Testing sensorimotor timing across age and music experience in a real-world environment

Hélène Serré; Keith Harrigian; Se-Woong Park; Dagmar Sternad

doi:10.1038/s41598-026-38073-x

. 2026 Feb 11;16:8300. doi: 10.1038/s41598-026-38073-x

Testing sensorimotor timing across age and music experience in a real-world environment

Hélène Serré ^1,^✉, Keith Harrigian ², Se-Woong Park ³, Dagmar Sternad ^1,⁴

PMCID: PMC12966301 PMID: 41673201

Abstract

Rhythmic ability has been studied for more than a century in laboratory settings testing timed finger taps. While robust results emerged, it remains unclear whether these findings reflect behavioral limitations in realistic scenarios. This study tested the synchronization-continuation task in a museum with 335 visitors of a wide variety of ages (5-68yrs), music experiences (0-40yrs) and educational and cultural backgrounds. Adopting a dynamic system’s perspective, three metronome pacing periods were anchored around each individual’s preferred tempo: the preferred tempo, 20% faster, and 20% slower. Key laboratory findings were replicated and extended to a large age range and individuals with only informal music training: timing error and variability decreased during childhood and increased in older adults, and they were lower even with only moderate music experience. Consistent with an oscillator perspective, timing at non-preferred tempi drifted toward their preferred rate. Overall, these findings demonstrate that some of the main results on sensorimotor timing from controlled experiments are robust to noisy, naturalistic settings.

Subject terms: Neuroscience, Psychology, Psychology

Introduction

The human sense of timing is one of the oldest phenomena that garnered interest in experimental psychology dating back to Stevens, Scripture and others in the 19^th century^1–3. Over more than 100 years, a plethora of studies scrutinized how humans time their motor commands, specifically in rhythmic timing, as it underlies the human fascination with music⁴. Legions of experiments have employed the extremely simple movement – finger tapping – to test the ability to react, predict and entrain to auditory or visual stimuli. In 1886, Stevens introduced the synchronization-continuation paradigm, where participants tap their index finger on a key in synchrony with a metronome¹. A synchronization segment is followed by a segment where the external cue is withheld, and the participant continues tapping at the same pace^5–10. This finger tapping task has remained a widely-used experimental paradigm until the present day^11,12, and has been used to address a large variety of questions including neuroimaging methods in humans and animals^13,14. Different theoretical models have been proposed to account for the findings, from Wing and Kristofferson’s clock-motor timing¹⁵ to network and oscillator perspectives^5,16–18.

This body of experimental research has accumulated a considerable number of observations, grounded in a relatively confined set of variables. The main measurements are the time difference or error between the beat and its corresponding tap, and the variability of the inter-tap interval (ITI)¹⁵. This paradigm has revealed many fine-grained features of sensorimotor timing. The present study re-examined these features in a large diverse cohort in the “Living Laboratory” of the Museum of Science in Boston to test the robustness of these observations in a naturalistic environment (Figure 1). In this outreach activity, we invited a very diverse sample of museum visitors of all ages, music experience, educational, professional, cultural and ethnic background. Our experimental approach was grounded in a dynamic systems perspective, positing that the behavioral rhythm emerges from a neurom echanical oscillator with attractor properties^17,19. We therefore designed the experiment to individualize the timing protocol to each participant’s preferred period, or spontaneous motor tempo (SMT), and then defined pacing periods in relation to their SMT. To set the stage for our study, the main results from the previous plethora of studies are briefly summarized.

Fig. 1 — Experimental data collection in the Living Laboratory at the Museum of Science in Boston.

Spontaneous motor tempo

Sensorimotor synchronization is least variable when moving at one’s preferred rhythm, often called spontaneous motor tempo (SMT)^7,9,17. The SMT naturally exhibits inter-individual differences determined by biomechanical properties of the limbs^7,20,21, physiological factors, such as energy consumption^22,23, the tapping force²⁴, and the time of day^25,26. Individuals are consistent over the same task, but also across similar tasks like finger tapping and playing piano²⁷. A less variable SMT is observed in musicians, even though they are trained to synchronize with many different tempi⁵. Further, the SMT slows down with age and music training^9,10,28. Finally, when enforcing faster or slower paces, participants drift towards their SMT when the rhythmic cue disappears^5,7.

Our study also anchored the metronome intervals around the SMT, thereby relating people’s timing performance to their own preferred period. From the vantage point of a dynamical systems perspective, we hypothesize that the preferred tempo is a stable attractor^5,7,19, and when oscillating at a different period, but within the oscillator’s basin of attraction, participants perform with more error and variability (less stability) and drift towards their preferred period. The negative asynchrony, i.e., the tendency to tap prior to the metronome beat, is also expected to shorten for faster periods and lengthen for slower periods¹¹.

