observed in 50 ms only (Table 6). This suggested that the size of priming effects depended on
whether the required response was positive or negative, and appeared to be larger for positive response trials.
To summarize, our mouse movement analyses showed consistent congruency effects in all four durations (Table 2). Congruency effects were robust for positive response trials, as compared to negative response trials (Tables 3 and 4). The regression analyses further showed that the intercepts of the regression lines were significantly higher than zero in 20 ms, 50 ms,
and 100 ms prime durations for positive response trials (Table 6).
Table 6
Regression for AUC on d0 in each duration.
Duration
Response
Predictor
B
t
p-Value
20 ms
Positive
Constant
d0
Constant
d0
0.067
0.005
0.004
0.069
2.332
0.130
0.108
1.485
0.022
0.897
0.914
0.141
Constant
d0
Constant
d0
0.080
0.045
0.003
0.013
2.500
0.983
2.139
0.661
0.014
0.934
0.035
0.510
Constant
d0
Constant
d0
0.114
0.014
0.038
0.002
2.282
0.704
0.681
0.110
0.025
0.483
0.498
0.913
Constant
d0
Constant
d0
0.080
0.015
0.050
0.009
1.117
0.568
0.791
0.366
0.267
0.571
0.431
0.716
Negative
50 ms
Positive
Negative
100 ms
Positive
Negative
150 ms
Positive
Negative
50
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
Table 7
Response times for congruent and incongruent trials (ms).
Durations
20 ms
50 ms
100 ms
150 ms
Together
N
Incongruent
Congruent
t
p
Cohen’s d
88
96
91
92
1077.11
1118.65
1103.95
1129.15
1077.40
1112.49
1100.54
1123.67
0.10
2.04
0.93
1.51
0.92
0.04
0.36
0.14
0.01
0.21
0.10
0.16
367
1107.68
1103.91
2.27
0.02
0.12
Table 8
Comparison of response time for positive response trials.
Duration
Incongruent
Congruent
t
p
20 ms
50 ms
100 ms
150 ms
N
88
96
91
92
1075.97
1108.20
1098.04
1122.73
1072.60
1101.63
1086.62
1112.95
0.80
1.56
2.44
2.14
0.43
0.12
0.02
0.04
Overall
367
1101.60
1093.79
3.53
<0.001
Table 9
Comparison of response time for negative response trials.
Duration
Incongruent
Congruent
t
p
20 ms
50 ms
100 ms
150 ms
N
88
96
91
92
1078.26
1129.10
1109.86
1135.57
1082.20
1123.34
1114.45
1134.39
0.92
1.28
0.90
0.23
0.36
0.20
0.37
0.82
Overall
367
1113.76
1114.04
0.12
0.91
3.2. Response time analysis
Compared with the mouse movement data, the overall congruency effects measured by response times were weaker; the
mean RT in incongruent trials was significantly longer (M = 1107.68, SD = 109.85) than that in congruent trials (M = 1103.91,
SD = 110.19)3; F(1, 363) = 4.94, MSE = 2502.87, p = 0.03, partial g2 = 0.01. There was no interaction between prime duration and
congruency; F < 1.0 (see Table 7).
The t-tests applied to each duration showed that the mean response times were significantly shorter in congruent trials
than in incongruent trials in the 50 ms duration condition: t(95) = 2.06, p = 0.04, d = 0.21, 95% CId [0.07, 0.50]. A similar
trend was present for the remaining three conditions except for the 20 ms group, where the average RTs in congruent
and incongruent trials were nearly identical.
Congruency effects were found for positive response trials (Table 8), but not for negative response trials (Table 9). The
interaction between response type and congruency was significant; F(1, 366) = 6.40, MSE = 937.82, p = 0.01, partial
g2 = 0.02. There was no interaction between prime duration and congruency, neither for positive nor for negative response
trials; F < 1.
The regression analysis showed that the intercept was not different from zero; b0 = 1.02, t = 0.34, p > .73. The slope was
not different from zero; b0 = 1.59, t = 1.12, p > .26. In addition, prime duration did not interact with congruency effects,
F < 1.0. Further regression analyses performed separately for positive and negative response trials in each duration found
intercepts above zero in the 100 ms condition (Table 10). These findings are consistent with the idea that the magnitude
of priming was subject to response types, and larger priming effects were found for positive response trials.
