Differentiating the reported time of intent and action on the basis of temporal binding behaviors and confidence ratings

Abstract

The reported time of intent (W) and the reported time of action (M) have been used as indices of consciousness during simple voluntary actions. However, it is unclear whether W is exclusively inferred from M. Past studies have suggested that W is inferred from M by demonstrating that W varies when judged in conjunction with M. The current study offers counterevidence by showing that W is independent of M under some circumstances related to temporal binding. Participants performed a voluntary keypress that elicited a tone (briefly delayed at 5 and 60 ms). Subsequently, they reported W or M and indicated the confidence of their report. Binding strength was measured as the extent to which the W and M reports gravitated toward the time of the tone. Moreover, the binding strength was evaluated in conjunction with time course and knowledge to assess whether the strength increases due to repeated exposure or weakens if informed of the tone delay manipulation, respectively. We observed that the binding strength associated with W increased over time, and being informed of the tone manipulation did not affect W’s binding behaviors. In contrast, M’s binding behaviors did not change over time but being informed of the tone manipulation may release M from binding. The corresponding confidence ratings associated with W were uniform whereas those associated with M fluctuated over time. Collectively, the results suggest that binding behaviors associated with W and M differ, and that W is not simply derived from M.

During a decision-making process, the relationship between intent and action is often viewed as unidirectional, with intent leading to an action. In a model attempting to link these phenomenon, Ajzen et al. (2004) proposed that the intent to act is facilitated by factors such as attitude toward the decision, subjective norm (e.g., how others feel about the matter at hand), and the perceived behavioral control. According to their model, then, intent involves a decision component that precedes action in time.

Libet et al. (1983) attempted to quantify intent and action using an experimental paradigm requiring participants to perform an unplanned, volitional act. Following the action, participants subsequently reported their estimate of the time of intent (W) or the time of action (M). Generally, it has been demonstrated that the reported time of intent precedes the reported time of action by approximately 100–150 ms, consistent with the general perspective that intent causes an action (Braun et al., 2021; Libet et al., 1983).

The reasons why participant’s subjective report of W precedes the subjective report of M by about 100 ms remains unclear; and in particular, it has been questioned whether the W report reflects the true timing of intent or whether the W report is artificially produced (Breitmeyer, 1985). Along these lines, Gomes (1998) has proposed that W is inferred from the perceived time of action. Support for this view can also be drawn from several studies in which participants performed the Libet task and judged W and M in different orders (Dominik et al., 2017; Sanford et al., 2020). When the M judgment preceded the W judgment, the W value was significantly earlier than M, implying that the W report is retrospectively anchored around M. On the other hand, when the W judgment preceded the M judgment, the W and M values were the same. As such, the results from these two conditions suggest that when the perceived timing of action is available (e.g., by having to report M first), the W report is anchored and inferred from M, rendering W as dependent and closely tied to the M value.

However, other studies have suggested the W and M reports are distinct. In one of our earlier studies, participants performed a modified Libet task in which they received a tone upon keypress. This tone was briefly and systematically delayed. We observed that the W report was strongly inferred from the timing of the tone, displaying a form of temporal binding behavior in which the timing of the event (in this case, time W) is falsely perceived to be later and closer to the time of the tone (Banks & Isham, 2009). In a separate study, we also observed the M report was inferred from the timing of the tone, but not to the same extent (Banks & Isham, 2010). These results hint to the possibility that W and M, while inferred from the same source—namely, the tone—may nevertheless operate on separate mechanisms, and therefore W and M are distinct under some circumstances.

To determine whether W and M have distinct characteristics, we performed the aforementioned tone manipulation task (Banks & Isham, 2009) and examined the temporal binding patterns for W and M at the sensory and cognitive levels. To assess how W and M differ at the sensory level, we compared W and M’s temporal binding strength over time. The binding strength can be demonstrated by at least two behaviors: when the reported times shift systematically with the timing of the misleading tone, and when this shift is more pronounced over the time course of the experiment (Sugano et al., 2010). Past studies have indicated that an increase in binding strength over time reflects sensory adaptation (Stetson et al., 2006). The more exposure to the tone manipulation, the more we gravitate toward it and rely on it to judge the W and M report.

At the cognitive level, we asked whether the W and M’s binding strength would be reduced if the participants became knowledgeable of the tone manipulation. With knowledge, previous work has reported that being alerted about the rules or structures can alter one’s awareness (Dienes & Scott, 2005). Thus, by raising awareness about the tone manipulation in the current study, the tone’s capacity to influence the W and M reports could be modulated. In the current study, we examined whether explicit knowledge about the tone manipulation would have a top-down influence on the W and M reports. We reasoned that if W and M were inferred from one another, their binding behaviors would be similar in all experimental conditions despite being informed of the tone manipulation. On the other hand, if W and M were independent, we would anticipate the conditions under which W and M bind to differ and dissociate. On the premise that W is cognitively driven (e.g., having an impression of “I hear the tone at this time, therefore, I must have intended the action a moment earlier”), being informed of the experimental structure should weaken the temporal binding effect. On the premise that the M report is based on multiple inputs including sensorial, we anticipated M to be less influenced by knowledge. Such dissociation would differentiate W and M as independent entities.

The effect of the order in which W or M is reported first was also examined. Based on the results observed by Dominik et al. (2017), if W were exclusively inferred from M, we could expect the binding strength or binding pattern of W to vary with the those of M when the W task followed the M task.

