Studying sense of agency online: Can intentional binding be observed in uncontrolled online settings?

Highlights

• Intentional binding is robust to the use of an online sample.

• The tone-binding effect seems to be larger than the action-binding effect.

• Explicit judgements of agency correlate with intentional binding.

• Personality traits do not correlate with intentional binding.

Abstract

Intentional binding is often used as an implicit index of the sense of agency. However, intentional binding research has primarily been conducted in controlled lab environments. During the COVID-19 global pandemic, there has been a shift to implementing studies using online platforms and it is an open question whether the intentional binding effect can be found using an online experimental set-up and participant sample. Here, we address this question by asking online participants to complete the Libet clock version of the intentional binding task, which we make freely available to researchers as a jsPsych (De Leeuw, 2014) plugin. Intentional binding was observed in the form of later keypress estimates and earlier auditory tone estimates, when the auditory tone followed the keypress. These findings confirm that intentional binding can be assessed in online contexts. We discuss these findings in relation to the broader intentional binding literature.

Introduction

The sense of agency, the fundamental feeling of control over one’s actions and the consequences of those actions, is a core part of the human condition (e.g., Moore & Obhi, 2012). In fact, a consistent lack of a sense of agency is associated with brain disorders, such as corticobasal syndrome and schizophrenia (e.g., Moore, 2016, Wolpe et al., 2014). As such, it should come as no surprise that the past two decades has seen a surge of research by cognitive scientists and neuroscientists in this area (for a recent review, see: Haggard, 2017). A key factor in this surge was the discovery of intentional binding: the temporal illusion of perceiving the gap between one’s actions and the consequences of those actions as closer in time compared to when such actions are not intentional or do not occur at all (e.g., Haggard et al., 2002, Haggard, 2017, Moore and Obhi, 2012).

Initial evidence for intentional binding was provided by Haggard et al. (2002). In their study, they used the Libet clock method wherein participants were tasked with observing a conventional clock face in order to judge the onset time of one of two events: their keypress or an auditory tone. In the operant condition, their keypress would always be followed by an auditory tone. In the baseline condition, their keypress or the auditory tone would occur alone. As such, participants made four types of perceptual judgements: the onset time of their keypress with no subsequent audio tone (i.e., Baseline-Keypress), the onset time of an audio tone with no preceding keypress (i.e., Baseline-Tone), the onset time of their keypress with the audio tone following (i.e., Operant-Keypress), and the onset time of the audio tone after their keypress (i.e., Operant-Tone). Results showed that participant estimates for the Operant-Keypress were perceived later in time relative to the Baseline-Keypress, while the Operant-Tone were perceived as earlier in time relative to the Baseline-Tone. Furthermore, in a critical control experiment, a separate set of participants were induced to make involuntary movements (via TMS) that resulted in perceptual shifts in the opposite direction. Given these results, Haggard et al. (2002) concluded that, “Our results show that truly operant intentional actions (voluntary action, then tone) elicit perceptual attraction or binding effects. This effect associates or binds together awareness of the voluntary action with awareness of its sensory consequence, bringing them closer in perceived time.” (pg. 384). They speculate that this intentional binding may be related to “normal agency” (i.e., the sense of agency).