Effects of music experience

Many studies were motivated by how music practice affects the sense of time and tested professionals with many years of extensive training^29,30. Unsurprisingly, sensorimotor synchronization is enhanced in musicians due to their natural aptitude and musical training¹⁹. The extensive practice fine-tunes the neural mechanisms underlying timing and coordination, leading to superior motor timing, documented in lower ITI variability and shorter asynchronies^19,31–35. Beyond synchronization, maintaining a steady tempo without cues is also considered a significant quality when playing music. Despite its importance for musicians, this drift in the continuation segment has received little attention in musically-trained individuals⁵. Based on previous studies in a controlled environment^5,19, we expect music experience to enhance tempo flexibility around the SMT¹⁹, and to show less drift towards their preferred period.

Effects of age

Synchronization abilities improve significantly as children grow up and typically reach adult levels by 8 to 10 years of age^{9,10,28,35,36}. Before that age, ITI variability is higher and asynchrony is longer. Also, children drift toward their SMT faster than adults⁹. Not unexpectedly, older adults (>50 years old) are more variable than younger adults⁹. The SMT also evolves with age, particularly exhibiting a slowing in older adults. While it has been shown that children exhibit faster SMT, these drifts were calculated in reference to the same set of intervals across all participants (150, 225, 337, 506, 759, 1,139, and 1,709 ms), not scaled to their preferred period⁹. Setting the metronome periods with respect to the individually preferred period, we expect to confirm the faster drift in children.

This brief review of findings presents our current understanding of timing. However, it is noteworthy that these effects are very subtle. The ITI variability (reported as coefficient of variation) in 5-75 year-olds varies from 0.2 in the youngest participants to 0.07 in adults for intervals between 200ms and 1600ms^9,37. The difference in mean asynchrony between professional musicians and non-musicians is maximally 40ms³¹. These results are typically obtained from tightly controlled experiments, often conducted in sound-proof cabins with precisely tuned tapping keys, and with participants that can follow precise instructions and are usually attentive and receive financial reward for their participation. While these procedures have yielded reliable data and revealed interesting features of timing and their limitations, such scenarios are a far cry from realistic environments^34,38. Hence, the question arises whether these timing features are strong enough to survive or even matter in the complex uncontrolled real-world environments.

The present study draws upon hundreds of hours of data collected in the environment of an outreach activity in the Museum of Science in Boston. Using a large inter-generational pool of museum visitors who volunteered for a finger tapping task with a synchronization-continuation design, the current study pursued three aims: 1) test the robustness of previously reported basic timing effects in a noisy real-world environment, 2) examine the effects of age in a diverse sample of the population, and 3) investigate the effect of on average largely informal music experience.

Museum visitors completed the finger tapping task with synchronization and continuation segments in three different conditions: at their spontaneous motor tempo (SMT), at a tempo 20% faster and another one 20% slower than their SMT. This design was motivated by a dynamical systems perspective, where the SMT or preferred period is regarded as a reflection of a neural oscillator that is specific to individuals and presents an attractor for adjacent intervals^5,7,17,19,39.

The data collection was performed in an open space of the larger museum area, the Living Laboratory, where museum goers passed by and were spontaneously recruited to participate and learn about ongoing research (Figure 1). Challenges in recording the tapping force data were overcome by a novel algorithm that sensitively detected contacts despite a high degree of variation and noise. Results confirmed that ITI error and variability improved during development and deteriorated for older adults, both in the synchronization and continuation segments. Both sensorimotor synchronization and rhythm maintenance were consistently better with music experience. Our hypothesis based on the dynamical systems perspective on drift was supported: participants indeed drifted towards their preferred period, although these drifts did not differ between musicians, nor across age. Our results demonstrated that even in the little controlled conditions of a museum environment, the main timing effects stood the test of time. The findings are discussed in light of previous studies and through the lens of the dynamic systems’ theoretical framework.

Results

The experiment was part of the Living Laboratory program of the Museum of Science in Boston, an initiative designed to educate the public through active involvement in scientific experiments (Figure 1). All subjects in this experiment were museum visitors and data collection occurred during 4-hour-long exhibitions set up every weekend over 8 months (except the holidays). The 26 exhibit sessions were conducted by a total of 9 experimenters rotating across the sessions. Each data collection session required the team work of 3 students—one for recruiting museum visitors and obtaining informed consent, one for providing information about our ongoing research and administering a short questionnaire about age and muscial training, and one for operating the experimental apparatus to collect the data.

Participants were first asked to tap their finger on the force sensor for 20s at their preferred rhythm, which was analyzed to determine the spontaneous motor tempo (SMT). Two period conditions, 20% faster and 20% slower, were calculated based on the individual SMT. These on-line calculations took maximally 3min. Then, participants were asked to tap to a metronome beat for 15s and maintain the pace for 25s after the metronome terminated. The data collection for each subject was limited to 15min per museum policy. A total of 335 volunteers participated in the study (180 female, 155 male, ages 5-68). Given the uncontrolled nature of the experimental data collection (see Figure 1), some data had to be eliminated from further analysis (see details in the Methods), ending with a total of 236 subjects (130 females, 106 males, ages 6-68).