To further demonstrate that the reported timing of intent and action are dissociable, we also examined confidence ratings as a possible secondary measure of implicit awareness related to temporal error when reporting W or M. Past studies have supported confidence ratings as an alternative measure of consciousness; that confidence ratings reflect implicit knowledge (e.g., Cheesman & Merikle, 1986; Dienes, 2007) and phenomenal consciousness (Block, 1995; see Norman & Price, 2015, for review) not captured by verbal reports (but see Rosenthal, 2019, for a different perspective). Given the possibility that we are able to detect a temporal discrepancy between the timing of W or M and timing of the tone but are not able to explicitly and verbally report it, such implicit detection may manifest in the form of confidence reports. Therefore, if W were dependent on M, the corresponding confidence for W should be the same for M. Alternatively, if the confidence ratings are different, then it would suggest that the W report does not rely exclusively on the M report. As such, we may be able to differentiate between W and M on the basis of implicit temporal awareness, further supporting the working hypothesis that W does not exclusively rely on M.

Experiment 1

Methods

Participants

Forty participants were recruited for the experiment (18 years or older; 20 participants in each of the two conditions which include the Uninformed, and the Uninformed/Informed conditions). The number of participants was based on a priori power calculation for a mixed analysis of variance (ANOVA), with one between-subjects independent variable and three within-subjects variables, with alpha error probability = .05, and power = .9 (G*Power; Faul et al., 2007). This results in a sample size estimate of 36 participants. Participants received course credits or monetary compensation for the study ($10). The study protocol was approved by the Institutional Review Board of the University of California, Davis.

Apparatus and procedure

The experiment followed a protocol adapted from Libet et al. (1983) and Banks and Isham (2009). Participants sat in front of a computer screen 60 cm away. In each trial, a circular clock face was presented. The clock was 90 mm in diameter, with 60 evenly spaced tick marks numbered from 0 to 59. A cursor moved in the clockwise direction along the circumference, completing one revolution in 3 seconds. Each trial consists of two revolutions. Participants fixated their gaze at the center of the clock and rested their right index finger on the response button (space bar on the keyboard). They were instructed to press the button spontaneously and to not plan the time of the button press. They could choose not to make a button press on any trial. When the button was depressed, the computer emitted a 200-ms beep either after a shorter (5 ms) or a longer delay (60 ms). The delays were randomized. Upon trial completion, the participants were asked to report the number on the clock marked by the moving clock cursor at the instant they felt the intent to respond (W) or the instant they executed the action (M). Participants did not receive extensive training to perform the button press nor how to report the timing of intent or action. They were provided with the operational definitions that the timing of intent referred to the moment one feels the urge or a decision to press the button, and the timing of action referred to the moment in which the button is pressed. Importantly, we did not emphasize that the timing of intent should precede the timing of action, but the assumption could be implied. Additionally, they were asked to rate how confident they were of the report being accurate; a Likert scale was used, with 1 representing the lowest level of confidence and 7 representing the highest level of confidence. See Fig. 1 for a pictorial depiction of the trial structure. The specifications of each condition were as follows.

Fig. 1

Trial structure for all experimental conditions. Participants pressed a button whenever they felt the urge to do so. Upon the button press, a tone was delivered after a brief delay (5 or 60 ms). There were two Timepoints (Timepoint 1 and Timepoint 2). Each timepoint consisted of a W block and an M block. Participants reported the timing of intent (W) or the timing of action (M) by reading from the Wundt clock (order randomized). They also indicated how confident they were of their temporal report, on a scale of 1 (lower confidence) to 7 (higher confidence). The Uninformed group received no additional instructions (Experiment 1). The Uninformed/Informed group was informed of the tone manipulation at Timepoint 2 (Experiment 1) and the fully Informed group was informed at the beginning of Timepoint 1 (Experiment 2)

Uninformed condition

W and M were judged in separate blocks of trials. The experimental session for each participant consisted of 80 trials (40 trials each for the W and M judgment). These trials were further divided into four alternating blocks of W and M (e.g., 20 trials of W, 20 trials of M, 20 trials of W, and 20 trials of M in alternating orders). Subsequently, for the analyses, the first 40 trials (First W block + First M block) made up the data for Timepoint 1 (T1) and the last 40 trials (Second W block + Second M block) made up Timepoint 2 (T2). The order of the W and M blocks were counterbalanced across participants. Participants were not informed of the two tones having different delays.

Uninformed/Informed condition

At Timepoint 1, participants were not informed of the tone manipulation; thus, the trial structure was the same as in the Uninformed condition. In the second half, at Timepoint 2, the participants were informed of the tone manipulation. Specifically, they were informed that there was a brief delay between keypress and the tone, and that the amount of delay varied.

Analyses

The design of the experiment consisted of five independent variables. The within-subjects variables were Report Type (W and M report), Delay (5 ms and 60 ms delay between keypress and tone delivery) and Timepoint (Timepoint 1 denotes the first half and Timepoint 2 denotes the second half of the experiment). The between-subjects variable was Knowledge Condition (Uninformed and Uninformed/Informed) and Order (W-before-M and M-before-W). The dependent variables were the reported time (backward-referenced from the moment of action), and the corresponding confidence ratings. Regarding the reported time, a greater reported time value indicates that the perceived moment of W or M was earlier and was further away from the time of action and the tone (i.e., weaker binding effect). A smaller reported time value indicates that the perceived moment of W or M was later, closer to the time of action and the time of the tone (i.e., stronger binding effect).

The time report (and corresponding confidence ratings) was subjected to two separate ANOVAs. To examine the binding behavior with respect to the Timepoint and Knowledge variables, the time report was subjected to a 2 Report Type (W and M) × 2 Delay (5ms and 60 ms) × 2 Timepoint (Timepoint 1 and Timepoint 2) × 2 Knowledge Condition (Uninformed or Uninformed/Informed) mixed ANOVA. To examine the effect of order, the time report from Timepoint 1 only was subjected to a 2 Report Type (W and M) × 2 Delay (5 ms and 60 ms) × 2 Order (W before M, or M before W) mixed ANOVA. Post hoc pairwise comparisons were conducted as appropriate.

