Proceedings of the 7th International Conference on Music Perception and Cognition, Sydney, 2002

C. Stevens, D. Burnham, G. McPherson, E. Schubert, J. Renwick (Eds.). Adelaide: Causal Productions.

A STATISTICAL ANALYSIS OF TEMPI IN BACH’S D MINOR PARTITA

Richard Pulley
University of New South Wales

Bach’s Works for Solo Violin: Style, Structure, Performance1,

ABSTRACT where he states that the above approaches provide no definitive
answers, these methods are valuable—yet to a certain extent
In many music writings, authors have made assertions regarding subjective.
the tempo that is appropriate for a particular work, or the tempo
that a particular performer has chosen. A performance may be This paper outlines a statistical analysis that is more objective in
judged as being too fast or too slow; or sometimes a determining whether a particular tempo is average or extreme. It
chronological trend may be evident, such as “modern players contains no formula to determine the correct tempo, but it
play this work faster than fifty years ago.” Often these provides a way to benchmark performances so they may be
statements are subjective perceptions, based upon the listener’s compared and their relative tempi assessed. This information
gut feeling, and are sometimes supported anecdotally by may then be used alongside traditional musicological methods.
reference to only a small number of performances.
The six Sonatas and Partitas of Johann Sebastian Bach were
This paper describes how sound recordings may be used to first published in 1802. They have been central in a violinist’s
objectively measure and compare different tempi throughout the repertoire, and performances of these masterworks have varied
twentieth century, referring in particular to J.S. Bach’s Second greatly from age to age and violinist to violinist. Since
Partita in D minor for solo violin. Specifically, a simple Joachim’s recordings of 1903, most notable violinists have left
statistical technique will be presented to help better understand recordings of these works that reveal different playing styles
tempo relationships. over the past century that may be analyzed and compared. The
D minor Partita was chosen for this study because its popularity
Eighteen recorded performances spanning 65 years were increases the number of CD recordings available.
collected, including both “mainstream” and “historically
informed” performances. Segments from each movement—long 2. EXCERPT SELECTION
enough to be timed accurately, but short enough to eliminate
repeats that were not always followed)—were chosen for One cannot simply take the duration given on the back of a CD
statistical analysis. The mean and standard deviation were cover, and use this to determine the tempo used. These
calculated for all instances of each excerpt, which normalized durations include an unspecified amount of silence at the
the timings to comparable values, rather than absolute timings beginning and the end of the track, and cannot be used to
that provide no context or frame of reference with which to compare two performances when one contains the repeats and
compare other excerpts. The normalized results were presented one does not.
graphically.
Therefore, the selection of suitable excerpts was the first step. In
The normalized results demonstrated clearly which each movement, an excerpt was chosen that did not include any
performances were average, and which were extreme in tempo. repeated sections that weren’t universally followed. The
Some chronological trends were present, though the selections were also made to be long enough that any errors in
philosophical divide between “mainstream” and “historically measurement were negligible, and a true sense of the tempo had
informed” performances was found to be more significant. been established. In all cases, this was longer than 30 seconds.
These differences can be subjectively perceived (within certain See [Timing.doc] for these selections.
limits—for more details on these issues see Epstein, 1995) as
well as objectively quantified. The statistical analysis, therefore, Each performance of each movement was digitally stored onto a
provided one way of attaining objective information, and thus computer. Using computer software (Creative WaveStudio), the
an alternative approach to the more intuitive technique of tempo selected excerpt was then cut out and saved as a separate file.
assessment of recorded performance found in some writings in All leading silence was removed, and the excerpt was cut at the
aesthetics and musicology. beginning of the next note after the excerpt. The computer
software gave the duration for this excerpt, which was then
1. INTRODUCTION recorded and tabulated with the other durations (see
[Timing.doc]).
There exist a number of different approaches for determining
appropriate tempi for particular works. For eighteenth-century 3. ANALYSIS
works, guidelines are contained in contemporary treatises, such
as the link between meter and tempo, the meanings of the Italian These numbers mean little in themselves: what is important is
terms such as moderato etc., and the typical speeds of dance their speed relative to the other performances. Otherwise, no
movements. These enable performers to ascertain a suitable
range of tempi. Despite Lester’s pessimism in his recent book, 1
Joel Lester, Bach’s Works for Solo Violin: Style, Structure,
Performance, (New York: Oxford University Press, 1999), 150.