Spontaneous Motor Tempo (SMT)

A custom-written algorithm calculated the mean SMT including the faster and slower periods and set up the metronome pacing protocol for the immediately ensuing data collection. The SMT was on average 518ms, which is close to the 500ms that is ubiquitously reported in the literature as the preferred interval for production^9,36,40 and perception⁹. Table 1 summarizes the medians and first and third quartiles of the SMT, together with all subsequent performance measures.

Table 1.

Distributions of the computed timing variables (median and first and third quartiles) for the three period conditions and the two segments. They are reported in milliseconds and in percentage of the metronome interval M, with the exception of asynchrony, also reported in ms, drift and ITI variability that are reported in arbitrary units.

Variable	Paced			Unpaced
	Faster	Preferred	Slower	Faster	Preferred	Slower
SMT (ms)	414 (350, 486)	518 (438, 608)	622 (525, 729)
ITI Error (%M)	3.45 (2.75, 4.36)	3.12 (2.3, 4.05)	3.16 (2.54, 4)	5.90 (4.1, 8.3)	5.53 (4.13, 7.9)	5.70 (4.26, 8.35)
ITI Variability	0.029 (0.022, 0.037)	0.026 (0.02, 0.032)	0.027 (0.02, 0.033)	0.031 (0.023, 0.04)	0.031 (0.023, 0.042)	0.029 (0.024, 0.038)
Asynchrony (%M)	-14.1 (-17.7, -9.8)	-16.2 (-22.1, -11.3)	-17.2 (-22.9, -11.5)
Asynchrony (ms)	-68.25 (-94.25, -47.34)	-81.69 (-110.16, -56)	-86.62 (-112.22, -61.25)
Drift (rad/tap)				0.15 (-0.06, 0.41)	0.034 (-0.18, 0.29)	-0.06 (-0.28, 0.22)

Open in a new tab

Figure 2A shows the preferred periods as a function of age, separated for individuals with and without music experience. The figure also highlights the wide variations of the SMT between 233ms and 1122ms, partly due to the uncontrolled environment, partly due to the wide range of individuals who spontaneously interrupted their museum stroll and participated in the experiment. The main effects of age (β=6.22, LRT=4.1, df=1, p=0.04), age-squared (β=-0.10, LRT=6.27, df=1, p=0.01) and music experience (LRT=8.26, df=1, p=0.004) were significant. SMT varied with a parabolic shape over age, and was on average 67ms longer for subjects with music experience (music - no music = 67.5, t=2.87, p=0.004).

Fig. 2 — Distribution of spontaneous motor tempo (SMT) and related pacing periods. A: SMT as a function of age split by musical experience. One point represents one participant (SMT average over one trial). The solid lines indicate the parabolic regression, separated by music experience. Shaded areas represent the 95% confidence interval (estimate -/+1.96 SE). B: Distribution of the three pacing periods. The histograms are displayed with stacked bars.

This inter-individual variability resulted in a wide overlap of the period’s distribution between the different pacing conditions, as seen in Figure 2B. For example, 500ms could be categorized as the fast, preferred or slow period across individuals. Disregarding potentially extreme choices, this overlap illustrates that motor timing is individually grounded: the words ‘preferred’ or ‘spontaneous’ reflects very different durations depending on the person.

ITI error

The error of the intertap interval (ITI) was given by the median of the unsigned deviations of the tapping period from the metronome period for each trial. It was expressed as percent of the target duration to allow pooling all different periods. The main effects of age (β=-0.02, LRT=14.32, df=1, p<0.0001), age-squared (β=0.00028, LRT=10.68, df=1, p=0.001), music experience (LRT=11.62, df=1, p=0.0006) and the interaction between segment and period condition (LRT=9.26, df=2, p=0.01) were significant. Figure 3A shows the ITI errors across age, with a parabolic regression, suggesting that ITI error decreased from children to adults and increased again at older age. The paced segment showed overall smaller errors than the unpaced continuation segment (paced–unpaced=-0.62, t=-24.42, p<0.0001). Figure 3B highlights how musically experienced subjects exhibited lower errors than subjects without music experience (music–no music=-0.16, t=3.44, p=0.007). As seen in Figure 3C, within the paced segment, the faster condition exhibited higher ITI errors than the preferred condition (β=0.15, t=4.69, p<0.0001) and the slower condition (β=0.1, t=3.32, p=0.003).