4. Discussion
In this study, we applied Opstal’s number judgment task but replaced their letter-pair primes (e.g., ‘‘A, a’’) with pictures
with positive or negative connotations. In our setting, the prime-target relation was quite abstract, and we investigated congruency effects by comparing mouse-movement trajectories in congruent and incongruent trials.
The AUCs in incongruent trials were greater than those in congruent ones, indicating that the positive or negative connotations of primes influenced the same/different judgment on numbers. Furthermore, the coefficients for the regression
3
Since response time data are sensitive to outliers, we reanalyzed the data following the suggestion by Ratcliff (1993). We calculated the inverse of RT (1/RT)
without applying any arbitrary cut-off criterion. Overall results of this extra analysis were consistent with the regular response time analysis reported here.
51
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
Table 10
Regression for response time on d0 in each duration.
Duration
Response
Predictor
T
p-value
20 ms
Positive
Constant
d0
Constant
d0
4.952
2.864
6.089
3.895
0.752
0.315
0.912
0.422
0.454
0.754
0.364
0.674
Constant
d0
Constant
d0
4.789
1.140
5.620
0.089
0.610
0.269
0.672
0.020
0.543
0.789
0.503
0.984
Constant
d0
Constant
d0
21.537
4.507
24.671
8.945
2.003
1.046
2.148
1.943
0.048
0.299
0.034
0.055
Constant
d0
Constant
d0
24.857
5.973
13.345
5.758
1.748
1.119
0.828
0.952
0.084
0.266
0.410
0.343
Negative
50 ms
Positive
Negative
100 ms
Positive
Negative
150 ms
Positive
Negative
B
analysis were nearly zero, suggesting that the visibility of the primes had little influence on the magnitude of priming. Overall, our mouse movement data revealed more robust congruency effects than response time data, and positive response trials
yielded greater congruency effects compared to negative response trials. Below, we discuss these results with respect to
three essential issues: the question of subliminal priming, the complexity of semantic priming, and the mouse movement
method applied to study priming.
4.1. The question of subliminal priming
Our d0 measure ranged from 0.20 to 3.00, with the median of 1.97. Because primes were visible for most participants, it
is difficult to claim that priming effects emerged from subliminal processing. On the other hand, since participants ignored
primes in the number judgment task while paid attention to primes in the awareness test, the visibility of primes was lower
in the number judgment task than in the awareness test (Naccache, Blandin, & Dehaene, 2002). It is reasonable to suggest
that the semantic priming occurred very rapidly (Fisk & Haase, 2011). Nominally, a longer prime presentation means a higher
level of awareness, and greater priming effects are expected (Vermeiren & Cleeremans, 2012). Yet, we found no correlation
between the magnitude of priming and d0 . Give the absence of interaction between prime duration and congruency effect,
our regression analysis showing that congruency effects were still significant at a relative low level of visibility (d0 = 0) does
not necessarily reflect a subliminal processing.
4.2. The complexity of semantic priming
Most masked priming tasks can be classified into three types given the extent to which primes and targets overlap: repetition priming, cued priming, and semantic priming. For repetition priming, targets and primes are more or less identical,
thus priming effects are contaminated by a target-response mapping (Damian, 2001). Possibly, participants are merely influenced by perceptual features of primes rather than semantic processing (Abrams & Greenwald, 2000). Cued priming has similar problems. In this setting, participants perform different judgment tasks according to an instructional cue, preceded by a
masked priming cue (Weibel, Giersch, Dehaene, & Huron, 2013). For example, a cue (e.g., ‘‘A’’ or ‘‘S’’) was explicitly presented
prior to a target (i.e., ‘‘sparrow’’); ‘‘A’’ asked for a semantic judgment (e.g., living or non-living), while ‘‘S’’ for a phonological
judgment (e.g., bisyllabic or not). In each trial, the explicit cue was preceded by a masked prime (e.g., ‘‘A’’ or ‘‘S’’). Again, cues
were also used as primes, so the priming effects could result from cue-response mapping instead of the semantic processing.
