Agreement in Musical Experts on the Identification of Beat Levels and their Salience

Agreement in Musical Experts on the Identification of Beat Levels and their Salience

Schroeder, J., Simmons, G.

Yale University, Cognition of Musical Rhythm, Virtual Lab


1.1  Introduction

This experiment aimed to look at the salience of beat (or pulse) levels, or subdivisions, in certain songs. Salience is a measure of how perceivable each beat level is, and is made of up a number of different variables, including volume, timbre, and pitch. The purpose of studying the number of salient pulse levels, or subdivisions, was to explore whether their variance might affect the perception of a song’s groove. A pulse level is a steady beat in the music, a stream of musical events that happen in equal and predictable intervals, and has also been defined more anecdotally as a beat that you might feel compelled to tap along or move to. However, in many pieces of music, there are several possible pulse levels that one could focus on. We theorized, after reading Janata, Tomic, and Haberman (2011), that having more pulse levels accessible in the music might be connected with a higher groove rating. There are, of course, many different factors that make up the perception of groove; in this study we wanted to isolate this one factor as best as possible to see what the relationship is. [Genevieve]

1.2  Previous Research

Our initial inspiration was drawn from the “Sensorimotor Coupling in Music and the Psychology of the Groove” study by Janata, Tomic, and Haberman (2011). Many other studies have investigated the meaning of ‘groove’ and the rhythmic properties related to it, by comparing microtiming deviations (Gouyon, Hornstrom, Madison, Ullen, 2011) or just categorizing the prominent factors “regular-irregular, groove, having swing, and flowing” (Madison, 2006).

Methods included assessing correlations between listeners’ ratings and a number of quantitative descriptors of rhythmic properties for one hundred music examples from five distinct traditional music genres (Gouyon, Hornstrom, Madison, Ullen, 2011) and in terms of differences in ratings across sixty-four music examples taken from commercially available recordings (Madison, 2006).

Janata et al. explored the urge to move in response to music using phenomenological, behavioral, and computation techniques. Assuming that groove is a psychological construct, they posited that the “degree of experienced groove is inversely related to experienced difficulty of bimanual sensorimotor coupling under tapping regimes with varying levels of expressive constraint and that high-groove stimuli elicit spontaneous rhythmic movements” (Haberman, Janata, Tomic, 2011). [Genevieve]

1.3  Present Research

The questions we set out to attempt to answer was if the saliency of beat level pulses affected perceived groove rating of a set of songs with already-established groove ratings from the 2011 study of Haberman, Janata, and Tomic.

Our initial proposal was to have a panel of musical experts rate levels of beat levels in songs for confirmation and then have the ability to choose songs with a variety of beat levels before giving subsequent subjects songs to rate grooviness of, but because of difficulties in collecting data and inconsistencies between experts’ opinions, we have decided to use only the first part of our initially proposed project. [Genevieve]


1.1  Participants

There were 5 participants, all students from the Yale School of Music, as well as one professor. The participants were contacted by email, and were not offered any sort of compensation. [Jordan]

1.2  Stimuli

The experiment consisted of a Qualtrics survey, built with the Qualtrics website, and contained fourteen 30 second excerpts of songs of various style and genre, which were supplied by Petr Janata, and had been used in Janata et al. (2011). Each of the fourteen excerpts constituted a trial, and the number of beat levels present in the song, the salience of each of those beat levels, and the primary instrument that contributed to the creation of each beat level were used as variables. Salience was rated on a scale from 0 – 10, and the labelling of instrumentation was left up to the subjects. The tempos shown below were found by Stefan Tomic and Petr Janata using the method described in Tomic & Janata (2008), and a few were halved, due to the fact that they were obviously associated with a faster metric levels. One song (Step it Up Joe) was excluded in later analyses due to a lack of information provided for us by Janata et al.

Song Artist Genre Tempo (bpm) Groove Rating
Superstition Stevie Wonder Soul 99 108.7
Yeah! Usher feat. Lil’ John & Ludacris Soul 211 89.7
Freedom of the Road Martin Sexton Folk 25 59.7
What a Wonderful World Louis Armstrong Jazz 36 66.4
Beauty of the Sea The Gabe Dixon Band Rock 63 32.1
Thugamar Fein an Samhradh Linn Barry Phillips Folk 33 29.3
The Child is Gone Fiona Apple Rock 195 62.3
Mama Cita (Instrumental) Funk Squad Soul 95 101.6
Citi Na GCumman William Coulter & Friends Folk 20 35.2
Summertime Ella Fitzgerald & Louis Armstrong Jazz 99 67.9
Goodies Ciara feat. Petey Pablo Soul 50 92.3
Step it Up Joe Mustard’s Retreat Folk X X
In the Mood Glenn Miller & His Orchestra Jazz 162 96.9
Squeeze Robert Randolph & The Family Band Rock 58 63.4

Figure 1: This figure details the information about each song as collected and found by Janata et al. (2012). [Jordan]

1.3  Task & Procedure

Participants were asked to complete a survey which presented 14 excerpts of songs each 30 seconds long in a random order. Participants were then asked to identify the salience of each beat level, on a scale of one to five with the first being the slowest, and the last being the fastest. They were instructed to only put record beat levels that they believe existed clearly in the music, not subdivided and less-natural beat levels that they were able to find due to musical training. They were also asked to provide the instrument that contributed the most to the creation of each beat level.