Fig. 3 — Timing (ITI) error (in percent of the metronome period M) as a function of age, music experience, segment and period condition. A: Timing error as a function of age. One point represents one participant averaged over all conditions and segments. The solid lines indicate the parabolic regression, separated by music experience. Shaded areas represent the 95% confidence interval (estimate -/+1.96 SE). B: ITI error for participants with and without music experience, pooled across all period conditions. The total number of participants in each group is given in the bars. C: ITI error in the paced and unpaced segments separated by the three period conditions. Asterisks indicate significant differences (* p<0.05; ** p<0.01; *** p<0.001). Error bars represent the standard error.

ITI variability

The ITI variability was defined by the quartile variation coefficient. Similar to the ITI error, age (β=-0.02, LRT=21.8, df=1, p<0.0001), age-squared (β=0.0002, LRT=13.5, df=1, p<0.0001) and music experience (LRT=10.65, df=1, p=0.001) had significant effects on ITI variability. There was also a tendency for the interaction between segment and period condition (LRT=5.53, df=2, p=0.06). Figure 4A shows how ITI variability first decreased and then increased with age, expressed as a function of age-squared. Musically experienced participants displayed lower variability (music–no music=-0.12, t=-3.29, p=0.001; Figure 4B). As seen in Figure 4C, in the paced segment, ITI variability was higher in the faster condition compared to the preferred (faster–preferred=0.1, t=3.32, p=0.003) and the slower conditions (faster-slower=0.08, t=2.69, p=0.02). The unpaced segment showed higher variability than the paced segment (paced – unpaced=-0.1, t=-5.83, p<0.0001).

Fig. 4 — ITI variability as a function of age, music experience, segment and period condition. A: ITI variability as a function of age, split by music experience. One point represents one participant averaged over all conditions and segments. The solid lines indicate the parabolic regression, with and without music experience. Shaded areas represent the 95% confidence interval (estimate -/+1.96 standard error). B: ITI variability in participants with and without music experience, pooled for all period conditions. C: ITI variability in the paced and unpaced segments separated by the three period conditions. Asterisks indicate significant differences (* p<0.05; ** p<0.01; *** p<0.001). Error bars represent standard error.

Asynchrony

The asynchrony was the signed difference between the metronome and its corresponding tap onset, with negative values indicating that the tap advanced the metronome. This metric could only be computed for the paced segment. The overall average of asynchrony was -81ms (+-40ms), which was expectedly negative signaling that the tap preceded the metronome beat. This high number compared to reported values in the literature^41,42 was likely due to our continuous force signals, and the algorithm that detected tap onsets and not peaks (see methods)⁴¹. Asynchrony was normalized by the SMT to enable comparison between individuals. As seen in Figure 5, the factors age (β =-0.002, LRT=5.56, df=1, p=0.02), age squared (β=0.00004, LRT=7.26, df=1, p=0.007), and the interaction between music experience and period condition (LRT=10.47, df=2, p=0.005) were significant. Asynchrony became shorter as age increased (Figure 5A) and was shorter with music experience (Figure 5B), but only in SMT (music – no music=2.56, t=2.19, p=0.03), with a tendency in the slower condition (music – no music=2.37, t=1.84, p=0.07). Figure 5B also shows that within individuals with music experience, asynchrony increased for the faster and preferred conditions (faster – preferred=2.21, t=5.3, p<0.0001) and tended to increase between preferred and slower conditions (preferred-slower=0.93, t=2.17, p=0.08).

Fig. 5 — Asynchrony as a function of age, and period condition. A: Asynchrony as a function of age and split by musical experience. One point represents one participant averaged over all period conditions on the paced segment. The solid lines indicate the parabolic regressions. Shaded areas represent the 95% confidence interval (estimate -/+1.96 SE). B: Asynchrony as a function of music experience and period condition. Asterisks indicate significant differences (^.p<0.1; * p<0.05; ** p<0.01; *** p<0.001). Error bars represent standard error.

Individuals with no music experience had a higher asynchrony in the preferred than in the faster condition (faster-preferred=5.84, t=5.28, p<0.0001), but asynchrony in the slower condition was not significantly higher than in the preferred condition. Asynchrony, quantified in milliseconds without the normalization, displayed the same statistical pattern or results within each music condition. However, since musically experienced individuals had a slower SMT, their asynchrony expressed in ms did not differ from individuals without music experience.

Drift

Expressed in degrees per tap, the drift should be negative or positive, depending on whether the participant tapped faster or slower than the metronome period. Figure 6 shows that, as hypothesized, drift significantly differed in the three period conditions (LRT=54.58, df=2, p<0.0001): it was positive in the faster condition (β=9.55, t=6.38, p<0.0001), showing that subjects slowed down their tempo, when they were no longer paced by the fast metronome. In contrast, in the slower condition drift was negative, reflecting the tendency to speed up, although this effect did not reach significance (β=-1.67, t=-1.08, p=0.6). However, this drift in the fast condition was significantly larger compared to the preferred period (faster–preferred=5.89, t=4.54, p<0.0001), just as the drift in the slow period was significantly different from the preferred period (preferred–slower=5.32; t=3.92, p=0.0003). Counter to our hypotheses, neither age, nor music experience significantly impacted the drift.