Bayes factor analysis

The Bayes factor analysis was performed to complement null hypothesis significance testing JASP (Version 0.16; Quintana & Williams, 2018; Rouder et al., 2009). For results below the significance threshold (p < .05), we used a Bayes factor, BF₁₀, to indicate the degree of favorability toward the alternative hypothesis. For results that were not below the significance threshold, the Bayes null factor, BF₀₁, was used. The Cauchy prior r scale was set at the default value of d = .707. The larger the Bayes factor, the greater the evidence in support of the corresponding hypothesis being tested.

Results

Of the 40 participants recruited for the study, two in the Uninformed/Informed condition were excluded from the analysis due to their reported times (W and M) exceeding two standard deviations, resulting in a distribution of 20 participants in the Uninformed condition and 18 participants in the Uninformed/Informed condition.

Timepoint, knowledge, and binding

As anticipated, and consistent with previous observations (e.g., Libet et al., 1983), we observed that the W report (101 ms before keypress, SE = 9) was earlier than the M report (44 ms before keypress, SE = 4), F(1, 36) = 43.262, p < .001, η_p² = .546. Moreover, we also observed a main effect of Delay, F(1, 36) = 29.241, p < .001, η_p² = .448, illustrating that the tone manipulation effectively influenced the reported timing of action and intent such that when the tone was delivered after a brief 5-ms delay, the mean reported times of W and M (76 ms before keypress, SE = 6) was judged to be earlier than when the tone was delivered after a 60-ms delay (69 ms before keypress, SE = 6). This main effect serves as evidence of replication that post-action events, such as a tone, can effectively influence both the W and M reports (e.g., Banks & Isham, 2009). However, there was no interaction effect between Report Type and Delay, F(1, 36) = 1.705, p = .200, η_p² = .045, nor a three-way interaction effect between Report Type, Delay, and Knowledge conditions, F(1, 36) = .194, p = .662, η_p² = .005. The lack of these interactions further suggest that the tone manipulation was effective and influenced both the W and M reports despite the participants having been informed of the tone delays (but see Experiment 2).

Another critical observation was related to the variable Timepoint. In the Uninformed condition, we anticipated the binding strength to increase over time (i.e., the mean reported times would be closer to the time of the tone at Timepoint 2 than at Timepoint 1). Such findings would align with the perspective of adaptation over time (e.g., Stetson et al., 2006). Note that in the Uninformed/Informed condition, Timepoint 2 also marked the timepoint in which the participants were explicitly informed of the tone manipulation. The ANOVA results revealed an interaction between Report Type and Timepoint, F(1, 36) = 16.407, p < .001, η_p² = .313. For the W report, it was observed that temporal binding was strengthened over time (Fig. 2). The W report at Timepoint 1 (W_Time1 = 110.9 ms before keypress, SE = 10.66) was judged to be earlier whereas the W report at Timepoint 2 was judged to be later and closer to the time of the tone (W_Time2 = 89.40 ms before keypress, SE = 8.32), t(37) = 3.848, p < .001, d = .624, BF₁₀ = 62.67. The M report, on the other hand, did not differ over time. The M report at Timepoint 1 (M_Time1 = 42.50 ms before keypress, SE = 3.76) was not statistically different from the M report at Timepoint 2 (M_Time2 = 46.10 ms before keypress, SE = 3.55), t(37) = 1.546, p = .131, d = .251, BF₀₁ = 1.93. In short, though opposite of what we had anticipated, the W reports displayed a sensory adaptation-like behavior by shifting to the tone over time whereas such behavior for the M reports was not as clearly visible in the current study, suggesting the possibility that the W report relies more on the sensorial tone input than did M.

Fig. 2

W and M reports (a) and the corresponding confidence ratings (b) as a function of Delay, Timepoint, and Knowledge. The significant p values are denoted in red, and all p values are two-tailed

In addition, we also predicted that the W report, but not the M report, would be affected by the informed knowledge such that the W reports would be released from binding. We did not observe such effect of Knowledge on W. However, an examination of the M data in the Uninformed/Informed condition revealed that the M report might be released from binding after being informed; that is, the reported M time in the Informed block (49.42 ms before keypress, SD = 2.23) was marginally earlier than the M report in the Uninformed block (41.33 ms before keypress, SD = 2.59) of the Uninformed/Informed condition, t(17) = 2.22, p = .04, d = .523, BF₁₀ = 1.71. Although the results did not achieve statistical significance once corrected, the data trend and M’s potential release from binding was worth mentioning.

The insignificant effects related to Knowledge in the current experiment may be due to the fact that the participants in the Uninformed/Informed group were told of the tone manipulation too late. We further address this concern in Experiment 2. Based on the findings thus far, we conclude that the observed increase in the binding strength of W over time, but not in M’s binding strength, suggests that W and M operate differently and that W does not infer exclusively from M.

Timepoint, knowledge, and confidence ratings

The confidence ratings data were expected to provide additional insights regarding the W and M reports. One of the goals was to determine whether W and M could be differentiated on the basis of confidence level. If these reported times were inferred from one another, we anticipated that their confidence ratings would be similarly judged. The analysis would also reveal whether confidence ratings reflect an implicit awareness about temporal discrepancy. If so, the confidence ratings would be lower if the tone presentation was delayed (i.e., 60-ms condition). We subjected the confidence ratings to a 2 Report Type (W or M) × 2 Delay (5 or 60 ms) × 2 Timepoint (T1 and T2) × 2 Knowledge (Uninformed, Uninformed/Informed; between-subjects) mixed ANOVA. The results are presented in the second panel of Table 1. There was a main effect of Report Type, F(1, 36) = 12.7, p = .001, η_p² = .261. Post-hoc pairwise comparison revealed a greater confidence when reporting W (4.99, SE = .09) than when reporting M (4.72, SE = .09), t(37) = 3.160, p = .003, d = .512, BF₁₀ = 16.23. The observations suggest that in general, participants felt they were more accurate at reporting the timing of intent (W) than the timing of action (M).

