Content uploaded by Claudia Fritz
Author content
All content in this area was uploaded by Claudia Fritz on May 22, 2017
Content may be subject to copyright.
1
Listener evaluations of new and Old Italian violins
Claudia Fritza, Joseph Curtinb, Jacques Poitevineaua, Fan-Chia Taoc
a Lutheries–Acoustique–Musique Team, Institut Jean Le Rond d‟Alembert UMR 7190, Université Pierre et Marie
Curie Univ Paris 06, CNRS, Sorbonne Universités, 75005 Paris, France;
b Joseph Curtin Studios, Ann Arbor, MI 48103;
c Research & Development, D‟Addario & Company, Farmingdale, NY 11735
Published in PNAS, May 2017 (doi: 10.1073/pnas.1619443114)
Abstract
Old Italian violins are routinely credited with playing qualities supposedly unobtainable in new
instruments. These qualities include the ability to project their sound more effectively in a concert
hall – despite seeming relatively quiet under the ear of the player, compared with new violins.
While researchers have long tried to explain the “mystery” of Stradivari‟s sound, it is only
recently that studies have addressed the fundamental assumption of tonal superiority. Results
from two studies show that under blind conditions experienced violinists tend to prefer playing
new violins over Old Italians. Moreover, they are unable to tell new from old at better than
chance levels. The current study explores the relative merits of Stradivari and new violins from
the perspective of listeners in a hall. Projection and preference are taken as the two broadest
criteria by which listeners might meaningfully compare violins: Which violins are heard better,
and which are preferred? In two separate experiments, three new violins were compared with
three by Stradivari. Projection was tested both with and without orchestral accompaniment.
Projection and preference were judged simultaneously by dividing listeners into two groups.
Results are unambiguous. The new violins projected better than the Stradivaris, whether tested
with orchestra or without; the new violins were generally preferred by the listeners; and the
listeners could not reliably distinguish new from old. The single best-projecting violin was
considered the loudest under the ear by players, and on average, violins that were quieter under
the ear were found to project less well.
Significance Statement
Old Italian violins are widely believed to have playing qualities unobtainable in new violins,
including the ability to project their sound more effectively in a hall. Because Old Italian
instruments are now priced beyond the reach of the vast majority of players, it seems important to
test the fundamental assumption of their tonal superiority. A recent study found that under blind
conditions violin soloists generally prefer new violins, and are unable to distinguish between new
and old at better than chance levels. This paper extends the results to listeners in a hall. We find
they generally prefer new violins over Stradivaris, consider them better-projecting, and are no
better than players at telling new and old apart.
2
Introduction / context
Violinists generally agree that individual violins vary considerably in their ability to project –
projection referring to how well an instrument can be heard at a distance, especially over a
background of competing musical sound. The paradigmatic test for projection is the violin
concerto, where a soloist is expected to carry over a full orchestra, often in a very large hall. Old
Italian instruments are commonly believed to have the advantage over new ones in this regard.
“What made the old instruments great was their power in a hall,” wrote the distinguished violinist
Earl Carlyss [1] in response to a 2010 blind study [2]. Somewhat paradoxically, Old Italian
violins are also commonly described as being relatively quiet under the ear of the player,
compared with new instruments. According to concertmaster Frank Almond (who plays a
Stradivarius): “a peculiar (and sublime) aspect of great old Italian instruments is that the sound
somehow expands and gains more complexity from a distance, especially in a concert hall.” He
contrasts this with many modern instruments, which seem to have a large sound under the ear, but
may not “carry past the sixth row.” [3]. Similarly, renowned cellist Steven Isserlis (who plays a
Stradivarius cello) wrote, “A famous (and curious) feature of Stradivarius instruments is that their
tone seems to increase with distance. As a rule, if my tone sounds small to me, it means that it is
travelling out into the hall . . .” [4].
Most celebrated violinists since the early 1800s have played instruments by Stradivari or
Guarneri del Gesu, and this is often taken as evidence that these violins possess some
combination of playing qualities not found in newer instruments. Over the past two centuries,
numerous informal playing and listening tests have challenged the notion [5,6,7]. More recently,
a pair of well-controlled studies invited players to blind-test new violins against those by
Stradivari and other Old Italian makers [2,8]. In both studies the most-preferred violin was new,
and there was a general preference for new instruments over old. Moreover, players appeared
unable to reliably distinguish between new and old.
These studies focused on judgments made by violinists while playing. Because projection
must by definition be judged by listeners at a distance, many questions remained unanswered. Do
Stradivari violins in fact out-project high-quality new instruments? Are better-projecting
instruments typically quieter under the player‟s ear? Are the instruments that soloists prefer to
play the ones that are found to project best in a hall? And can listeners tell whether a soloist is
playing an Old Italian violin rather than a new one?