Fig. 6 — Average drift in the three period conditions. Asterisks indicate significant differences (* p<0.05; ** p<0.01; *** p<0.001). Error bars represent standard error.

Discussion

Over more than 100 years, legions of experiments have employed the extremely simple movement – finger tapping – to test the ability to react, predict and entrain to auditory or visual stimuli. The synchronization-continuation paradigm, first introduced by Stevens in the 19^th century, has generated a body of work on rhythmic timing with a set of well-established findings. These findings were obtained in rigorously controlled experiments, with individuals that were motivated and frequently rewarded. The metrics that describe timing acuity were in the order of milliseconds, and the effect sizes are often much below 100ms. What do these results mean for behavior in the real world? Motivated by the desire to add ecological validity to these findings, this study tested this widely used protocol in a more naturalistic environment. We therefore took the experimental paradigm into a “Living Laboratory” that was far from the typical rigorously controlled laboratory setting and involved a wide variety of individuals with different ages, and different educational, professional, ethnic and cultural backgrounds.

Robustness of basic timing effects

The study took place in a museum exhibition hall and included more than 300 participants between 4 to 68 years old with diverse musical backgrounds. This musical experience included a wide variety of instruments (piano, guitar, violin, saxophone, clarinet, flute, etc.) and ranged from 1 to 40 years of practice, typically confined to leisure activities (with some exceptions). Our results replicated the main findings from the literature: (1) subjects exhibited a SMT close to 500ms; (2) ITI error and variability decreased during childhood and increased again in adolescents and older people; (3) individuals with music experience showed lower ITI error and variability and slightly slower SMTs; (4) the tapping tempi drifted towards the SMT when pacing stopped.

Support for a neural oscillator

When paced at rhythms different from their SMT, individuals drifted back to their individually preferred tempo after the external pacing was stopped. This finding is remarkable as the subjects in our study exhibited an extremely wide range of preferred periods, from 233ms and 1122ms. This drift towards the individually preferred tempo presents strong support for the hypothesis that the preferred period acts as an attractor, i.e., timing is not absolute, but is individually grounded. Natural periods have been recognized as inherent to locomotion, with the preferred stride frequency accounted for by biomechanical and energetic factors^43–45. With this same dynamical interpretation, Yu and colleagues⁷ revealed a drift towards the natural period in a manual task. However, this task involved swinging different hand-held pendulums that had significant inertia and supported biomechanical explanations: the self-chosen period was close to the pendulums’ natural period that required the lowest control and energy expense. When tapping a finger on a light key, biomechanics is not as prominent a predictor of the preferred tempo as the finger’s inertia and required force are almost negligible. Hence, the observed drifts towards the preferred periods could not be explained by mechanical properties of the end-effector. It is therefore likely that the individual tempo ‘choice’ originated in neural dynamics¹⁷.

Variability and asymmetry in sensorimotor timing

Previous studies by Sternad and colleagues^7,39 showed that at periods between 800ms and 1200ms, ITI variability in the paced segment followed a U-shape with its minimum at the SMT. Similarly, Zamm and colleagues found this U-shape in variability for musicians playing paced melodies on a keyboard (with SMTs below 600ms)⁵. In our study, the ITI variability did not exhibit this expected U-shape. However, the faster conditions showed striking consistency in most of the dependent measures, with higher ITI error and variability in the paced segment and positive drift in the unpaced segment. One speculation is that the museum environment and the experience of being a subject in a scientific experiment may have increased arousal in some individuals, provoking an increase in their SMT. If some participants went faster than their ‘real’ preferred period during the initial SMT measurement, their ‘estimated’ SMT may have been somewhere between the faster and ‘preferred’ condition, while the slower condition may not have been much beyond the SMT. This could explain why the slower intervals did not show the expected increase in error and variability to generate the U-shape.

Changes with age

In line with previous work^9,35,46, ITI error and variability also showed a U-shape across the age range. In addition, asynchrony became shorter with age. Both of these results were seen regardless of the period condition, suggesting that timing is subject to general developmental and aging changes. However, drift did not replicate previous findings that showed a decrease from childhood to adulthood⁹. This discrepancy may arise from different calculations: McAuley and colleagues⁹ quantified drift as the average error over all taps during continuation for target intervals from 150ms to 1700ms. In our analyses, the drift was quantified by the relative phase between tap and beat in the last converging portion of the continuation segment (last 8 taps). This choice recognized the drift as a non-stationary process, following earlier results that had shown that drift was an exponential return to the preferred period⁷, consistent with the behavior of a first-order dynamical system. Given that our drift calculation was more in line with the presumed transient process, it is suprising that it did not capture the age effect otherwise present in a previous study. That said, our dataset only included 8 children under 8 years, and only 8 people older than 60 years, while their dataset included 46 participants >60 years and 68 participants <10 years. Further, periods in our study were only 20% away from the SMT; age differences may have only shown up for periods farther away from the SMT.