Figure 2: This figure shows the basic setup of each trial. An additional space was provided below in each trial for miscellaneous or explanatory comments. [Jordan]

1.4  Data Collection & Analysis

The data was collected through the Qualtrics website, and then exported into an excel sheet. An analysis was conducted by looking at the descriptive statistics on the experts’ agreement on the number of beat levels in each song, as well as the most salient beat level of those. Tempos of the most salient beats were found using an online tap metronome. This was conducted by the authors, and though there was a certain amount of subjectivity involved, the experts’ provided information concerning the instrumentation of each level, as well as its placement on the scale of Slowest Beat Level to Fastest Beat Level was carefully consulted to make decisions regarding the tempo.  These measures were then used to compare to the groove ratings and tempos found in Janata et al. (2011). [Jordan]


1.1 Population Sample

Our population sample consisted of four (4) Yale School of Music students, and one (1) professor and researcher of music. By these criteria, we judged them to be musical “experts”, defined as having many years of experience and a solid basis of theoretical and practical application of music. We believe our results allow us to generalize to the population of people who have this foundation of musical knowledge, but also to comment on the general population as a whole. No further descriptive statistics such as age, gender, etc. were obtained as their study was only intended to be exploratory at the outset of this experiment. [Jordan]


The # of Beat Levels

Expert # 1 Expert #2 Expert #3 Expert #4 Expert #5
Superstition 3 2 4 3 3
Yeah! 2 4 3 3 5
Freedom of the Road 2 4 4 3 4
What a Wonderful World 2 3 5 3 4
Beauty of the Sea 3 4 2 2 4
Thugamar Fein an Samhradh Linn 2 3 2 2 4
The Child is Gone 2 3 4 3 5
Mama Cita (Instrumental) 2 3 2 3 4
Citi Na GCumman 2 2 3 2 5
Summertime 2 2 4 2 5
Goodies 2 3 3 3 2
Step it Up Joe 2 3 3 2 4
In the Mood 3 3 3 3 3
Squeeze 3 3 4 3 4

Figure 3: Table of the numbers of beat levels assigned to each song by each expert.

As the table shows, there were a wide variety of beat levels identified by each individual expert. Different experts had different trends of what number of beat that they consistently identified, such as Expert #1 only alternating between identifying 2-3 beat levels per song. Expert #4 only identified 2-3 beat levels per song as well, and Expert #2 only had three songs with 4 beat levels identified. Experts #3 and #5 both had more variety between beat levels identified. Only Expert #3 and #5 identified any song as having 5 beat levels, although this was not consistent across songs and Expert #3 only identified “What a Wonderful World” as having 5 beat levels. The only song consistent in number of beat levels between all five experts was “In The Mood”, but the experts still disagreed on the particular order of instrumentation as organized by tempo in distinguishing each beat level.

Descriptive Statistics of the # of Beat Levels by Song

Range Mean Standard Deviation
Superstition 2 to 4 3 0.707106781
Yeah! 2 to 5 3.6 1.140175425
Freedom of the Road 2 to 4 3.4 0.894427191
What a Wonderful World 2 to 5 3.4 1.140175425
Beauty of the Sea 2 to 4 3 1
Thugamar Fein an Samhradh Linn 2 to 4 2.6 0.894427191
The Child is Gone 2 to 5 3.4 1.140175425
Mama Cita (Instrumental) 2 to 4 2.8 0.836660027
Citi Na GCumman 2 to 5 2.8 1.303840481
Summertime 2 to 5 3 1.414213562
Goodies 2 to 3 2.6 0.547722558
Step it Up Joe 2 to 4 2.8 0.836660027
In the Mood 3 3 0
Squeeze 3 to 4 3.4 0.547722558

Figure 4: Descriptive statistics based on the number of beat levels the experts assigned each song, organized by song.

The majority of the ranges seem to be centered around 3, with the mean of all the songs falling between 2.6 – 3.6.  The mean for all the beat levels perceived by the experts, across all songs, was M = 3.04285714, SD = .32749465. No song was identified as having less than 2 beat levels.

Figure 5: A histogram of the frequencies of each mean # of beat levels of the songs

Figure 6: This bar graph displays the means of the number of beat levels given by the experts for each song.

The histogram in Figure 5 appears to be relatively normal, except for gap in the center for a mean number of beats of 3.2, which tells us that statistical analyses are valid, but also that our results may be more random than we expected, as a normal distribution is the distribution of a random sampling. A mean number of beat levels of 3 and 3.4 beat levels are the most frequent, occurring four times each – 3.4 falls just outside one Standard Deviation from the overall mean, so it’s odd that there are 4 songs with that mean, but we had only 5 participants, which may explain some of the randomness of the data. “Yeah!” falls furthest from the overall mean, but is still less than 2 Standard Deviations away.

Descriptive Statistics of the Number of Beat Levels by Expert

Range Mean Number of Beat Levels Perceived Standard Deviation
Expert #1 2 to 3 2.28571429 0.468807231
Expert #2 2 to 4 3 0.67936622
Expert #3 2 to 5 3.28571429 0.913873533
Expert #4 2 to 3 2.64285714 0.497245158
Expert #5 2 to 5 4 0.877058019

Figure 7: This table shows descriptive statistics of the number of perceived beat levels, but this time is organized by expert, rather than song.

Figure 8: This figure shows how many beat levels each expert perceived in each song.

As we can see from these two figures, it is clear that different experts had different concepts of beat levels and used different strategies to find the beat levels in a given song. After all, for the same 14 clips of songs, one expert (Expert #5) perceived mostly 4 and 5 beat levels in the songs, while two others (Expert #1 & #4) perceived no more than 3 beat levels in all of the clips. We can also see that there was wide disagreement between the experts on almost every song – the only song that the experts unanimously agreed on was “In the Mood”, which they perceived as having 3 beat levels. It must also be taken into account that the mean and standard deviation across all the reported numbers of beat levels is M = 3.04285714, SD = .32749465. We can see that Expert #2 is closest to this mean. For the all the results from each expert in table form, please check the appendix.