Table 1 Experiment 1 ANOVA results (Report Type × Delay × Timepoint × Knowledge) for time reports and corresponding confidence ratings

Two-way and three-way interaction effects provided further insights, relating confidence ratings to structural knowledge, tone manipulation, and adaptation over time. As summarized on Table 1, there was a three-way interaction between Report Type, Timepoint, and Knowledge, F(1, 36) = 18.133, p < .001, η_p² = .335. Relatedly, there were three two-way interactions: Report Type and Timepoint, F(1, 36) = 11.273, p = .002, η_p² = .238; Report Type and Knowledge, F(1, 36) = 17.123, p < .001, η_p² = .322; and between Timepoint and Knowledge, F(1, 36) = 9.266, p = .004, η_p² = .205. In the Uninformed condition, the corresponding confidence ratings for W marginally increased over time, t(19) = 1.993, p = .061, d = .446, BF₁₀ = 1.19 (confidence at W_Time1 was 5.00, SE = .13 and at W_Time2 was 5.32, SE = .15). In contrast, the confidence ratings for M decreased significantly over time, t(19) = 3.943, p = .001, d = .882, BF₁₀ = 41.17. That is, the confidence ratings for M_Time1 was 5.16 points (SE = .18) and this essentially declined for M_Time2 to 4.03 points (SE = .16). Within this Uninformed condition, it appears that over time, the participants felt a consistent level of confidence about the W report, but in contrast, the same participants became less confident in their reporting of M (Fig. 2).

In the Uninformed/Informed condition, the corresponding confidence ratings for W did not vary with Knowledge, t(17) = .694, p = .497, d = .164, BF₀₁ = 3.32. The mean confidence rating for the uninformed block at W_Time1 (4.78, SE = .141) was not significantly different from the ratings for the informed block at W_Time2 (4.86, SE = .16). However, the confidence ratings of the M reports increased, at least marginally, over time, t(17) = 2.104, p = .051, d = .496, BF₁₀ = 1.43, with the confidence ratings of 4.74 (SE = .19) for the uninformed block M_Time1 and of 4.98 (SE = .17) for the informed block M_Time2.There was no main effect nor an interaction surrounding Delay (Table 2), suggesting that temporal discrepancy was not captured by confidence ratings in the current study.

Table 2 Experiment 1 ANOVA results (Report Type × Delay × Knowledge × Order) for time reports and corresponding confidence ratings (data drawn from Timepoint 1 of the Uninformed and Uninformed/Informed conditions)

Order effect and temporal binding

Based on Dominik et al. (2017), we also examined whether the order in which W and M were being reported would affect the W and M judgments. We collapsed together the time reports from Timepoint 1—the time period in which all participants were uninformed of the tone manipulation—from both the Uninformed and Uninformed/Informed conditions. Subsequently, these time reports were subjected to a 2 Report Type (W or M) × 2 Delay (5 or 60 ms) × 2 Order (W-before-M, M-before-W; between-subjects) mixed ANOVAs. A replication of Dominik et al.’s results would yield an interaction effect between Order and Report Type such that the W report would be earlier than the M report if M were reported first (i.e., the M report would serve as an anchor in which the W report is inferred from); and W and M reports would be the same if W were reported first. The ANOVA results revealed no significant effects related to Order. Particularly, there was no interaction between Report Type and Order, F(1, 36) = .755, p = .391, η_p² = .021, illustrating that the order in which W and M were judged did not influence the W and M reports in the current study. Additionally, there was no main effect of Order, F(1, 36) = 1.711, p = .199, η_p² = .045; no interaction between Delay and Order, F(1, 36) = .839, p = .366, η_p² = .023; and there was no three-way interaction between Report Type, Delay, and Order, F(1, 36) = .980, p = .329, η_p² = .026.

Order effect and confidence ratings

We did not observe any main effects or interactions related to the Order variable, p > .30 (see Table 2 for ANOVA results), suggesting that confidence ratings associated with the W and M did not vary with the order in which the W and M tasks were presented.

Results summary

We observed that the W and M’s binding strength differentiated on the basis of Timepoint and structural Knowledge, but not on the basis of Order. With respect to Timepoint, the W report shifted in the direction of the tone more so than the M report. In fact, M was not impacted over time. On the basis of Knowledge, we observed that being informed did not release W from binding; the W_Time2 in the informed condition shifted closer to the time of the tone despite having been informed of the manipulation. M behaved differently from W. Not only did the M reports remained consistent over time in the Uninformed condition, but there was also a hint that M might be displaying an anti-binding behavior in the Uninformed/Informed condition. That is, upon being informed the M_Time2 reports were further away from the tone compared to the M_Time1 report in the Uninformed/Informed condition.

W and M also differentiated on the basis of confidence ratings. The corresponding confidence ratings for W were consistent over time, suggesting that the level of confidence was uniformly distributed despite the W reports progressively shifted closer to the timing of the tone. On the other hand, the corresponding ratings for M declined over time in the Uninformed condition, suggesting the possibility that the observers might have implicitly detected a temporal discrepancy between the timing of their action and the ensuing tone. Combined evidence of both the M and W behaviors and their corresponding confidence ratings suggests that some attributes of W and M do not always covary.