To exclude the alternative explanation of stimulus–response mapping, primes should be novel stimuli that have never been
explicitly presented to participants. However, even for many studies employing novel primes, the mapping problems still
exists, because the association between primes and targets was straightforward in terms of the task demand. For example,
the cognitive strategy adopted for judging targets (e.g., same/different judgment) can also be applied to primes, which creates an indirect contamination by stimulus–response mapping between targets and primes. Essentially, the ‘‘mapping’’ problem is that the practiced processing of targets facilitates the unintentional processing of primes, which in turn, may
exaggerate the priming effects.
To reduce the stimulus–response mapping, we made the link between the primes and targets obscure. The same/different
judgment on number pairs can hardly be applied to a single picture like a red light. Primes and targets are only weakly associated via their positive or negative connotations. Unless those connotations were automatically processed, congruency
effects could not be observed. As predicted, such a weak prime-target association still generated salient priming, suggesting
that the semantic information of masked pictures can influence the same/different judgment in a relatively complex way.
52
K. Xiao, T. Yamauchi / Consciousness and Cognition 27 (2014) 42–52
Admittedly, any priming task that creates congruency between primes and targets cannot completely eliminate the stimulus–response mapping. Though we found that the priming effects were larger for positive than negative responses, the
mechanism underlying this disparity is unclear. We speculate that ‘‘different’’ judgments require different processes from
‘‘same’’ judgments; and to make ‘‘different’’ judgments, some extra processing should involve. We plan to address this question in a future study.
4.3. Measuring priming using a mouse movement method
In the current study, the results of RTs were roughly consistent with those of AUCs. However, larger congruency effects
were found in the mouse trajectory measures than RT measures. This implies that AUCs are more sensitive to semantic priming effects compared with RTs. However, note that our mouse movement method did not record reaction times in a traditional way (i.e., pressing response keys rather than clicking mouse). It is difficult to compare the relative merit of the two
measures in this study. At this stage, we merely suggest that mouse movement methods can supplement the traditional
reaction time techniques.
The mouse movement method records the dynamic real-time information of the hand motion, which reflects continuous
cognitive processes (Freeman et al., 2008). As an emerging psychophysical technique, the mouse movement method has
been applied to study visual perception, social stereotype, inductive reasoning, lexical judgment, and problem solving
(Dale, Kehoe, & Spivey, 2007; Spivey & Dale, 2006; Yamauchi, 2013; Yamauchi, Kohn, & Yu, 2007), showing its advantage
to revealing subtle cognitive processes. In this regard, we suggest that the mouse movement method can provide a viable
tool to study semantic priming.
5. Conclusion
To conclude, we find that masked pictures can influence number judgments when the prime-target association is indirect,
which adds more evidence to the presence of complex semantic priming. Such congruency effects are not influenced by
prime duration. In addition, we find that a mouse movement measure, which assesses the attraction of cursor trajectories,
can help revealing priming effects. We recommend that mouse movement methods can work as complementary tools for
semantic priming research, in addition to reaction time techniques.
References
Abrams, R. L., & Greenwald, A. G. (2000). Parts outweigh the whole (word) in unconscious analysis of meaning. Psychological Science, 11(2), 118–124.
Aglioti, S., DeSouza, J. F., & Goodale, M. A. (1995). Size-contrast illusions deceive the eye but not the hand. Current Biology, 5(6), 679–685.
Cumming, G. (2012). Understanding the new statistics: Effect sizes, confidence intervals, and meta-analysis. New York: Routledge.
Damian, M. F. (2001). Congruity effects evoked by subliminally presented primes: Automaticity rather than semantic processing. Journal of Experimental
Psychology Human Perception and Performance, 27(1), 154–165.
Dale, R., Kehoe, C., & Spivey, M. J. (2007). Graded motor responses in the time course of categorizing atypical exemplars. Memory & Cognition, 35(1), 15–28.