To answer these questions, we ran two experiments in which neither players nor listeners
knew the identity of the test instruments. Projection and preference were taken as the broadest
criteria by which listeners might meaningfully separate two violins in a hall: Which violin is
heard better, and which is preferred? The first study was held in 2012 in a small concert hall near
Paris, France, immediately following the above-cited preference test [8]. The second was held in
2013 in a larger hall in New York City. These studies, referred to here as the Paris and New York
experiments, involved 55 and 82 listeners, respectively. In each case three new violins were
tested against three by Stradivari. In Paris, the violins were tested with and without orchestral
accompaniment. In New York, all tests were unaccompanied.
3
Materials and Methods
General methodology
In each experiment, a group of listeners was presented with a series of pair-wise comparisons.
Each pair consisted of a new violin and a Stradivarius (the selection of the instruments is detailed
below), and was played by a soloist behind an acoustically transparent screen. Players used their
own bows, as they typically would when evaluating instruments in real-life. Modified welders‟
goggles together with much-reduced ambient lighting made it impossible for players to identify
instruments by eye.
The Paris experiment was conducted in the 300 seat concert hall 'A coeur de ville' in Vincennes
(a hall well-regarded for its acoustics), and the New York experiment in the 860 seat Great Hall
at the Cooper Union building. Listeners were free to sit anywhere in the halls, though in Paris the
first five rows were excluded.
Seven internationally renowned soloists (named Soloist 1-7 in SI) took part in the Paris
experiment, six of whom participated in the player-preference experiment [8] conducted in the
preceding two days. In New York, one soloist from the Paris experiment was joined by another
distinguished violinist (SI Text). In both tests, the number of soloists who normally performed on
an Old Italian instrument was about equal to the number normally performing on a new one (SI
Text).
In Paris we invited experienced listeners, by which we mean listeners with expertize relevant to
our subject; they included violin makers, players, musicians, audiophiles, music critics,
composers, and acousticians. The New York experiment was a public event within Mondomusica
(an international exhibition of handcrafted musical instruments); listeners were people who saw
the advertisement and were interested in the topic. In both experiments, listeners who returned
incomplete evaluation sheets were omitted from the analysis. The distribution of the remaining
listeners (55 for Paris, 82 for New York) is provided in SI. At the beginning of the experiment,
listeners were informed orally that participation was voluntary, and that all answers would be
anonymized. Under these conditions, CNRS (the employer of the principal investigator) does not
require any formal approval from a licensing committee prior to the experiment.
Detailed procedure for the Paris experiment
This experiment was divided into three parts: Parts 1 and 2 were concerned with the relative
projection of old and new violins; Part 3 tested whether listeners could tell if the violins being
played were new or old. The full score sheet given to the participants is available in SI.
1) Part 1
a) Testing Projection
Relative projection was tested by a series of pair-wise comparisons, structured as follows. A
soloist played a short excerpt on each violin of a pair, then did the same thing again. This so-
called ABAB format ensured listeners could hear each violin both before and after the other. The
whole sequence was done twice for each pair, first using a solo excerpt, and then a concerto
excerpt accompanied by the orchestra (details about the orchestra can be found in SI). It was
4
made clear to listeners that the same pair was being played throughout. In order to test each of the
Stradivari violins against each of the new, nine new/old pairs were formed. These were played in
random order by Player 1, then after a short break by Player 2, but with the order of the violins
within the pairs reversed, and the order of the pairs modified. Note that in choosing soloists for
this test, we selected Player 1 as one who regularly plays a new violin (Soloist 1, see SI), and
Player 2 as one who regularly plays an old (Soloist 2).
To facilitate testing, a researcher placed each pair of violins in turn on a table. A large sign
(legible even in the relative darkness) placed in front of each instrument indicated to the soloist
whether it was violin A or violin B. To help listeners keep track, a large video screen indicated
which violin of which pair was being played (e.g., Pair 5, violin A).
Musical excerpts were selected from the standard repertoire, and were well-known to all soloists.
The unaccompanied excerpts were chosen to cover a good deal of the instrument‟s range within
about ten seconds. Soloist 1 played the opening solo (bars 23-27) from the first movement of the
Tchaikovsky Violin Concerto op. 35. Soloist 2 played the opening solo (bars 90-94) from the first
movement of the Brahms Violin Concerto in D major, op. 77. The accompanied excerpts were a
bit longer (~20 s), and were passages in which the solo violin might easily be „covered‟ by the
orchestra: Soloist 1 played bars 187-202 from the third movement of the Brahms Concerto, and
Soloist 2 played bars 41-43 from the 2nd movement of the Sibelius Violin Concerto, op. 47.