While several studies reported a slowing of the preferred period with age^9,25,47, in our data SMT was faster among the oldest participants. However, the same increase in SMT was also shown in McAuley et al.’s study⁹ who reported a cubic trend over a very large age range (4-95 years). Nevertheless, it is worth noting that the variance explained by these estimations was respectively 12% in McAuley’s data, and 7% in our data. Hence, although aging influenced the spontaneous motor tempo, other factors such as the environment might have overriden that effect. As mentioned above, a loud and crowded place like a museum may have increased participants’ arousal, and thereby shortened their SMT²⁵.

Positive effects of music experience on sensorimotor timing

In line with previous work¹⁹, ITI error, variability and asynchrony were lower for participants with some music experience, suggesting that musical training increases the flexibility to entrain to both preferred and non-preferred rhythms. Counter to our expectations, our “musicians” did not perform significantly better at maintaining a steady tempo, but drifted in the absence of external cues.

This absence of drift differences may be due to the heterogeneity of music profiles, which included very different experiences and backgrounds. Importantly, we did not require systematic or even professional training. Dalla Bella and colleagues⁴⁸ defined a musician as having at least 7 years of formal training, but found that also informal music experience significantly enhanced motor and perception performance. In the same vein, Scheurich and colleagues¹⁹ reported that a couple of years of systematic training could already significantly improve sensorimotor timing. In our study, 25% of our ‘musicians’ practiced for less than 3 years with a range of practice from 1 to 40 years, with a median at 7+/-8 years. Note also that the experience that our museum participants reported included a wide variety of instruments.

And yet, it is noteworthy that even with this widely varied music experience in our cohort, sensorimotor timing was improved in terms of ITI error and variability, and asynchrony to some extent. In contrast to synchronizing to a beat, tempo maintenance may be a more difficult skill that requires more experience than informal music training. But all taken together, considering a multi-dimensional continuum of music practice may be more appropriate than the binary discrimination between musicians and non-musicians applied in most studies on timing.

Limitations

Our naturalistic setting intentionally relaxed the usual laboratory control and allowed for much more data collection “noise”. The portable equipment also limited the sampling frequency of the force signal to 200Hz which corresponded to roughly 2% of the shortest periods (~250ms). However, ITI error and asynchrony were defined between the force onset and the metronome that had a higher temporal resolution (2000Hz). Nevertheless, the relatively low sampling rate of force may introduce some granularity which reduced the sensitivity of the analysis. Despite this shortcoming, most results were sufficiently robust.

While our study recruited a large variety of participants, we could not target specific individuals in our recruitment per museum policy. Hence, the age distribution was uneven, with fewer children and elderly volunteering than teenagers and middle-age adults, which weakened the generalizability of the findings across the life span. The information about music experience was also provided in a very time-limited adhoc fashion in a brief questionnaire just prior to the data collection. Nevertheless, both age and music experience had significant effects on performance, highlighting the robustness of these results. For any future study on age and music experience it should also be noted that it is not trivial to disentangle the effect of age from the effect of years of music experience. In the ideal scenario, one would need to compare individuals with the same age, and our dataset was not sufficient to address this confound.

The SMT was assessed directly before the data collection proper using only a single trial, which likely impacted the robustness of this metric. Due to practical constraints, we were unable to implement a more fine-grained and reliable method for assessing the SMT, such as the approach proposed by Kaya and Henry¹⁸. In the same vein, to keep the experiment within the permitted 15min, there were only two trials per condition, and the continuation segments were too short to more thoroughly study non-stationary processes such as the drift. Future studies could conduct a detailed characterization of individual music background, a greater number of trials, and a more robust assessment of SMT across multiple sessions. Such “improvements”, however, would bring the study yet again close to the tightly controlled laboratory conditions, rather than the natural adhoc scenarios in which humans have to time their actions in everyday life.

Conclusions

Our study design was motivated by testing laboratory results on sensorimotor timing in a naturalistic environment. Our explicitly uncontrolled study was able to replicate the main previous results on sensorimotor timing, giving them ecological validity that allowed some conclusions and applications in the real world. Some results were consistent with the oscillator hypothesis, particularly the drift observed during the unpaced segment. The fact that even short-term and more casual music experience positively affected sensorimotor timing reinforced that some music training should be fruitful for children to improve their sensorimotor timing. Clinical work with Parkinson patients already showed that music-based gait training not only improved gait parameters, but also perceptual and motor timing performance^49,50. This speaks to providing musical training for neuro-diverse populations, such as autism⁵¹. Considering the rapid development of portable and adaptive technologies⁵², there is an opportunity to exploit these advances for educational and rehabilitative purposes³⁸.