1.3  Expert’s Perceived Instrumentation and Salience Ratings

Tempos of Each Expert’s Most Salient Beat Levels

Person 1 Person 2 Person 3 Person 4 Person 5
Superstition Bass Drum + Snare: 50
Clavinett + High Hat: 99
Bass: 100 Vocals: 100
Guitars: 200
Voice: 100 Bass, Kick, + Snare: 100
Yeah! High Hat  + Synth: 210 Percussive Click: 52 Vocals: 210 Bass Drum + Clap: 105 Kick: 52
Cymbals, Voice + Synth: 210
Freedom of the Road Bass Drum + Snare: 49 Bass: 25
Drums: 146
Harmony Piano: 25
Drums: 146
Vocals: 146
Bass Drum + Snare Drum: 49 Kick + Snare: 49
What a Wonderful World Bass Drum + Snare: 72 Drums: 72 Vocals: 72 Voice: 72 Kick, Snare, Horns, + Strings: 72
Beauty of the Sea Keyboard: 60
Keyboard: 120
Synth: 120 Saxophones: 60 Strings (Synthesizer): 60 Organ: 60
Thugamar Fein an Samhradh Linn Downbeats Every 6: 17 Cello: 34
Bagpipes: 34
Wind Cello: 34 Rolled Chords: 17 Cello: 34
The Child is Gone Piano: 65
Drums: 195
Piano: 65 Vocals: 65
Drums + Keyboard: 195
Piano Chords: 65 Drums, Bass, + Piano: 195
Mama Cita (Instrumental) Bass Drum + Bass:  95
High Hat + Percussion: 190
Drum: 95 Keyboard: 95
Percussion: 190
Bass Drums:  95 Kick + Bass: 95
Cabasa: 190
Citi Na GCumman Guitar’s Bass Notes: 38 Guitar Whole Note: 38
Guitar Melody: 114
Guitar Arpeggios: 114 Strum: 38
Summertime Strings: 34
Trumpet: 70
Drums: 70 Trumpet: 70 Drum Pattern: 70 Strings: 34
Bass: 70
Goodies Claps: 51 Drums: 101 Vocals: 101 Snare Drum:  101 Vocals: 204
In the Mood Bass: 162 High Hat: 162 Saxes + Trumpets: 162 Double Bass: 162 Drum + Bass: 162
Squeeze Bass Guitar + Bass Drum + Snare: 120
High Hat: 244
Drum Kit: 244 Bass: 120
Drums: 244
Guitar: 475
Bass Guitar: 120 Bass: 120
Rhythm + Solo: 240

Figure 9: A table showing the instrumentation and tempo of the beat level (or beat levels, as is the case with a few songs and experts) experts perceived as the most salient. Tempos are in BPM, and when an expert rated a song as having two or more equally salient “most salient” beat levels, all have been included.

Figure 10: This graph depicts the tempo found by the authors of the “Most Salient Beat Level” identified by each expert, compared to the tempos found by Janata et al. (2011). In order to graph the information, only one “Most Salient Beat Level” could be shown, though as the table above shows, some experts identified more than one “Most Salient Beat Level” – in these cases we have arbitrarily chosen to graph the slowest one.

As we can see from Figure 9 & 10, many of the experts were able to agree on a most salient beat level – all of the experts and Janata agreed unanimously on the tempo of the most salient beat level of “In the Mood”, placing it at around 162 BPM, as well as on the most salient beat level of “Mama Cita (Instrumental)”. However, as is shown in the table, in the case of “Mama Cita (Instrumental)”, three of the experts (Experts #1, #3, and #5) also identified a second most salient beat level at double the first tempo. This is an example of how different beat levels seemed to follow simple ratios such as halves and thirds. The keyboard, drum, and bass in “Mama Cita” (Instrumental) were identified by all five experts as being the most salient beat level at the tempo we further analyzed to be 95 BPM. Additionally, those three experts all identified a second, equally most salient beat level, at 190BPM (double time of the slower, more salient beat level) for high hat, percussion, and cabasa. This can also be seen in “Superstition,” where the most salient beat levels identified were even at the four different tempos of 50, 99, 100, and 200 BPM, all easily subdivided into each other.

There was some disagreement though; all of the experts deemed either percussive sounds or vocals as the most salient beat levels for “Goodies.” However, each expert identified each at a different tempo, based on our interpretations of their rankings of each salient instrument. Expert #1’s ‘claps’ are half the tempo of Experts #2 and #4 percussion sounds, which are actually the same tempo of Expert #3’s vocals because they are listed before percussion in the five levels. Expert #5 has vocals listed after percussion beat levels though, leading us to interpret that theirs was the faster subdivision of vocals at the tempo of 204 BPM.

Experts tended to differ particularly in identifying in the instrumentation of specific beat levels, or assigned different instruments different tempos, but for the most part always had at least one level that could be found in the results for other experts as well. For example, in “What a Wonderful World,” the most salient beat level, although identified as some combination of bass drum, strings, horn, and voice, always was within a consistent tempo of 72 BPM. Many times differences in instrumentation were only due to strategies in identifying specific names of instruments, i.e. labelling a beat level as ‘bass drum and snare’ versus ‘kick and snare.’

In conclusion, we found a wide variance in our data – sometimes the experts agreed, even unanimously in the case of “In the Mood”, and sometimes they disagreed not only on the most salient beat levels, but on the instrumentation and tempos of those beat levels. However, we generally found that the experts identified tempos that fell into simple beat ratios, and if not the same as Janata’s ratings, were at least multiples or factors of them.

1.4  Analysis & Figure 3

Figure 11: This graph compares our mean number of beat levels identified to the groove ratings of the same song clips from Janata et al. (2011)

As we can see from this graph, though there was no statistical analysis possible (due to the lack of a “known value” for the beat levels) we can say that we failed to reject the null hypothesis that the number of beat levels in a song is not correlated with the grooviness of the song. In other words, it is not clear at all whether having more beat levels in a song might contribute to hearing the song as having more groove. In our findings, the song with the highest groove rating ended up having roughly an average number of beat levels, while songs that received a very low groove rating had the same average number or even higher numbers of beat levels. “Citi Na CGumman,” with one of the lowest groove ratings of 35.2, had up to five levels of beat levels identified.


It’s clear from our results that beat levels are a far more subjective and varied measure than we anticipated. Our assumption was that school of music students and professors, with their expertise and experience with music, would be more likely to identify similar numbers of beat levels for each song.  However, as shown by the results, only one song was unanimously agreed upon, “In the Mood” – the rest showed a wide range of variety, with some experts identifying only 2 beat levels for some songs and others identifying up to 5 beat levels for the same songs. As noted in the Present Research section of this structured abstract, this unexpected variability made it difficult and of questionable worth to continue on with our original experiment.  However, with the data we gathered from the experts, we were still able to conduct several analyses exploring the variability between experts we found more thoroughly, and comparing our results with those of Janata et. al (2011).