Experiment 2

In Experiment 1, the informed portion coincided with Timepoint 2, raising a concern that any observed effects related to Timepoint and Knowledge would be difficult to interpret. To address this concern, Experiment 2 examined another Knowledge condition in which the participants were fully informed from the beginning. As such, any effects related to knowledge would be free from any contributions from the Timepoint variable.

Methods

Participants

Twenty participants (18 years and older) were recruited for the fully Informed condition. Datasets from three participants were identified as outliers and were excluded from the analysis. The data were analyzed in conjunction with the Uninformed data (N = 20) examined in Experiment 1. Thus, a total of 37 datasets were included in Experiment 2 analysis. The number of participants met the criterion based on a priori power calculation for a mixed ANOVA with one between-subjects independent variable and three within-subjects variables with alpha error probability = .05, and power = .9 (G*Power; Faul et al., 2007). This results in a sample size estimate of 36 participants. Participants received course credits or monetary compensation for the study ($10). The study protocol was approved by the Institutional Review Board of the University of California, Davis.

Analyses

Time reports were subjected to two separate ANOVAs to address the potential effects of Timepoint, Knowledge, and Order on W and M’s binding behavior. As in Experiment 1, to address the effects of Timepoint and Knowledge, the W and M data, along with the corresponding confidence ratings, were subjected to a 2 Report Type (W and M) × 2 Delay (5ms and 60 ms) × 2 Timepoint (Timepoint 1 and Timepoint 2) × 2 Knowledge (Uninformed or Informed). To examine the effect of Order, only data from Timepoint 1 were analyzed. Different from Experiment 1, the analysis of the Order effect in Experiment 2 also included the variable Knowledge, resulting in a 2 Report Type × 2 Delay × 2 Knowledge mixed ANOVA.

Results

Timepoint, knowledge, and binding

As in Experiment 1, to evaluate the temporal binding behavior on the basis of knowledge and timepoints, we subjected the reported times to a 2 Report Type (W and M) × 2 Delay (5 ms and 60 ms tone delays) × 2 Timepoints (Timepoint 1 and Timepoint 2) × 2 Knowledge (Uninformed or Informed, between-subjects) mixed ANOVA. There was no four-way interaction effect between the variables, F(1, 35) = 2.780, p = .104, η_p² = .074. Furthermore, there was no three-way interaction between Report Type, Delay, and Timepoint, F(1, 35) = 3.080, p = .080, η_p² = .081, nor between Delay, Timepoint and Knowledge, F(1, 35) = .611, p = .440, η_p² = .017, between Report Type, Timepoint, and Knowledge, F(1, 35) = 1.450, p = .237, η_p² = .040, and between Report Type, Delay, and Knowledge, F(1, 35) = 3.167, p = .084, η_p² = .084.

The significant results observed in this analysis were simple main effects and two-way interactions. Consistent with Experiment 1, we observed that the W report (128 ms before keypress, SE = 18) was earlier than the M report (57 ms before keypress, SE = 6), F(1, 35) = 15.541, p < .001, η_p² = .307. We also observed a main effect of Delay, F(1, 35) = 19.821, p < .001, η_p² = .362, illustrating that the tone manipulation effectively influenced the reported timing of action and intent. When the tone was delivered after a brief 5-ms delay, the mean reported time (98 ms before keypress, SE = 10) was earlier than when the tone was delivered after a 60-ms delay (87 ms before keypress, SE = 9). These two main effects serve as evidence of replication that W is generally perceived as preceding M (e.g., Libet et al., 1983), and that postaction event, such as a tone, can influence the judgments of the timing of intent and timing of action.

In addition to these simple main effects, we also observed an interaction effect between Report Type and Delay, F(1, 35) = 9.280, p = .004, η_p² = .210 such that the W report was greater influenced by the tone delay manipulation than the M report; that is, the W_5ms (92.70 ms, SE = 9.12) was judged as earlier than the W_60ms (83.30 ms, SE = 9.35); t(36) = 3.576, p = .001, B₁₀ = 30.768, whereas the M_5ms (58.10 ms, SE = 5.99) did not differ from the M_5ms (53.5 ms, SE = 6.98), t(36) = 1.564, p = .126, BF₀₁ = 1.861. The results provide key evidence that the tone manipulation affects the W and M reports differently, leading to an interpretation that W is not always derived from M.

Although we did not observe an interaction effect between Knowledge and Report Type, F(1, 36) = 2.312, p = .137, η_p² = .060, suggesting that W and M were reported in the same manner despite being informed of the tone manipulation, the ANOVA revealed a main effect of Knowledge, F(1, 35) = 6.553, p = .015, η_p² = .158, such that the mean reported time in the Informed condition (116.9 ms before keypress, SE = 18.70) was significantly earlier than the mean reported time in the Uninformed condition (68.00 ms before keypress, SE = 7.69), t(35) = 2.560, p = .015, BF₁₀ = 3.690. The results showed that being informed of the tone manipulation has a broader anti-binding influence on both W and M rather than on one report type alone.

As in Experiment 1, we observed a main effect of Timepoint, F(1, 36) = 8.084, p =.007, η_p² < .188. The mean reported time at Timepoint 1 (96.1 ms before keypress, SE = 11.00) was earlier than at Timepoint 2 (84.9 ms before keypress, SE = 98.00), t(36) = 2.905, p = .006, BF₁₀ = 6.277. However, unlike Experiment 1, there was no significant interaction effect between Report Type and Timepoint, F(1, 35) = 3.116, p = .086, η_p² = .082. We speculate these differences might be attributed to when the participants were informed. In Experiment 1’s Uninformed/Informed condition, the participants were told of the manipulation midway into the experiment whereas in Experiment 2’s Informed condition, the participants were made aware of the manipulation at the start of the experiment. The interaction effect observed in Experiment 1 might be attributed to the abrupt paradigm shift when the participants were informed halfway through, resulting in a different temporal binding behavior. In contrast, the absence of an interaction effect in Experiment 2 might reflect how information about the tone manipulation became increasingly less effective, and in turn W and M judgments became increasingly more reliant on the tone itself over time