Desender, K., & Van den Bussche, E. (2012). Is consciousness necessary for conflict adaptation? A state of the art. Frontiers in Human Neuroscience, 6(3), 1–13.
Fisk, G. D., & Haase, S. J. (2011). The relationship between the objective identification threshold and priming effects does not provide a definitive boundary
between conscious and unconscious perceptual processes. Consciousness and Cognition, 20(4), 1221–1231.
Freeman, J. B., Ambady, N., Rule, N. O., & Johnson, K. L. (2008). Will a category cue attract you? Motor output reveals dynamic competition across person
construal. Journal of Experimental Psychology: General, 137(4), 673–690.
Freeman, J. B., & Ambady, N. (2010). MouseTracker: Software for studying real-time mental processing using a computer mouse-tracking method. Behavior
Research Methods, 42(1), 226–241.
Friedman, J., & Finkbeiner, M. (2010). Temporal dynamics of masked congruence priming: Evidence from reaching trajectories. In ASCS09: Proceedings of the
9th conference of the Australasian society for cognitive science. Sydney: Macquarie Centre for Cognitive Science (pp. 98–105).
Greenwald, A. G., Draine, S. C., & Abrams, R. L. (1996). Three cognitive markers of unconscious semantic activation. Science, 273(5282), 1699–1702.
Kiesel, A., Kunde, W., Pohl, C., Berner, M. P., & Hoffmann, J. (2009). Playing chess unconsciously. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 35(1), 292.
Macmillan, N., & Creelman, C. D. (1991). Detection theory: A user’s guide. New York: Cambridge University Press.
Naccache, L., Blandin, E., & Dehaene, S. (2002). Unconscious masked priming depends on temporal attention. Psychological Science, 13(5), 416–424.
Opstal, F. V., Gevers, W., Osman, M., & Verguts, T. (2010). Unconscious task application. Consciousness and Cognition, 19(4), 999–1006.
Ortells, J. J., Marí-Beffa, P., & Plaza-Ayllón, V. (2012). Unconscious congruency priming from unpracticed words is modulated by prime-target semantic
relatedness. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 1–20.
Proctor, R. W. (1981). A unified theory for matching-task phenomena. Psychological Review, 88(4), 291.
Ratcliff, R. (1985). Theoretical interpretations of the speed and accuracy of positive and negative responses. Psychological Review, 92(2), 212.
Ratcliff, R. (1993). Methods for dealing with reaction time outliers. Psychological Bulletin, 114(3), 510.
Schoups, A. A., Vogels, R., & Orban, G. A. (1995). Human perceptual learning in identifying the oblique orientation: Retinotopy, orientation specificity and
monocularity. The Journal of Physiology, 483(Pt 3), 797–810.
Song, J. H., & Nakayama, K. (2009). Hidden cognitive states revealed in choice reaching tasks. Trends in Cognitive Sciences, 13(8), 360–366.
Spivey, M. J., & Dale, R. (2006). Continuous dynamics in real-time cognition. Current Directions in Psychological Science, 15(5), 207–211.
Sternberg, S. (1966). High-speed scanning in human memory. Science, 153(3736), 652–654.
Van den Bussche, E., Van den Noortgate, W., & Reynvoet, B. (2009). Mechanisms of masked priming: A meta-analysis. Psychological Bulletin, 135(3), 452.
Vermeiren, A., & Cleeremans, A. (2012). The validity of d0 measures. PLoS One, 7(2), e31595.
Weibel, S., Giersch, A., Dehaene, S., & Huron, C. (2013). Unconscious task set priming with phonological and semantic tasks. Consciousness and Cognition,
22(2), 517–527.
Yamauchi, T., Kohn, N., & Yu, N. Y. (2007). Tracking mouse movement in feature inference: Category labels are different from feature labels. Memory &
Cognition, 35, 852–863.
Yamauchi, T. (2013). Mouse trajectories and state anxiety: Feature selection with random forest. In Proceedings on humaine association conference on affective
computing and intelligent design (ACII 2013) (pp. 394–404). IEEE Computing Society. doi: http://dx.doi.org/10.1109/acii.2013.72.