Through these analyses we discovered that each expert often had their own quirks and trends – this makes sense when you consider that each expert likely had their own specific strategy for discerning the number of beat levels in each clip. Across the experts, we also noticed a habit of recognizing a combination of instruments as creating beat levels not articulated by one instrument alone.  For example, in “Yeah!”, Expert #4 identified the instrumentation of the most salient beat level as “Bass Drum + Clap”.  What we can hear when listening to the clip is that the Bass Drum and the Clap both move at slower tempos individually [which Expert #2 separated, as is shown in Figure 7, (we attributed Expert #2’s instrumentation of a “percussive click” as the Clap heard by Expert #4) and which we found to have tempo roughly half that of the two instruments combined], but when heard together, and perceived as one beat level, they combine to form a faster metric structure, the downbeats of each instrument falling on beats of a faster tempo.

Experts, in addition to disagreeing about the number of beat levels present in the songs, also disagreed about the most salient beat, and the instrumentation of it. As noted above, this was sometimes caused by some experts’ combinations of instruments, but sometimes different experts simply seemed to be listening for different auditory stimuli in order to obtain the number of beat levels.  When comparing Figure 6 with Figure 7, we can see that Expert #1 seemed to focus more on the different percussion present in a song to find salient beat levels, while Expert #2 seemed to look more consistently to the strings in order to find the salient beat levels.  From these results we can infer that these two experts were likely using different strategies, but also may have different concentrations or focuses within music, one being more attuned to percussion and the other to string instruments.

In addition to the variability found between experts, we ran into the problem that all of the song’s averaged beat levels fell between 2.6 and 3.6.  Despite the variability of reported beat levels, sometimes ranging from 2 beat levels all the way to 5 beat levels, the averages showed little variation. It may be that this difference is still significant, but it debunks an implicit assumption we possessed that there would be a wider variety of averages, and that it would be more clearly in different categories. This also seems to imply that most songs would fall within this range, as we sampled a variety of genres and tempos of music.

Though were not able to correlate the number of salient beat levels with the perceived groove of a song, we do not believe that the two are necessarily unrelated. There were several implicit assumptions that became clear only after analyzing the results, but that might have accounted for the variability as well as the findings we drew from those results.  For example, a better method, and one that might have produced more consistent results, would have been for the authors to analyze the songs beforehand and produce a list of instruments contained in each song, which the participants could have then put into order from slowest beat level to fastest.  This practice would have at least eliminated some of the variability within the instrumentation, as many experts put down varying names for the same beat level.  The way we chose to measure salience also had its flaws; 0 – 10, while seemingly a simplistic subjective measure, was found difficult to attach to a measure of salience, as there is no unit for salience, and no value that immediately corresponds across the two measures. (For example, from 0 – 10, what does a salience of 5 mean?  The beat level is halfway salient?). As stated above, a question asking participants to order the salient beat levels, and using that as a comparative measure may have been a more effective method. Lastly, our method of finding the tempo of each beat level was flawed.  Though the authors were as accurate as possible, following the results given by the experts as closely as possible, it would have been more accurate to have the participants give the tempo for each beat level they were identifying themselves, by use of a within survey tap metronome.  Using this method would have eliminated any possible bias (no matter how much the authors tried to uphold the integrity of the results, it is possible, though unlikely, that we misinterpreted some of the information when finding the tempos), and made the results much clearer.  In another direction, a future study may have each expert rate the song’s grooviness in addition to finding the beat levels – perhaps the connection is not between the structural beat levels within the song and the song’s groove, but between a person’s personal perceived number of beat levels (however many beat levels that person is hearing/believes to exist in the song) and that person’s personal assessment of the song’s groove.  Future studies may also want to ask the experts specific questions about their training, in order to better understand their backgrounds and the strategies they may be employing in the task.  By implementing these measures, we believe that a more successful experiment, more sound and distributed to a larger number and wider variety of people may be conducted in the future to better explore the relationship between the number of beat levels and groove.



The authors would like to thank our musical experts for taking the time to complete our survey, as well as Professor Eve Poudrier for advising us throughout the process. We would also like to extend our gratitude to Petr Janata, who was the leading researcher for the original study our experiment was based off of, and who very generously provided us with the stimuli he and his team used and the data they collected.


Janata, P., Tomic, S. T., & Haberman, J. M. (2011). Sensorimotor Coupling in Music and the Psychology of the Groove. Journal of Experimental Psychology, Vol. 141, No. 1, 54–75


Madison, G. (2006). Experiencing Groove Induced by Music: Consistency and Phenomenology. Music Perception: An Interdisciplinary Journal Vol. 24, No. 2 (pp. 201-208)

Madison, G., Gouyon, F., Ullen, F., Hornstrom, K. (2011). Modeling the tendency for music to induce movement in humans: First correlations with low-level audio descriptors across music genres. Journal of Experimental Psychology: Human Perception and Performance, Vol 37(5), 1578-1594.

Tomic, S. T., & Janata, P. (2008). Beyond the Beat: Modeling Metric Structure in Music and Performance. The Journal of the Acoustical Society of America 124.6: 4024. Web.


Expert #1

1st Beat Level and Salience 2nd Beat Level and Salience 3rd Beat Level and Salience 4th Beat Level and Salience 5th Beat Level and Salience
Superstition Bass Drum + Snare: 10 Clavinet + High Hat: 10 Horns + Clavinet + High Hat: 9
Yeah! Bass Drops: 8 High Hat + Synth: 10
Freedom of the Road Bass Drum + Snare: 10 High Hat + Guitar: 8
What a Wonderful World Bass Drum + Snare: 10 Guitar Arpeggios: 9
Beauty of the Sea Everything else: 8 Keyboard: 10 Keyboard: 10
Thugamar Fein an Samhradh Linn Stuff On Downbeats Every 6: 10 Cello: 3 Cello: 6
The Child is Gone Piano: 10 Drums: 10
Mama Cita (Instrumental) Bass Drum + Bass: 10 High Hat + Percussion: 10
Citi Na GCumman 8 10
Summertime Strings: 10 Trumpet: 10
Goodies Claps: 10 Vocals + Synth: 10
In the Mood Open High Hat: 7 Bass: 10 High Hat + Horns: 9
Squeeze Bass Guitar + Bass Drum + Snare: 10 Bass Guitar: 10 High Hat: 10

Figure 12

As stated above, Expert #1 never perceived more than 3 beat levels in a given song, and tended to rate the salience of those beat levels relatively highly, even at times rating two or more beat levels as “The Most Salient”. Expert #1 also tended to focus on the percussion parts in the piece in order to extract the salient beat levels although they tended to deconstruct the drum kit, for example differentiating between high hat, and bass drum and snare.