Timepoint, knowledge, and confidence ratings

We examined whether W and M could be differentiated on the basis of confidence level. The confidence ratings were subjected to a 2 Report Type (W or M) × 2 Delay (5 or 60 ms) × 2 Timepoint (Timepoint 1 and Timepoint 2) × 2 Knowledge (Uninformed, Informed; between-subjects) mixed ANOVA (Table 3, right panel). As in Experiment 1, there was a three-way interaction effect between Report Type, Timepoint, and Knowledge, F(1, 35) = 13.543, p < .001, η_p² = .279, along with a two-way interaction between Report Type and Timepoint, F(1, 35) = 14.885, p < .001, η_p² = .279, and a two-way interaction between Report Type and Knowledge, F(1, 35) = 21.036, p < .001, η_p² = .375. Given that the same Uninformed data analyzed in Experiment 1 were reanalyzed here, the statistical results for the Uninformed data thus were the same as those reported in Experiment 1 (i.e., W-based confidence ratings did not differ significantly across the two Timepoints), t(19) = 1.993, p = .061, BF₀₁ = 1.193 (W_Time1 = 5.00, SE = .10; W_Time2 = 5.32, SE = .10). In contrast, the confidence level for M declined (M_Time1 = 5.164, SE = .21; M_Time2 = 4.03, SE = .14), t(19) = 3.943, p < .001, BF₁₀ = 41.172. Overall, greater confidence was given to W (5.16, SE = .05) than to M (4.60, SE = .11), t(19) = 4.540, p < .001, BF₁₀ = 136.01.

Table 3 Experiment 2 ANOVA results (Report Type × Delay × Timepoint × Knowledge) for time reports and corresponding confidence ratings

The key observation was in the Informed condition, which revealed a different pattern from the Uninformed condition. We observed that neither the W nor M ratings varied with Timepoint. Specifically, W ratings at Timepoint 1 (W_Time1 = 5.00, SE = .24) and Timepoint 2 (W_Time2 = 5.13, SE = .26) did not differ from one another, t(16) = .930, p = .183, BF₀₁ = 2.753. Similarly for M, the ratings also did not change significantly between Timepoint 1 (M_Time1 = 5.32, SE = .22) and Timepoint 2 (M_Time2 = 5.42, SE = .22), t(16) = .791, p = .441, BF₀₁ = 3.051. Comparison between W and M ratings showed marginally lower confidence level associated with the W reports (5.07, SE = .24) than to the M reports (5.37, SE = .21), t(16) = 2.108, p = .051 but Bayes factor analysis revealed this did not meet the significant criterion (BF₁₀ = 1.446).

The results from both Experiments 1 and 2 suggest that the W-based confidence ratings are stable and are minimally affected by knowledge about the tone manipulation. M-based confidence, on the other hand, seem to be adjusted based on knowledge. In one case, when the participants became informed of the tone manipulation midway into the experiment, their M-based confidence ratings increased upon this knowledge (Experiment 1). In the other case, when prospectively informed, the M-based confidence remained the same (Experiment 2). These observations provide metacognitive differences between the W and M reports, adding to the perspective that W and M do not always covary.

Order effect and temporal binding

We examined the possible effect of Order on temporal binding by subjecting the time reports acquired at Timepoint 1 to a 2 Report Type (W or M) × 2 Delay (5 or 60 ms) × 2 Knowledge (Uninformed or Informed; between-subjects) × 2 Order (W-before-M, M-before-W; between-subjects) mixed ANOVAs (Table 4). Note: Unlike Experiment 1, the inclusion of the Knowledge variable in this analysis was necessary because the Timepoint 1 data consisted of both informed and uninformed trials. The ANOVA revealed a four-way interaction effect between Report Type, Delay, Order, and Knowledge, F(1, 33) = 12.995, p = .001, η_p² = .283. Further examination of each Knowledge condition revealed a three-way interaction between Report Type, Delay, and Order in the Informed condition, F(1, 15) = 14.186, p = .002, η_p² = .486, but not in the Uninformed condition, F(1, 18) = .312, p = .584, η_p² = .017. Within the Informed group, we observed the following. For the W-before-M group, the W report shifted with the tone delay such that the W_5msTime1 report (236.67 ms before keypress, SE = 71.98) was earlier than the W_60msTime1 report (177.78 ms, SE = 58.82), t(8) = 3.170, p = .013, BF₁₀ = 5.106, illustrating that the tone manipulation was influential on W. The M report on the other hand was not influenced by the tone delay manipulation; that is the M_5msTime1 report (55.33 ms before keypress, SE = 9.62) was not statistically earlier than the M_60msTime1 report (65.11 ms, SE = 10.20), t(8) = .562, p = .589, BF₀₁ = 2.722. In contrast, for the M-before-W group, we did not observe a tone delay influence on either the W nor the M report. The M_5msTime1 report (95.38 ms before keypress, SE = 27.57) was not significantly different from the M_60msTime1 report (86.53, SE = 32.73), t(7) = 1.162, p = .283, BF₀₁ = 2.86. Similarly, The W_5msTime1 report did not differ from the W_60msTime1 report, t(7) = .300, p = .773, BF₀₁ = 2.330.

Table 4 Experiment 2 ANOVA results (Report Type x Delay x Knowledge x Order) for time reports and corresponding confidence ratings (data drawn from Timepoint 1 of the Uninformed and Informed conditions).