Methods

Participants

A total of 335 volunteers participated in the study (180 female, 155 male, ages 5-68). The inidivudals were of all educational, professional, cultural and ethnic backgrounds. The study was approved by the Institutional Review Board of Northeastern University and the Museum of Science in Boston. All methods were performed in accordance with the relevant guidelines and regulations. All subjects gave informed consent prior to participating and nobody received any remuneration. Everyone also completed a one-page survey that collected demographic attributes, any known physical and neurological disorders, and information regarding previous music experience. This initial reporting should not have taken more than 5 minutes as the total time for study participation was not to exceed 15 minutes per museum policy.

The experiment was part of the Living Laboratory program of the Museum of Science in Boston, an initiative designed to educate the public through active involvement in scientific experiments. All subjects in this experiment were visitors at the Museum of Science in Boston, Massachusetts, that were attracted to the outreach efforts by a collection of students of Northeastern University. Data collection occurred during the 4-hour exhibitions that were set up every weekend between October 2015 and May 2016. The experiment was conducted by a total of 9 experimenters, including 6 undergraduate students, 2 graduate students, and 1 postdoctoral fellow. Each data collection session required the team work of 3 students—one for recruiting museum patrons, one for providing general information about ongoing research, and one for operating the experimental apparatus for data collection. The 9 students rotated their responsibilities each weekend. To ensure that they all followed the same procedures, they underwent several introduction sessions.

Subject elimination

Given the uncontrolled nature of the experimental data collection and because the experimental apparatus had to be assembled and set up anew every weekend, the subjects’ data had to be closely scrutinized and several subsets of data needed to be eliminated from further analysis, ending with a total of 236 subjects (130 females, 106 males, ages 6-68). The following details the elimination procedures:

Based on visual scrutiny of the signals, 11 subjects were removed due to incorrect execution of the task. These subjects either tapped too lightly on the force sensor, were not able to synchronize to the auditory metronome due to distraction, or they forgot to continue tapping after the auditory metronome was silenced. An additional 20 subjects were excluded during post-processing due to data acquisition and processing errors. Specifically, 5 subjects were discarded because the experimental apparatus had been improperly configured. The remaining 15 of these 20 aforementioned subjects were discarded, because there was too much noise in the digital signal to properly identify tap initiations. An additional 22 subjects were excluded prior to statistical testing due to abnormalities related to their SMT. Specifically, 1 subject had a SMT of 1.8 s that differed by more than 5 standard deviations from the average SMT. The other 21 subjects were discarded, because the SMT that was calculated at the start of the experiment differed by more than 10% from the value that was calculated for verification after the end of the experiment. The 10% threshold was chosen based on empirical measurements of the SMT variability in the prior literature^9,53. 46 additional subjects consistently presented outliers defined in the section ‘Statistical analyses’ and were deleted from further analysis.

Following all these eliminations, a total of 236 subjects (130 females, 106 males, ages 6-68) remained in the final analysis. The full age and gender distribution for the analyzed subjects can be found in Figure 7. 128 subjects reported their number of years of music experience between 1 and 40 years, summarized as a median of 6 years and an IQR of 8 years. The frequency of practice per week ranged from 1 to 4 times, with a median of 3 (mean of 2.6) and an IQR of 2 (STD of 1.16). Piano playing and singing were the most practiced music activities, in 33 and 26 subjects, respectively. Note that the number of years overlapped with their age, such that, for example, a 30 year old participant could have maximally 25 years of musical practice.

Fig. 7 — Distributions of 236 participants of age and gender. The insets shows the split by music experience.

Experimental design and procedure

Subjects were seated comfortably at the research station within the Living Laboratory at the Museum of Science (Figure 1). The student experimenter explained the task and provided a pair of noise-cancelling headphones to wear during the testing. The metronome beeps were then played via the headphones. To mitigate the effect of distractions within the public data collection area, subjects were instructed to point their gaze toward the experimental apparatus. Nevertheless, siblings, parents, and friends frequently gathered around the subjects, watching but also potentially distracting them (Figure 1).

In the first stage of the experiment, subjects were instructed to tap their finger on a small force plate for 20s at a pace that they would be able to maintain for at least 5min. Upon completion of this first task, the intervals between the onset of two subsequent taps (ITIs) were computed. The arithmetic mean of the ITIs during the middle 10s of that recording served as an estimate for each subject’s spontaneous motor tempo, or SMT. Based on this estimate, three testing periods were determined as 0.8 (20% faster), 1.0 (SMT), and 1.2 (20% slower) than the SMT.