Expert #2

1st Beat Level and Salience 2nd Beat Level and Salience 3rd Beat Level and Salience 4th Beat Level and Salience 5th Beat Level and Salience
Superstition Drums: 6 Bass: 8
Yeah! Percussive Click: 9 Low Drum: 7 Synth: 8 Strings: 8
Freedom of the Road Bass: 9 Guitar: 6 Piano: 8 Drums: 9
What a Wonderful World Strings: 7 Guitar: 8 Drums: 6
Beauty of the Sea Saxes: 5 Synth: 8 Synth: 4 Synth: 3
Thugamar Fein an Samhradh Linn Guitar: 5 Cello: 8 Bagpipes: 8
The Child is Gone Strings: 6 Piano: 9 High Hat: 8
Mama Cita (Instrumental) Piano: 5 Drums: 9 Percussion: 6
Citi Na GCumman Guitar’s Bass Notes: 8 Guitar’s Midrange Notes: 6
Summertime Strings: 6 Drums: 9
Goodies Sign Wave Noise: 7 Drums: 8 Vocals: 5
In the Mood High Hat: 8 Bass: 7 Saxes From 10-12 sec: 8
Squeeze Rhythm Guitar: 6 Drum Kit: 8 Electric Guitar: 7

Figure 13

Expert #2, in comparison to Expert #1, seemed to tune in more to string instruments when searching out the salient beat levels within a piece. Unlike the other Experts, Expert #2 refrained from combining several instruments into the creation of beat levels, citing only one instrument for each beat level.

Expert #3

1st Beat Level and Salience 2nd Beat Level and Salience 3rd Beat Level and Salience 4th Beat Level and Salience 5th Beat Level and Salience
Superstition Drums: 8 Trumpet: 7 Vocals: 10 Guitars: 10
Yeah! Synth: 9 Vocals: 9 Percussion: 10
Freedom of the Road Harmony Piano: 10 Guitar: 9 Drums: 10 Vocals: 10
What a Wonderful World Bass: 6 Trombone Fill: 7 Strings: 9 Vocals: 10 Guitar: 8
Beauty of the Sea Saxophones: 10 Synth: 9
Thugamar Fein an Samhradh Linn Strummed: 9 Wind Cello: 10
The Child is Gone Bass: 5 Violin: 8 Vocals: 10 Drums + Keyboard: 10
Mama Cita (Instrumental) Keyboard: 10 Percussion: 10
Citi Na GCumman Guitar Whole Note: 10 Guitar Arpeggiation: 9 Guitar Melody: 10
Summertime Strings Vib: 9 Bass: 6 Bass Answer: 8 Trumpet: 10
Goodies Bass: 8 Vocals: 10 Percussion: 9
In the Mood Bass: 5 Drums: 7 Saxes + Trumpets: 10
Squeeze Keyboards: 6 Bass: 10 Drums: 10 Guitar: 10

Figure 14

Expert #3 often neglected to identify an instrument for a first and even second beat level, focusing on faster and more salient instruments through which to identify each level. The Bass was most frequently associated with the slowest identified beat level. Like the rest of the experts, the instrumentation for “In The Mood” consists of some combination of percussion, horns, and bass.

Expert #4

1st Beat Level and Salience 2nd Beat Level and Salience 3rd Beat Level and Salience 4th Beat Level and Salience 5th Beat Level and Salience
Superstition Bass Drum: 3 Drum Set + Voice: 10 Guitar Combo: 8
Yeah! Voice (Accented Syllables): 3 Bass Drum + Clap: 9 High Hat: 6
Freedom of the Road Bass Guitar + Bass Drum: 7 Bass Drum + Snare Drum: 10 High Hat + Guitar: 5
What a Wonderful World Bass Drum: 3 Drum Set + Voice: 10 Guitar Arpeggio: 7
Beauty of the Sea Strings (Synthesizer): 6 6
Thugamar Fein an Samhradh Linn Rolled Chords: 8 Bagpipes: 6
The Child is Gone Drum: 2 Piano Chords: 10 High Hat: 6
Mama Cita (Instrumental) Chord Change: 1 Bass Drums: 10 High Pitched Instrument: 3
Citi Na GCumman Chord Changes: 4 Guitar Arpeggios: 8
Summertime Strings + Bass Chords: 5 Drum Pattern: 8
Goodies Bass Drum (Synthesizer): 2 Snare Drum: 8 Vocals: 6
In the Mood Brass Pattern: 4 High Hat: 8 Double Bass: 10
Squeeze Drums: 4 Bass Guitar: 10 Solo Guitar: 10

Figure 15

Expert #4, like Expert #1, never perceived more than 3 beat levels within a given piece, though s/he had a tendency to rate the salience of some beat levels as much lower, perceiving some beat levels only as just above having no salience whatsoever. Expert #4 noted in the comments section of the survey (the only expert that did so) that some instruments that make up a salient beat level do not come in immediately with the beginning of each song clip, and that “the fastest level seems to be a combination of faster attacks rather than being defined by a main instrument,” which is an observation that analyzing all of the experts together can show as a difficult part of the survey to be clear upon.

Expert #5

1st Beat Level and Salience 2nd Beat Level and Salience 3rd Beat Level and Salience 4th Beat Level and Salience 5th Beat Level and Salience
Superstition Bass, Kick, + Snare: 9 Clav, Brass Line, Voice: 8 Clav, Brass Line, Voice: 7
Yeah! Kick: 7 Whistle: 8 Kick: 9 Cymbals, Voice, + Synth: 9 Ring + Voice: 5
Freedom of the Road Bass + Slide Guitar: 9 Kick + Snare: 10 Guitar, High-Hat, + Voice: 9 Hi-Hat, Voice, + Bass: 7
What a Wonderful World Harmony: 7 Kick, Snare, Horns, + Strings: 10 Guitar + Hi-Hat: 9 Voice: 5
Beauty of the Sea Phrasing: 2 Implication of Organ: 9 and 10 Low Hum: 9 Organ: 9
Thugamar Fein an Samhradh Linn Plucked String: 5 Cello: 8 Pipes: 4 Drone: 2 Drone: 0
The Child is Gone Phrasing: 2 Electric Guitar: 7 Drums, Bass, + Piano: 10 Strings: 2 Cymbal + Voice: 4
Mama Cita (Instrumental) Keyboard: 7 Kick + Bass: 9 Melody + Bass: 7 Cabasa: 9
Citi Na GCumman Phrasing Emphasis: 2 Bass Register: 7 Strum: 10 Chordal Rhythm: 1 Melody: 9
Summertime Harmony: 4 Strings: 9 Bass: 9 Trombone: 7 Brushes: 6
Goodies Synth Whistle: 0 Synth Whistle: 0 Kick + Snare: 8 Vocals: 9
In the Mood Sax Line: 8 Drums + Bass: 9 Sax Solo: 8
Squeeze Phrasing: 6 Bass: 8 Drums: 9 Rhythm + Solo: 8