Order effect and confidence ratings

We observed an interaction effect between Delay and Order, F(1, 33) = 9.927, p = .003, η_p² = .231. In the W-before-M condition, the mean confidence rating for the 5-ms delay trials (5.19, SE = .15) was the same as the mean rating for the 60-ms delay trials (5.15, SE = .15), t(18) = .885, p = .388, BF₀₁ = 2.978. In contrast, in the M-before-W condition, the mean confidence rating for the 5-ms delay trials was lower (4.82, SE = .15) than the 60-ms delay trials (4.97, SE = .15), t(18) = 2.904, p = .010, BF₁₀ = 5.340.

Results summary

The goal of the current study was to identify instances in which W and M differentiate in the context of temporal binding. In the current experiment, we observed that the W and M’s binding strength differentiated on the basis of Delay, with W being more influenced by the tone manipulation than M (consistent with Banks & Isham, 2009, 2010). The focused examination of the Order effect also revealed that the interaction was more prominent in the W-before-M condition than in the M-before-W condition, suggesting the involvement of Order as proposed by Dominik et al. (2017).

With respect to metacognition, and similar to the Uninformed/Informed condition in Experiment 1, we did not observe an interaction effect on the confidence ratings between Report Type and Timepoint in the Informed condition in Experiment 2 (but the interaction effect was observed in the Uninformed condition). Thus, we reiterate here the interpretation that when informed, observers are simply confident in both of their W and M judgments over time. However, when uninformed, the W and M-based confidence ratings are differentiable: while observers are consistent in their confidence of the W performance, their confidence level declines in the M performance over time. This observation further implies that we may be more implicitly sensitive to the temporal discrepancy between the timing of action and the tone than between the timing of intent and the tone.

Discussion

The current study examined how the reported time of intent (W) and the reported time of action (M) differentiate under different circumstances related to temporal binding. Strength of binding was evaluated based on whether the reported times shifted systematically with the timing of the misleading tone, and whether this shift was more pronounced over time and when the participants were unaware of the tone being briefly delayed. Different binding strengths and behaviors associated with the W or M reports would be indicative that W and M are differentiable.

Differentiating the perceived timing of intent (W) and timing of action (M) on the bases of binding behaviors and confidence ratings

The overall results suggest that the binding strength for W and M differed under different experimental conditions. For instance, W shifted systematically with the timing of the tone more so than did M. W’s susceptibility to the tone manipulation was consistent across all three structural Knowledge, emphasizing that being informed about the tone manipulation does not reduce the effect of the tone manipulation on the W report.

In addition to the timing of the tone, the W reports also became more bound to the tone as the experiment progressed. Importantly, this binding behavior was observed in all three Knowledge conditions, suggesting that despite being informed of the tone manipulation, the participants continued to infer from the tone to report W. Moreover, the associated confidence ratings for W were consistent over time (with a hint of an increase in the Uninformed condition). These observations suggest that the participants generally felt confident in the accuracy of their W report, perhaps up to the point where W was unaffected by the knowledge about the tone manipulation. We speculate this might be because intent is believed to be of an internal origin, and therefore, we feel more confident of its timing. So, despite having been informed of the tone manipulation, we consistently project great confidence to W-related attributes.

In contrast to W, the M reports did not vary over time in the Uninformed and Informed conditions. The M reports were consistent across the two timepoints (although, we note that numerical data from the Uninformed/Informed condition hinted to the trend that the M reports were further away from the tone after having been informed at Timepoint 2, suggesting a form of anti-binding; Matute et al., 2017). In addition, while the M reports remained consistent over time, the corresponding confidence ratings appeared to fluctuate under different knowledge conditions. In the Uninformed condition, the confidence ratings associated with the M reports declined over time. The reduction may reflect the participants’ implicit awareness of temporal discrepancy between the time of keypress and the time of the tone. Work by Rigoni et al. (2010) similarly reported no explicit evidence of discrepancy detection but did report evidence at the neural level via event-related potentials. The confidence ratings associated with M may similarly provide an indirect indication that a discrepancy has been detected.

Overall, W and M appear to differentiate on the basis of binding strength and confidence level. These differences imply that W and M are not always codependent and that they are metacognitized differently when the observers do not have external information or knowledge. If W were exclusively inferred from M, their binding patterns and the corresponding confidence ratings should have been equated.

A note on structural knowledge and W

We had predicted that knowledge of the tone delay manipulation would benefit the report of the timing of intent (W) more so than of the timing of action (M) given that the assumption that the M reports would rely more on sensorial input (e.g., tactile or visual input). However, we did not observe a strong impact of knowledge on the W report. In fact, the hint of anti-binding behavior was observed in the M reports rather than in the W reports in Experiment 1, and that the anti-binding of W was somewhat of a slow process as seen in Experiment 2. There are several speculations as to why knowledge played a minimum role on the W reports but might have benefited the M reports. First, it is possible that the participants felt they were provided with sufficient sensory information to judge the timing of their action. Thus, when informed of the tone manipulation, it was easier to dissociate from the sensory tone to judge M. On the other hand, there are not as extensive sensory inputs for the timing of intent, and despite the participants having been informed of the tone manipulation, the participants might have chosen to continue to rely on the tone as a reliable sensory anchor for time W (e.g., upon being informed at Timepoint 2 in the Uninformed/Informed condition; and at Timepoint 1 in the Informed condition), and overtime, this binding tendency grew, displaying a form of adaptation.