ISBN 1 876346 39 6 108

 ICMPC7, Sydney, Australia, July 2002

objective statement can be made on how extreme a particularly So that performances could be compared not only with other
slow recording is, let alone whether 74.5 seconds for the performances of the same excerpt, but across movements, each
Allemanda excerpt is more or less extreme than a 33.5 second value was normalized so that the vertical axis represented
recording of the Corrente. standard deviation units. For each individual measurement x,
the corresponding y value on the graph is calculated using the
To enable these sorts of comparisons to be made, the data were following formula:
analyzed, for each excerpt, according to the following two
statistical parameters:

1. the average (mean) duration for the performances;

2. the standard deviation for the performances. Someone playing all movements at average speed (for the
respective movements) would have their graph horizontal at
Comparing a particular excerpt’s duration with the average y=0. A faster than normal tempo is above this line (y > 0), and
duration ( ) simply reveals whether a particular recording is a slower one is below. The results are shown in Figure 2, with
faster or slower than the average. The standard deviation (σ) the recordings split into the following categories:
reveals how abnormal the duration of a particular recording is,
compared with the spread of measurements recorded. For 1. Recordings from 1930-1960;
example, if all recordings but one were within 5 seconds of the
average, then a recording that was 12 seconds slower than 2. Recordings from 1960-1980;
normal would be quite extreme, whereas if all the recordings 3. Recordings from 1980-2000 (excluding “historically
were within 12 seconds, then it would be less extreme. informed” recordings); and
A normal distribution was assumed (bell curve), which models 4. “Historically Informed” recordings on baroque
the way, over a large-enough sample, a set of values is instruments 1977-1999.
commonly distributed. A measure of how wide the spread of
values is, is the standard deviation: 4. INTERPRETATION
From these graphs, a number of interesting things may be
observed. First of all, one can tell at a glance which tempos
were the most extreme: the most extreme performance is clearly
Edinger’s Sarabanda, which is so slow that it lies at 3 standard
The square of the difference between each duration value and deviations below the mean duration for that excerpt. This means
the mean is summed together, and then divided by n-1, where n that using a normal distribution, it is estimated that this would
is the number of performances. The square root of this result be slower than 997 out of 1000 recordings! The second most
gives the standard deviation. The denominator is (n-1) because extreme tempo is the slowness of Szigeti’s Allemanda, followed
n is only used in a formula in probability with a known by the slowness of Milstein’s Corrente: both of which lie at
distribution, whereas n-1 is more appropriate in statistics for more than 2 standard deviations below the mean and thereby
estimating an unknown distribution. slower than 95-99% of recordings. Of the faster recordings, the
A value positioned at σ = 1 on the high (faster) side of the scale most extreme is the Allemanda played by Heifetz in 1952,
is faster than 65% of all recordings; for σ = 2, it is faster than which lies just short of two standard deviations above the mean.
95%, and for σ = 3, 99.7%. This is shown graphically in One small example can show how these results may shed new
Figure 1. light on previous writings. In the interesting research by
Elizabeth I Field2, at one point she examines various editions of
the Giga, and notes that “as far as metronome markings are
concerned, the more recent the edition, the faster the tempo ...
older editors had consistently indicated slower tempos.”3
However, is this a proven statement, or something that is only
anecdotal? From her own research, only three editors
(Herrmann in 1900, Hambourg in 1935 and Champeil in 1959)
indicate metronome markings and one other, Jacobsen in 1961,
inserted the tempo marking Moderato. This does not provide
enough information to indicate a pattern as Field suggests, as
the majority of the editions are silent. In fact, the preface of

2
Elizabeth I Field, Performing Solo Bach: An Examination of
Figure 1: Normal Distribution the Evolution of Performance Traditions of Bach’s
Unaccompanied Violin Sonatas from 1802 to the Present,
(Cornell University, 1999).
3
Field, 89-90.

ISBN 1 876346 39 6 109

 ICMPC7, Sydney, Australia, July 2002

Joachim and Moser’s edition of 1908 states that “the two was based on a very small number of instances, which can
Gigues in D minor and E major are to be played about easily lead to spurious conclusions.
moderato,” which indicates that they were of the same mind as
Jacobsen some 53 years later. 5. CONCLUSION
Examination of the results contained in this paper cast further The statistical analysis described in this paper is a
doubt on Field’s assertion. The normalized results for the Giga straightforward, objective approach for assessing tempo. It may
show that both fast and slow performances are found in each be used in conjunction with more traditional methods to provide
era. The problem with making the original assertion was that it support for (or question) statements regarding tempi that were
derived by other means.