Figure 16: Figures 12 – 16 show each experts’ perceived instrumentation and their salience within each song, in order of tempo slowest to fastest.

Expert #5 perceived the widest range of beat levels, and often perceived significantly more beat levels than the other experts, hearing 4 or 5 beat levels much more frequently. However, they also had much lower levels that other experts present, even listing in “Goodies” that the first two beat levels were composed of a synth whistle with a salience of “0”, and can be assumed to be stating the presence of the synth as auditory stimulation without contributing to the tempo.


Sample Structured Abstract Reports

Here are a few sample structured abstracts from students group projects in previous semesters:

– Acevedo, Lettie, Parnes, & Schartmann (2013), Effect of Tempo on Perceived Emotion of Musical Excerpts

– Broshy, Latterner, & Sherwin (2013), Interaction Between Melodic Pitch Content and Rhythmic Perception

– Davis, Fox, & Roth (2013), Effects of Rhythmic Consistency on Perceived Speech Effectiveness

– De Freitas, Jameson, & Strebendt (2013), Influence of Rhythmic Tempo on Sustained Entrainment to the Beat

– Guerra, Hosch, & Selinsky (2013), Tapping to Uneven Beats






Authors’ names here (alphabetical order)

Yale University, Cognition of Musical Rhythm, Virtual Lab (this is your affiliation)



This section introduces your topic and research question and situates it in the larger context of musical experience and music research.

1.1  Introduction

Start with one or two sentences that situate your topic in the larger context of musical experience. Follow with highlighting the particular aspect of musical experience your study aims to investigate, providing all necessary definitions. You can also state your research question (or a series of related questions) in a preliminary form here.

1.2  Previous Research

This section provides a brief summary of previous research that has addressed some aspects of your research question. The summary can be organized topically (grouping sources that have investigated related aspects together) or methodologically (grouping sources that have employed similar methods together). References to specific articles should be given in parentheses, including authors’ last name and publication date (in alphabetical order within the parentheses, if there is more than one).

1.3  Present Research

State your formal (i.e., operationalized) research question here. You may include one or two additional sentences for clarification or details, including the predicted outcome.



This section provides the details of your experimental design as concisely and clearly as possible so that readers can better assess your results. You may include figures to illustrate some aspects of the method (e.g., diagram of procedure, notation of sample stimuli, etc.).

1.1  Participants

Provide some basic information about your population sample (age, gender, musical training). This is the population you can generalize to.

1.2  Stimuli

Describe your experimental materials and how they were constructed. Be as specific as possible: What are the source materials? What are the variables? How many different conditions and trials?

1.3  Task & Procedure

What is the experimental paradigm, i.e., what is the specific task your participants had to perform, including the specific instructions, and how were the materials presented (apparatus, order, type of response, etc.)?

1.4  Data Collection & Analysis

How was the data collected and in what format? What are your measures? How was the data transformed (if applicable)? What kind of statistical tests did you conduct?



This section presents an analysis of your data and basic findings. Begin with some descriptive statistics of your population sample (e.g., gender, age, musical training, etc.). Follow with descriptive and/or inferential statistics based on performance measures. Each analysis will generate one main observation supported by some form of statistical analysis (e.g., comparison of means and standard variations based on one variable, e.g., % correct scores in slow vs. fast tempo). Each result should be accompanied by a well-formed graph and/or table.

1.1 Population Sample

Report descriptive statistics related to your population sample (the population you will be able to generalize to).

1.2  Analysis & Figure 1

Report your first finding and supporting data.

1.3  Analysis & Figure 2

Report your second finding and supporting data.

1.4  Analysis & Figure 3

Report your third finding and supporting data.



This section aims to provide a brief summary of your study and contextualize its results. How do your results inform your original question? What are some of limitations of your method? What is the next step?



References should use current APA style and be well demarcated by indenting all the lines after the first (as exemplified below).

Fidali, B. C., Poudrier, È., & Repp, B. H. (2013). Detecting perturbations in polyrhythms: effects of complexity and attentional strategies. Psychological Research, 77, 183-195.

Poudrier, È. (2009). Local polymetric structures in Elliott Carter’s 90+ for piano (1994). In B. Heile (Ed.), Modernist legacy: essays on new music (pp. 205-233). Farnham, England: Ashgate.

Poudrier, È., & Repp, B. H. (2013). Can musicians track two different beats simultaneously? Music Perception, 30, 361-390.

FINAL STEP: Structured Abstract

A structured abstract is one of a few standard formats used in the publication of conference proceedings. It is a great way to share findings with the community. To facilitate your writing of this abstract, I have prepared a template (posted on this page) that contains the basic instructions on how to do it. You can work directly on the template by posting it on your group project page; a document copy of the abstract is available here.

As each student will be graded individually, it is very important that you subdivide the work evenly between group members and that authors write their names in square brackets at the end of each section they contribute.

Structured abstract must be completed at the latest by Thursday, December 11 (end of Reading period), but as much of the work can be completed in advance of doing the data analysis, you are required to complete the writing of sections 1 and 2 by Thursday, December 4, 11:30 AM (class time).


STEP 5: Implementation (Drafts Due Tuesday, November 11)

The draft of your experiment is due next week, Tuesday, November 11. Try to make this draft as complete as possible, including embedding all testing instructions and questions on Qualtrics, as well as participants’ questionnaire and debriefing. If some of your stimuli are not ready, put place-holder questions. We will review your work in class on Tuesday and do some troubleshooting. The goal is to give your experiment a test-run by the end of the week, so that data collection can start at the latest on Monday, November 17.