Additionally, in Experiment 1, the participants were informed of the tone manipulation in the second half of a study session. It is possible that having been alerted to the manipulation midway into the experiment did not provide sufficient time for the participants to re-anchor or replace the cues previously used to formulate the M and W reports. Support of this interpretation can be drawn from the Informed condition which showed that W in the W-before-M condition was influenced by the tone at Timepoint 1, but W was dissociated from the tone influence at Timepoint 2. A theoretical framework that corroborates with this idea is the Attended Intermediate-Level Representation theory (AIR) offered by Prinz (2005). Temporal index of consciousness (i.e., W) in the Libet task is postulated to reside in the lower level of representation (e.g., Mylopoulous, 2015) and is brought forth to the intermediate, conscious level by attention (i.e., when the participants are asked to report W). When informed of the tone manipulation, it is possible that attention to W is heightened, but nevertheless insufficient to cause W to depart from the timing of the tone, and that time may be needed to pull away from binding.

Order effect

Based on Dominik et al. (2017), we also examined how the task order might affect temporal binding for W and M. We observed the interaction effect in the Informed condition such that the W and M reports followed the same reporting pattern in the M-before-W condition, but not in the W-before-M condition. Such finding though could be interpreted in two ways. First it could suggest that W was inferred from M when the W report followed the M report. However, the observed interaction could also reflect how W and M responded differently upon the knowledge of the tone delay manipulation. Given that the W report still relied significantly on the tone despite having been informed of the tone delay (as observed in the W-before-M condition), it might be the case that participants simply could not immediately ignore the tone, and that longer time was needed to overcome the tone influence. M, as we have seen from former findings (e.g., Banks & Isham, 2010), is not as influenced by the tone as W is. Thus, upon being informed of the tone manipulation, we speculate it was easy for M to be released from binding immediately (as seen in both the M-before-W and W-before-M conditions). In the M-before-W condition, it is possible that having performed the M task might have helped the participants learn to avoid the tone influence, and subsequently were able to do so in the W task that followed. Given these possibilities, the ordering of tasks appears to have an effect, but further examination is needed to understand its operation especially within our relatively complexed experimental setup.

Is W completely separate from M?

Although our findings suggest that W and M are differentiable, we believe that the W reports may still rely on the M reports in some instances. Previous work by Dominik et al. (2017) and Sanford et al. (2020) interpreted W and M to be undifferentiated, reasoning that the W reports are inferred from the M reports. Alternatively, we speculate that the inference from M is done only when the participants deem M to be a reasonable cue. In a traditional Libet task, as employed by Dominik et al. and Sanford et al., the only cue available for formulating W is the timing of action, and this may be why W is observed as dependent on M. Different from these traditional Libet tasks, the current study provided additional information—namely, the timing of the tone—that potentially served as another reasonable cue for a W report, allowing W to depart from M on several occasions including when compared over the course of the experiment and on the basis of confidence ratings.

Is there a real W?

The idea that W is reconstructed is not new, but whether there exists a veridical W remains an unanswered question. From our own research (Banks & Isham, 2009), we have suggested that W (as well as M) is partially inferred, and the final reports are likely the product of influences derived from external cues and internal awareness. The hints of this possibility are drawn from research that illustrates an implicit detection of a temporal mismatch when performing a temporal binding task via the tone manipulation procedure (Rigoni et al., 2010). In this event-related potential study, the authors observed that the participants displayed binding behaviors, suggesting that they were susceptible to the tone manipulation. However, the corresponding ERP signals revealed error-related negativity (ERN) and the feedback-related negativity (FRN), suggesting that the temporal discrepancy between the timing of action and the delayed tone was detected at the neural level. This neural evidence further supports the perspective that a veridical W may exist, but that it may only be indirectly accessible.

In the current study, we employed confidence ratings as a potential implicit measure that might lead to the identification of a veridical W. Unfortunately, we did not find that confidence ratings were sufficiently sensitive to detect a temporal discrepancy related to the W reports, but they might be sufficient for the M reports. We hope future investigations may be able to utilize our results to examine the extent in which metacognitive or implicit measures such as confidence ratings can be used as an index of temporal awareness.

Limitations and future directions

What are the underlying mechanisms of W and M?

The current study provides supportive evidence that W is differentiable from M. Although there is sufficient evidence to suggest that they differ under some circumstances as shown in the current study (see also a meta-analysis by Braun et al., 2021) which illustrates that W, but not M, is moderated by the number of trials, the mechanisms underlying these differences remain speculative and further investigation is needed. Additional evidence may provide insights toward identification of a veridical W as well.

What does confidence represent?

We observed that the confidence ratings associated with W was consistent whereas those associated with M fluctuated. However, it is important to note that such discrepancy needs additional investigation to fully support the perspective that W and M are distinct, and to ensure that the discrepancy is due to the complexity of introspection. A future study could consider another objective measure, such as error-detection ERP components, to assess whether the detection of the timing error of W and M differs, and whether this could be informative in differentiating W and M.

Alternative explanations to sensory adaptation

In the current study, we have demonstrated that W and M differentiate on the basis of the Timepoint variable. We built our hypothesis on the idea of sensory adaptation (Stetson et al., 2006) and interpreted our results to suggest that W and M binding strength differs because one may be more sensitive or responsive to the sensory input (i.e., to receive the tone repeatedly over time). Alternatively, it is also possible that the shift reflects a general tendency to become less anticipatory, perhaps achieving automaticity. We recognize that these explanations need further experimentation and may provide greater insights as to how W and M differ.

Motion biases

The temporal binding task used in our study relied on the assumption that the tone indicates the timing of action. However, an auditory-based representational momentum may also be involved. Representational momentum refers to an inaccurate perception that the final position of a moving object is further along its path than it actually is. The Wundt clock used in the current study, as well as in many other Libet-based paradigms, consists of a rotating spot that traverses in a circular motion, possibly engages a representational momentum effect. Work by Teramoto et al. (2010) reported a representational effect facilitated by sound. If being informed of the auditory tone manipulation could help correct the M report as observed in the Informed block, we could raise the question of whether such knowledge also helped minimize representational momentum bias when reading the clock.