Recordings 1930-60 Recordings 1960-80

2 2

1 1
Standard Deviation

0 Standard Deviation
0
1 2 3 4 5 1 2 3 4 5

-1 -1

-2 -2

-3 -3

Menuhin Heifetz36 Huberman Grumiaux Szeryng Suk

Heifetz52 Szigeti Milstein Ayo

Recordings 1980-2000 "Historically Informed" Recordings

2 2

1 1
Standard Deviation

Standard Deviation

0 0
1 2 3 4 5 1 2 3 4 5
-1 -1

-2 -2

-3
-3

Mintz Perlman Kuijken Wallfisch

Ricci Edinger Luca Podger

Figure 2: Tempo Comparisons of 18 Violinists

ISBN 1 876346 39 6 110

 ICMPC7, Sydney, Australia, July 2002

8. -----. Bach: Sonatas & Partitas for Violin. Vln. Henryk

5. ADDITIONAL FILES Szeryng. Deutsche Grammophon Double,
453 004-2, 1996. Recorded 1967.
[Timing.doc]
9. -----. Bach: Sonatas & Partitas. Vln. Josef Suk. EMI
6. REFERENCES Classics, CDFB 73644, 2000. Recorded 1970.

1. Epstein, D. Shaping time: music, the brain, and 10. -----. Bach: Sonatas and Partitas for Solo Violin. Vln.
performance. New York: Schirmer, 1995. Felix Ayo. Philips, 422 940-2, 1978. Recorded 1975.

2. Field, Elizabeth I. Performing Solo Bach: An 11. -----. Bach: Sonaten & Partiten. Vln. Nathan Milstein.
Examination of the Evolution of Performance Traditions Deutsche Grammophon, DG 457-701-2, 1998. Recorded
of Bach’s Unaccompanied Violin Sonatas from 1802 to 1975.
the Present. Cornell University, 1999.
12. -----. Bach: Sonatas & Partitas for Unaccompanied
3. Houle, George. Meter in music: 1600-1800. Violin. Vln. Sergiu Luca. Nonesuch, FD 73030, 1990.
Bloomington & Indianapolis: Indiana University Press, Recorded 1977.
1987.
13. -----. Bach: Sonatas and Partitas for Solo Violin. Vln.
4. Lester, Joel. Bach’s Works for Solo Violin: Style, Sigiswald Kuijken. Deutsche Harmonia Mundi, 77043-
Structure, Performance. New York: Oxford University 2-RG, 1981.
Press, 1999.
14. -----. Sonaten & Partiten. Vln. Shlomo Mintz.
5. Philip, Robert. Early Recordings and Musical Style: Deutsche Grammophon, DG 413 810-2, 1984. Recorded
Changing Tastes in Instrumental Performance 1900- 1983-4.
1950. Cambridge: Cambridge University Press, 1992.
15. -----. Sonaten und Partiten. Vln. Itzhak Perlman. EMI
Classics, 7 49483 2, 1988. Recorded 1986-7.
7. DISCOGRAPHY
16. -----. Partita No. 2 in D minor from Ricci: Celebrating
Recordings are listed chronologically according to their year of Six Decades On Stage. Vln. Ruggiero Ricci. One-
recording. Eleven, URS-92033, 1992. Recorded 1988.
1. Bach, J. S. Partita in B minor: Bourrée & Sonata in G 17. -----. Violin Sonatas & Partitas for Solo Violin Vol. 2.
minor: Adagio from Joseph Joachim: The Complete Vln. Christiane Edinger. Naxos, 8.550570, 1992.
Recordings (1903); Pablo de Sarasate: The Complete Recorded 1991.
Recordings (1904); Eugène Ysaÿe: A selection of his
recordings (1912). Vln. Joseph Joachim. Pavilion, 18. -----. Sonatas and Partitas for solo violin. Vln.
OPAL CD 9851, 1992. Recorded 1903. Elizabeth Wallfisch. Hyperion, CDD22009, 1997.
Recorded 1997.
2. -----. Bach: Sonaten und Partiten. Vln. Yehudi
Menuhin. EMI Classics, CHS 7 63035 2, 1989. 19. -----. Sonatas & Partitas vol. 1. Vln. Rachel Podger.
Recorded 1934-36. Channel Classics, CCS 12198, 1999.
3. -----. Partita No. 2 in D minor from Jascha Heifetz in
the Golden Thirties Vol. I. Vln. Jascha Heifetz.
Gramofono 2000, AB78511, 1994. Recorded 1935-
1937.

4. -----. Partita in D minor from Huberman. in Recital.

Vln. Bronislaw Huberman. Arbiter 105, 1997.
Recorded 1942.

5. -----. Sonatas & Partitas. Vln. Jascha Heifetz. BMG

Classics, 09026-61748-2, 1994. Recorded 1952.

6. -----. 6 Sonatas & Partitas for violin alone. Vln. Joseph

Szigeti. Vanguard Classics, OVC 8021/22, 1991.
Recorded 1955-6.

7. -----. Bach: Complete Solo Sonatas & Partitas for Solo

Violin. Vln. Arthur Grumiaux. Philips Duo, 2PH2 438-
736-2 , 1994. Recorded 1960-61.

ISBN 1 876346 39 6 111