Both experiments should include the following components:

– Statement about the pedagogical nature of the project and the anonymity of the data collection process.

– Statement that participants should free free to stop at any time, without adverse effect to them. You may also share that incomplete data sets will not be used in the analysis.

– Participants’ questionnaire (at any point you feel is most appropriate; an option we did not discuss is to have it in the middle).

– Debriefing & free response option: Explain what the experiment was seeking to explore more specifically at the end and provide participants with the option to send comments.

– Contact information in case participants want to receive a copy of the report (with a time period when it will be available).

As you near the end of the experimental design phase, you might want to take some time to review the methodology handouts that were distributed in class. Are there any concepts that relate to your study that is not clear enough to you? Is there any methodological issue arising from your planned procedure that resembles an issue described in the handouts?

Here are the instructions prepared by Pam Patterson on how to post your stimuli on classes*v2 and integrate them in your survey:

Instructions On Uploading Audio Files and Capturing the URL

Don’t hesitate to seek expert help from the support staff at Yale! Here are the contacts of people we have consulted with:

Mike Laurello, School of Music ( OR Will help you with most aspects of stimuli preparation and music technology for pre- and post-processing of data.

Scott Petersen, MusTLab ( Scott is the supervisor of the Department of Music’s technology lab (4th floor). You can contact him for any issue related to using the lab.

Sherlock Campbell, CSSSI ( Can help you with everything statistics as well as Qualtrics. Don’t forget that there are also consultants in the center who will be able to answer your questions and guide you through the steps of data analysis, if needed.

Pam Patterson, ITS ( Pam is an administrator for the course blog and can help you with any issues related to the course blog as well as using classes*v2 for your study (see above).

Rémi Castonguay, Gilmore Library ( Can help you with database research if you need additional background sources and when it is time to relate you findings back to previous findings.

Here are a few additional instructional resources:

David Huron on Types of Behaviors

NOTE that part 5, on “self-report”, is especially relevant for our work. You can find many more useful videos on empirical music research with David Huron here.

Yale’s link to

Also, two sources on doing web-based research; the first one is especially helpful as it provides some strategies to improve success and reliability of data collection:

Honing & Ladinig (2008), “The potential of the internet for music perception research: A comment on lab-based versus web-based studies

Germine, Nakayama, Duchaine, Chabris, Chatterjee, & Wilmer (2012), “Is the Web as good as the lab? Comparable performance from Web and lab in cognitive-perceptual experiments”


GROUP PROJECT – STEP 4: Experimental Design (Due Thursday, October 30)

Goal:  To operationalize your research question and build all the materials necessary for its implementation on the online platform. An as-complete-as-possible draft of your protocol is due on Thursday, October 30. The finalized plan and all materials will be due the following week.

Remember that In order to operationalize a research question, you need to come up with a theory from which you derive a set conjectures. (The theory and related conjectures should be informed by your background research.) You hypothesis is a formalize statement of one or two of these conjectures. The hypothesis is what you use as a guide to plan the experimental methods, including materials, task, and procedure. It must enable you to make a testable prediction, and thus must specify what will be measured. Ideally, it should also be followed by a scenario of the possible outcomes, and how each of these will be interpreted in relation to your question. The best experiments are often those whose findings are interesting no matter what the specific outcome is (e.g., see Levitin, 1999, “Experimental Design in Psychoacoustic Research”).

There are a few things to keep in mind that pertain to practical limitations:

1. Participants: Your population sample is likely to be small and not representative of the general population (a “convenient” rather than “random” sample). You should limit the number of variables and try to maximize the number of data points you will be collecting (e.g., using a repeated measures design, that is, a design where each condition is presented to all the participants). A convenient sample also means that you will not be able to generalize your findings to the entire population, but only to the population represented by your sample (thus, the importance of providing descriptive statistics of your sample along with the results).

2. Apparatus & Data Collection: This pilot study will be conducted online using Qualtrics. This software allows for several different formats for data collection: (1) multiple choice (single-forced or multiple responses); (2) rating scale (various formats are possible); and (3) free responses. It is possible to have more than one format, but be careful to not gather more data than you can handle. If you have only limited experience with Qualtrics, you should spend some time familiarizing yourself with it. There are various issues to consider when conducting experimental studies online, but an important one to think about from the start is low motivation (failing to complete the experiment). Thus, your experiment should not take too long to complete (25 minutes is probably the maximum time you can expect someone to spend on it) and you should make all efforts to not make it too tedious.

3. Performance Measures & Statistical Analysis: Be realistic and practical in your choice of measures and statistical methods. Most measures will require some form of processing, and unless you have very good programming skills, the road from data to measure could become very arduous. Remember that although inferential statistics have a lot of rhetorical power, descriptive statistics can also be very informative (e.g., see Windsor, 2004, “Data Collection, Experimental Design,” in Empirical Musicology).


1. Make a list of all the components of your experiment (this is a preview of the “methods” section). The final list should be detailed enough to serve as a recipe for building the components. All group members should review and edit the list as necessary by the deadline.

The list will include all necessary information pertaining to:

  • Task & procedure: what you will ask the participants to do and how the stimuli will be presented (ordering)
  • Stimuli: what are the source materials, where they can be found, and how they will be manipulated (i.e., independent variables), including all necessary tools (hardware, software, web tools, etc.); make sure to be in contact with Michael Lorello ( throughout this process!
  • Data collection & analysis: what kind of data will be collected, what will be measured and how, how the data will be analyzed (only descriptive statistics are required, but you may also use correlational and inferential statistics, if you know how. NOTE: We are scheduled for a statistics workshop on Tuesday, October 28 @ CSSSI; for additional reference, see Cozby (2011), “Understanding Research Results”.
  • Predicted outcome: What kind of results can you expect? Formulate how the different possible results would be interpreted in relation to your question.

2. Survey a few online experiments. How well did the design fit the experimental goal (may be presented only after the experiment is completed)? What may be some of the limitations of the methods use?

  • On the “Resources” page of this site, there are a few online experiments, some of which are still active.
  • You should still be able to access last semester’s group experiments on the MARL platform created by Mary Farbhood (NYU).

