Participants
Twenty-four participants volunteered for the study (21 males; average age 27.83, SD 5.42) for a payment of 3000 JPY. All had normal or corrected-to-normal vision and were naïve as to the hypothesis under investigation. They all gave a written informed consent. Three additional volunteers participated in the study but they were not included in the sample and analyzed (see “Data analyses” for inclusion criteria). All experimental procedures were approved by the ethics committee of the National Institute of Information and Communications Technology (NICT).
Apparatus and stimuli
Participants were tested using a personal computer (Lenovo, T400). Stimulus presentation and data collection were performed using Matlab software (R2013b) and the Psychophysics Toolbox (Brainard 1997; Pelli 1997). Tactile stimulation was delivered to the palmar part of the distal phalanx of the right hand thumb and fifth digit (12 mm2 contact surface) by solenoid tappers (Heijo Research Electronics), which were controlled with Matlab. In order to cover the noise made by key-presses and tappers, participants were presented with a background white noise via headphones.
All visual stimuli were presented on a display with 1440 × 900 pixels resolution and 60 Hz refresh rate, from a viewing distance of approximately 55 cm. Visual flash stimuli consisted of white flash presented within a Gaussian envelop of about 1.5° diameter. Other visual stimuli consisted of two black arrows (with a shape of an equilateral triangle) pointing either to the left or the right, and a geometric shape created by the superimposition of the left and right arrows. The size of these stimuli was ~ 1.01° width (Fig. 1).
Procedure
Participants completed two conditions: an action and a tactile condition. Each condition consisted of two blocks: cued and uncued blocks. Therefore, the experiment consisted of a 2 (ACTION: present, absent) by 2 (CUE: cued, uncued) factorial design. The order of the action and tactile condition, and the order of the cued and uncued blocks were counterbalanced across participants.
In each trial of the action condition, participants performed two tasks: two types of choice reaction time task (RT task) and a simultaneity judgment task (SJ task). In the RT task, participants were presented with an arrow at the center of the screen pointing either to the left or to the right (go-signal). The arrow was presented for two frames (33 ms). The orientation of the arrow was selected randomly and equiprobably. Participants were required to respond as fast and as accurately as possible to the arrow by executing a right-hand thumb key-press, when the arrow pointed to the right, and a right-hand fifth digit key-press, when the arrow pointed to the left. To manipulate the degree of action preparation in the RT task, two different types of cue-signal preceded the go-signal. In the cued block, before the presentation of the go-signal, participants were presented for 200 ms with an arrow pointing in the same direction as the go-signal (Fig. 1). The SOA between the cue-arrow and the go-signal-arrow was varied randomly and taken from an exponential distribution with average of 600 ms, and 500 and 800 ms for the shortest and longest possible SOAs, respectively. This was done to prevent any anticipation and temporally stereotyped response to the presentation of the go-signal. The uncued block was the same as the cued-action block except that participants were presented with a neutral cue (the result of the superimposition of the left and right arrows) before the onset of the go-signal (Fig. 1). Consequently, until the presentation of the go-signal, they did not know which action to execute. The trials in which participants pressed the wrong key were interrupted and repeated. Any key-press performed between 0 and 150 ms after the onset of the go-signal were considered as anticipation rather than reaction to the go-signal, which the trials were interrupted and repeated. Also, key-press performed later than the average reaction time + 480 ms was considered as a too late key-press and the trial was also interrupted and repeated.
For the SJ task, a visual flash (16 ms duration) was presented either before or after participants’ key-press executed in response to the go-signal. The flash was delivered randomly at one of ten different stimulus onset asynchronies (SOAs) ± 183 ± 150 ± 116 ± 83 ± 50 ms relative to the average action latency calculated on previous trials. Negative SOAs indicate that the flash was presented before the average reaction time. For the first trials, the average reaction time was calculated using the data from a short training session consisting of 20 trials, where participants were familiarized with the task. After the key-press was executed and the flash was presented, participants reported whether the flash occurred simultaneously with the action or not by pressing on one of two designated keys using their left hand. We chose simultaneity judgments, rather than temporal order judgements (“before” or “after” judgements) because the latter can lead to response biases that would affect directly our dependent measure (the point of subjective simultaneity, see “Data analyses”), such as tendencies to judge the flash more frequently as occurring after the action (see “Discussion”).