In the main part of the experiment, subjects were informed that they would hear a rhythmic metronome beep and should tap in synchrony with the beat to the best of their abilities for 15s (synchronization segment). This segment was followed by another 25s when the metronome was turned off and they should continue at the same tempo (continuation segment). Subjects completed a total of 6 trials, with 2 trials for each of the 3 metronome conditions. The 3 metronome periods were presented in random order. Raw data for a sample trial can be seen in Figure 8.

Fig. 8 — Example of a tapping trial in the slower condition. The two blue shades separated by the red dashed line highlight the paced and unpaced segments. Top: Raw force signal. Bottom: Inter-tap intervals (ITIs) clustered around the metronome beats at 550ms in this trial (dotted line). A drift toward the preferred period (460ms, dashed line) in the unpaced segment is visible.

Experimental apparatus

Tapping contacts were measured by a custom-made load cell with a strain gauge, glued on top of a beam that measured the deformation of the beam. To enhance sensitivity to the relatively low forces and deliver accurate and reliable signals, two holes were drilled into the beam under the strain gauge. The measured deformation was then converted into force and recorded at a sampling rate of 200Hz. Data was encoded using a data acquisition card with a pre-amplifier to enhance the signals (National Instruments, NI USB-6343).

To ensure that subjects tapped naturally, the experimenter did not provide explicit directions for how subjects should position their arm or tap on the force plate, such that the magnitude of force, the contact duration of the tap, and the actual finger movement between taps varied quite substantially between subjects. To accommodate the maximum number of museum visitors during the assigned 4-hour data collection session, the sensor could only be calibrated only once at the beginning of each daily session. Consequently, due to occasional noise and drift, standard processing algorithms for detection of the onset of the tap sometimes faced some challenges and warranted a special tap detection procedure.

Detection of tap onsets

Data was extracted from the force signals, with examples shown in Figure 8. When the signal rose from the baseline, the start of the tap was indicated. Sometimes, the finger contact was slightly prolonged giving rise to double-peaks confounding challenging a clear marking of the peaks. Hence, the continuous shape of the signal created some problems for a regular peak finding algorithm (Figure 9A). Therefore, the best landmarks for ITI calculations were the initial finger contact indicated by sharp rises from zero force between taps in the signal.

Fig. 9 — Comparison of the tap detection between two analysis methods of the continuous force signal. A: Applying the standard peak finder function could lead to detection of double peaks within a single tap. B: Using the Gaussian Hidden Markov Model (GHMM) shows how marking double peaks was avoided and how also the onsets of lighter contacts could be reliably detected.

A tap detection algorithm was developed using a Gaussian Hidden Markov Model (GHMM). In the first step, the digital signal was passed through a high-frequency moving-average filter using segments of 50 frames binned into 50 linearly-spaced force intervals to filter out high frequencies of events. The observed signal was treated as a system with two states – a baseline state that existed while the force sensor was at rest, and an active state that was invoked by a finger tap. To infer the latent states from the observed data, the GHMM was fitted to the time series within each window, using the Viterbi algorithm⁵⁴. Under this model, a change from the baseline state to an active state represented a tap initiation. To reduce the false positive rate of tap detection, the time series was segmented into time windows at three scales – two times the specific metronome period of that trial, five times that same metronome period, and the entire trial length. Active-state inferences from each overlapping time window were required to be in agreement between these three runs. To account for variations in force amplitude within the span of a single trial, the signal within each window was scaled to be within a range of 0 to 100. Inferred tap initiations that occurred in the middle of a 5ms span of monotonically decreasing force samples were discarded, as they were likely artifacts. Additionally, tap initiations that occurred less than 33% of the trial’s metronome period after a previously detected tap initiation were ignored.

To improve the precision of tap identification, a second latent state inference procedure using the GHMM was applied. Specifically, the digital signal was segmented into non-overlapping windows, split at the midpoints of consecutive taps that were inferred during the first pass. The signal within each window was again scaled to be within a range of 0 to 100; then a two-state HMM was used to infer the baseline and active states again. Similar to the first pass, tap initiations that occurred in the middle of a 5ms span of monotonically decreasing force samples were discarded, as were any tap initiations that occurred less than 33% of the trial period after a previously detected tap initiation. This algorithm extracted reliable tap initiations and inter-tap intervals, as illustrated in Figure 9B.

Preferred period or Spontaneous Motor Tempo (SMT)

Preferred period or Spontaneous Motor Tempo (SMT) was computed from one trial of 20s duration in the first stage of the experiment. The SMT was defined as the average duration (in ms) between tap intervals during the middle 10s of the 20s trial.

Performance measures

Five metrics were calculated to evaluate the rhythmic timing ability of subjects: SMT, ITI error, ITI variability, asynchrony, and drift. For the continuation or unpaced segment, the first 3s were excluded to eliminate disruptions or transients. Since the minimum number of taps within each segment was 8 in all participants, all measures were computed over the last 8 inter-tap intervals to ensure the same number of cycles across the 3 conditions and across all subjects.