3. Build the experimental materials: Divide the “building” tasks between your group members. As you work on the materials, you should put samples up on the group site to be reviewed by your colleagues. We will use some of the class time on October 30 to review experimental materials and do troubleshooting; make sure to have some sample materials ready, including a few sample questions on Qualtrics.

Here is a sample list of the information that may be necessary (based on the first experiment on the MARL platform “Spontaneous grouping”):

  • Study title: you will need to take into account how the title might impact your participants responses (i.e., how much do you want your participants to know about the purpose of the experiment before they complete it?).
  • Study description: a concise, one sentence description that says what your study is about.
  • Background: one paragraph that explains the context of your study (you may or may not want to be too specific, again depending on whether you think this might impact you participants responses).
  • General instructions: a concise preview of what your participants will be doing, e.g., “You will be asked to listen to sound clips and indicate how the sounds are grouped (multiple choice response).” You may add any other special instructions about how you want your participants to go through the experiment, e.g., do you want your participants to try to do it without taking a break? If there is more than one block of trials, do you want them to take a break between them? Do you want to ask them to do the experiment in a quiet space? (NOTE: There is a standard piece of technical instructions already prepared that reads “Once you have started the experiment, please do not use the “Back” browser button or refresh the page. By doing so you will interrupt the experiment and it will result in technical problems.”)
  • Part Description (Part 1 of #): This can be the same as the instructions in the previous page or a more/less detailed version (if your study has more than one part, you will need to provide one for each).
  • Trial Instructions (Part #): specific instructions for participants’ task, including how they will respond, e.g., “[Above play button:] Click on the play button to hear the sound. After the sound has finished playing, select a response. Press the continue button to continue to the next trial. [Below play button and above response boxes and continue button:] “How many tones per group did you hear?” You should also provide the response format, e.g., multiple-choice questions: 1, 2, 3, 4; Yes/No; happy/sad/calming/exciting. (NOTE: If your study has more than one part you will need to give similar information for each part.)
  • End (of Part #): You may thank your participants and/or given them some more information about the purpose of your study, how to contact you if they want to see the results when the study is completed, etc. (If your study is in many parts, you may instruct them to take a break at this point or to continue on.)

NOTE: We will work on the background survey and post-experiment questionnaire together. You may want to draft some of the questions you would like to have included.


Group Project STEP 3: Background Research (Multiple Deadlines)

Once a potential research question is identified, it is time to do some background research to see if other researchers have addressed your question before, what methods have been used, and what were the findings. Previous studies will help you to formulate your question in a more practical and specific way, and will even give you some ideas on how to go about designing an experiment. They will also make you aware of some of the issues to consider to avoid pitfalls. It is quite common for an experimental study to be modeled on a previous study, extending it in some modest way. This is a good strategy to make sure your methods are reliable and that your work is immediately relevant to the field.

Posting your research question on your group project homepage:

Before embarking on the background research proper, I would like you to take some time to re-formulate your research question, using the various issues that were raised during the class and the follow-up discussions. This task should be completed as soon as possible, so that the background research work can be divided in some logical way between group members.

Doing background research on your question:

1. Find at least 3-4 articles (1-2 per group member) that are directly relevant to your research question and the experimental study you will conduct to address this question. Some recommended databases to get started are: RILM, PsycArticles, and Google Scholar (J-STOR is a great place to get articles, but not so good for searches). Recommended journals for a more focused search: Music Perception (this is the top journal in this field), Empirical Musicology Review, and Psychology of Music (for a more detailed listing, see

2. Post the bibliographic references and abstracts on your group page as well as a link to the original source (if available electronically), and a comment on how this article will inform your experimental study. You may also include figures or any other materials from the articles that you think will be helpful to your project. (You should also add these two citations and the abstract to the Virtual Lab’s bibliography, for reference.) These references should be posted no later than Thursday, October 16, 11:30 AM.

3. Review the materials posted by your group members, and post comments/questions as needed in preparation for the in-class report and discussion (see below).

4. During class on Tuesday, October 21, each group will do a brief report to the class (about 10 minutes).  Each report will be followed by a brief open discussion, the main goal of which will be to start moving toward operationalization (i.e., defining the rhythmic element to be examined in such a way that it will be measurable, identifying the variables, and formulating a hypothesis). The report should include: (1) a clear statement of the research question, theory, and hypothesis; (2) some background information about state of research and paradigms previously used; and (3) how you intend to pursue your research. Michael Lorello, who will be providing us with technical support for the sound production part of the experimental design, will be present and will meet with each team following the presentations.

Three guiding questions for this task:

  • Based on your literature review, what is the research context in which your question or “problem” is situated?
  • What contribution to knowledge and understanding do you expect to make?
  • And, in preparation for the next step (experimental design), what methods (task, procedure, stimuli, measures) are suggested by this research?

STEP 1: RESEARCH QUESTION (Due 10/8 & 10/9)

The first step of the group experiment project involves writing, evaluating, and selecting a research question that will lead to the design and implementation of an experimental pilot study.

During class on Thursday, October 9, we will review the proposed questions, identify 4-5 viable research questions, and break out into two discussion groups. Each group (or “lab team”) will be tasked with selecting one research question for its experimental project. In preparation for the selection process, you will need to:

1. Identify at least three questions related to the materials we have reviewed in class so far or your individual research question. A good question for this project should have the potential to generate a testable hypothesis that can be experimentally explored within the time and technical constraints of this course. To determine the feasibility of your question, use the Group Tasks #6 & #7 distributed in class last week (from question… to protocol).

2. Post your two best questions on the Group Projects page no later than Wednesday, October 8, 9:00 PM. Each question should be as clear and concise as possible (no more than two sentences).

3. Review your colleagues posted questions no later than Thursday, October 9, 11:30 AM. In preparation for the class discussion and selection process, post questions/comments as needed using the reply function.

Good hunting!

About Group Projects

A significant portion of the coursework will be devoted to the design and implementation a pilot experimental study, and part of this work will be done during class meetings to facilitate troubleshooting and collaboration. Group members will contribute equally to each step of the project. Here is a preliminary list of deadlines:

10/9: Research question

10/21: Background research

10/30: Protocol

11/18: Data collection

12/2: Structured abstracts