Crucially, the average reaction time, that was used to determine the onset of the flash, was calculated separately for each condition (cued and uncued) and finger (thumb and fifth digit key-press), in order to make sure that the distributions of action-flash SOA in respect to the reaction time were similar across conditions and the action executed.
The procedure for the tactile condition was the same as for the action condition, except that participants did not execute any key-press in response to the go-signal, but instead, were delivered with a tactile stimulation on the finger indicated by the arrow. At the end of the trial they judged whether the flash and the tactile stimulation were simultaneous or not. In order to maximally match the temporal parameters of the action and the tactile conditions, both the onset and the duration of the tactile stimulation were determined by the individual action onset and duration recorded in the action condition. For instance, the onset and duration of the thumb tactile stimulation in one of the trials of the cued-tactile block was determined by the onset and duration of a thumb key-press in one of the trials of the cued-action block. However, if a participant began the experiment with the tactile condition, the onset time and duration of the tactile stimulation was derived from the mean and standard deviation of all previous participants’ action latency and duration.
The experiment lasted about 90 min. Each condition (action, tactile) consisted of 600 trials: 30 trials per action/touch-flash SOA per Cue [30 (trials) × 10 (SOAs) × 2 (cues: cued and uncued)].
Data analyses
Considering that visual flashes were presented on the basis of estimated action time (i.e., average action latency), we calculated the actual action-flash interval for each trial. Then, we divided each the stimulus-before-action trials and the stimulus-after-action trials into five time intervals of equal number of trials (total of 10 bins). This was done both for cued and uncued trials, for each participant (See Table 1 for mean and standard deviation of number of trials for each time bin, and mean and standard deviation of time intervals for each bin). The same procedure was performed for the tactile trials.
Table 1 Mean (SD) number of trials for each of the ten resampled bins in the action block and tactile block
Then, we calculated the proportion of judgments action/touch and flash simultaneous for each time interval (10 time bins) per participant. From these data, psychometric functions were calculated using a (Gaussian) nonlinear regression model (Eq. 1) implementing the Maximum Likelihood procedure as described in Myung (2003). Three parameters were fitted (1) mean α, (2) standard deviation σ and (3) a scale factor s, which refers to the amplitude of the Gaussian curve.
$$f\left( x \right)=s \cdot {e^{ - {{\left( {x - \alpha } \right)}^2}/(2{\sigma ^2})}}$$
(1)
Importantly, to make sure that our binning procedure did not introduce any bias in the observed results we also analysed the data without binning the data at all (see Supplementary Material).
Each individual point of subjective simultaneity (PSS), representing an estimate of the temporal offset between the action/touch and flash required to perceive these two events simultaneously, was evaluated as mean of the fitted Gaussian curve. Standard deviation (SD) of the Gaussian curve was used as an estimate of participants’ temporal sensitivity, i.e. how well they could detect asynchronies between the action/touch and the flash. Higher SD values indicate low sensitivity to asynchrony. Participants were not included in the sample size if their PSS or SD (averaged across condition) was higher than the largest action/touch-flash SOA, since the fit of the Gaussian curve would be unreliable. Moreover, participants with a psychometric function whose amplitude was lower than 0.5 were rejected from further analyses due to the same reason. According to these criteria three volunteers participated in the study and were not included in the sample size.
Our main interest was the PSS difference between cued trials and uncued trials. We hypothesized that if action specification in cued trials leads participants to perceive their actions as occurring earlier in time, a significant change in PSS values would be observed in cued as compared to uncued-action trials. Moreover, we expected this effect to be driven by action preparation process, rather than the non-motor expectations; therefore, we expected no difference in PSS values between cued and uncued trials in the tactile condition. We tested this hypothesis by evaluating the interaction between the factors ACTION (present, absent) and CUE (cued, uncued).
After a first visual inspection of the PSS differences between cued and uncued trials for both action and tactile blocks we noticed the presence of an outlier. One participant had a PSS difference between cued and uncued trials exceeding the group average by 2.5 standard deviations. It is well known that mean values can strongly be affected by outliers; hence we decided to use non-parametric two-tailed exact Wilcoxon signed rank test for our analyses. This test is less affected by outliers since it relies on ranks rather than mean values. Significance value was set at p < 0.05 for all statistical tests. We also confirmed that excluding the outlier and performing a parametric test did not change the conclusion drawn from the non-parametric test.