Follow up studies have corroborated the intentional binding effect and further clarified its relationship with the sense of agency (for a recent review, see Haggard, 2017). The aim of the current study is to add to this growing field by testing whether the intentional binding effect can be observed using an online sample. There are multiple factors that motivate the need for such a test. First, as far as we are aware, the vast majority of previous work on intentional binding (measured via the Libet clock method) has been conducted in a controlled laboratory setting; this is a problem as a person’s sense of agency is not confined to pressing a button in a quiet room with minimal distractions. Exceptions to this is work by Garaizar et al., 2016, Cubillas et al., 2020 who showed that action-binding effects (perceiving one’s action as later in time if it is paired with some outcome) were possible to find using an online sample. However, these studies exclusively looked at action-binding, so it is unclear whether tone-binding (perceiving the outcome of an action as earlier in time if it is preceded by an action) would be successfully replicated using an online sample. Furthermore, these studies compared the delay (e.g., immediate vs. 500 ms) between action and outcome as a measure of action-binding; however, it seems to be more common to compare action-outcome pairings with action-alone and outcome-alone baselines. As such, it is still an open question whether intentional binding can be observed in an online sample while using methodological parameters that more closely resemble the original findings (Haggard et al., 2002). Relatedly, with the COVID-19 pandemic disrupting normal lab functions (i.e., no in-person testing), researchers have pivoted to the use of online experiments to continue their work. However, it would be a mistake to assume that intentional binding effects will generalize to this new (noisier) context, as online experiments often mean that the experimenter has little to no control of the environment the participant is completing the experiment in. As such, it is important to test whether the basic intentional binding effect is found using an online sample. Finding that it does opens up future avenues of research using the online medium. Finally, although there are a number of studies showing intentional binding effects, the sample size is often limited due to time constraints and lab space; the use of small sample sizes may be problematic in ascertaining an accurate estimate of effect sizes (especially outside of a well-controlled lab setting). Overall, then, the use of a (relatively) large online sample to test the robustness of the intentional binding effect outside of a lab setting will provide us with novel insights into its generalizability, effect size, and its feasibility as a subject of experimental inquiry outside of well-controlled lab environments.

We created an online version of the Intentional Binding task using the Libet clock method (described above) via JavaScript/jsPsych (available for download at github.com/kinleyid/jspsych-libet; see methods section below for more details) wherein participants completed four conditions: Baseline-Key, Baseline-Tone, Operant-Key, and Operant-Tone. In the Baseline-Key condition, participants simply pressed a key while watching a rotating clock hand on a clockface and estimated the position of the clock hand at the onset of their key press. In the Baseline-Tone condition, they waited for an audio tone to play while watching a rotating clock hand on a clockface and then estimated the position of the clock hand when the onset of the tone occurred. In the Operant-Key and Operant-Tone conditions, participants pressed a key that also triggered an audio tone; in the former condition, participants estimated the timing of the keypress, while in the latter they estimated the timing of the tone. Note that intentional binding has also been studied using the “interval estimation” method (e.g., Engbert et al., 2007, Moore and Obhi, 2012), wherein participants simply estimate (often in milliseconds) the gap between an action they made (i.e., keypress) and a sensory consequence (i.e., an auditory tone). We opted to use the Libet clock method for this study as it has built-in baseline conditions with which to compare the experimental (i.e., operant) conditions. Furthermore, unlike the interval estimation method, the Libet clock allows for the assessment of both action binding effects (i.e., perceptual shift of voluntary keypress towards tone effect) and tone binding effects (i.e., perceptual shift of tone effect towards the voluntary keypress).

Given that this study is a direct replication of previous work on intentional binding, there are a number of directional predictions that we can make. In particular, we predict that estimates during the Operant-Key condition will occur later compared to the Baseline-Key condition while estimates during the Operant-Tone condition will occur earlier in time compared to the Baseline-Tone condition.

As a final point, we also had participants complete the Big 5 inventory as well as a measure of explicit sense of agency (i.e., “On a scale of 1–10, to what extent do you feel like you caused the tone?”). We added the Big 5 inventory to test whether personality traits influence intentional binding effects – a topic that, as far as we are aware, has not been explored using the Libet clock method. Lab experiments may also be limited in studying this topic as correlations often need large sample sizes relative to experimental methods. As such, the current study is in a unique position to shed light on this issue, which may lead to interesting insights for future research. We added the measure of explicit sense of agency (i.e., Judgements of Agency) because questions about the relationship between intentional binding and the sense of agency is currently being debated (e.g., Imaizumi and Tanno, 2019, Kirsch et al., 2019). One method that has been used to help answer this question has been to use explicit measures of the sense agency and to see if such measures correlate with intentional binding (presumably providing convergent validity for the measure). As such, we thought it prudent to add this measure. Overall, however, these measures were added for exploratory purposes and do not constitute the main question in this study.