While projection is presumed to be an intrinsic quality of an instrument, it may well be affected
by how good a “fit” a particular instrument is for a particular player. An experimental design
allowing separation of these two factors would, however, require the audience to rate nine pairs
of violins twice. The resulting 36 repetitions of the same excerpt would, we believe, have been an
unreasonable challenge to listener concentration. To help avoid boredom and fatigue, the second
soloist was therefore given a different excerpt. This trade-off between experimental design and
the capacities of the listeners made it impossible to later decouple the intrinsic qualities of the
violins from the influence of individual players.
For each pair of violins, listeners were required to answer the question: “Which violin projects
best?” The meaning of projection was left undefined, but a questionnaire (SI Text) on the subject
was sent to participants after the experiment ([9] provides a psycho-linguistic analysis of the
answers). Listeners recorded their evaluations by placing a mark at the position of their choice
along a continuous scale, where one end indicated that violin A was far superior to B in terms of
projection. A mark at the opposite end indicated the converse, and a mark at the midpoint
indicated the two projected equally well (SI Text). Though it was impossible for the soloists to
evaluate violins in this way while playing them, they did indicate (to CF) which instrument of
each pair they believed projected better, or that the two were equal.
b) Selection of test violins
In the Paris preference tests [8], ten soloists were presented with six new and six Old Italian
violins (including five by Stradivari), which had themselves been preselected from a pool of nine
Old Italian and 15 new violins made available by dealers, collectors, players, and makers (see [8]
for details of the pre-selection process). Over the course of two 75-minute individual testing
sessions, the soloists were asked to (1) eliminate any instruments they found unsuitable; (2)
5
choose from the remaining their four favorites, then rank these four in terms of overall
preference, and (3) select the single violin that would best replace their own violin for an
upcoming concert tour. Time limits restricted the number of violins that could be tested for
projection to three new and three old. The following factors were considered in selecting these
from the six old and six new violins which the soloists had evaluated:
1. The number of times an instrument was the top choice of a soloist. Five violins were
chosen by at least one soloist. Three were new (N5, N10, & N9) and two were by
Stradivari (O1 & O4).
2. Overall preference-score, based on soloists‟ choices of favorite violins. The three top-
scoring new were, in descending order, N5, N10 and N11. The three top-rated old were
O1, O8, and O4.
3. Measured acoustical output. Acoustical measurements were made of all instruments (SI
Text). Preliminary results were used for a quick measure of the relative sound output of
each instrument, per unit force at the bridge. The three new violins with the highest output
were (in descending order) N5, N11, & N10. The three old were O6, O4, & O1.
Violin
Times chosen as
favorite
Rank by overall preference
score within old/new category
Rank by acoustical output
within old/new category
N5
N10
N11
N9
4
1
0
1
1
2
3
3
1
3
2
4
O1
O4
O8
O6
3
1
0
0
1
3
2
4
3
2
4
1
Table 1: Data used in the selection of three new and three old violins for the listening
experiment
Based on the data in Table 1, we chose the following violins: N5, N10, N11, O1, O4 and O6. The
inclusion of O6 needs justification, as it was not chosen by any soloist and was rated 9th of 12 for
preference. It was selected mainly on the basis of its acoustical output, which was the highest of
all the old violins. Moreover, unlike the well-liked O8 (also a strong contender) it was a
Stradivarius, and so of most direct relevance to our hypothesis.
2) Part 2
Would the violins most frequently chosen by soloists in preference tests out-project those the
soloists most often rejected? To test this, we had listeners compare the two most frequently
chosen violins (N5 and O1) with two most often rejected (N2 and O12). The resulting four
new/old pairs were tested as in Part 1, with players and listeners being asked the same questions.
The solo excerpts were as above, but to increase musical variety, different orchestral excerpts
were used: Player 1 (Soloist 3, see SI) played a 20s excerpt from the second movement of the
6
Brahms concerto (bars 93-98), and Player 2 (Soloist 4) played a 16 s excerpt from the third
movement of the Sibelius concerto (bars 254-262).
3) Part 3
To test whether listeners could distinguish old violins from new, they were presented with a
succession of seven concerto excerpts, each played (with orchestra) by a different soloist
(Soloists 1-7, see SI). Unbeknown to listeners, the soloists played their own personal violins, thus
forestalling concerns that they might be insufficiently acquainted with the test instruments to
fully exploit their individual qualities. Each soloist had prepared a three-minute excerpt from
either the Mendelssohn, Sibelius, or Beethoven concertos. After each performance, listeners were
asked: “Do you think the violin is old or new? Why?”
Detailed procedure of the New York experiment
a) Testing projection and preference
The New York experiment tested how well the listeners‟ judgments of projection correlated with
judgments of preference. It was structured in much the same way as Part 1 of the Paris
experiment, with three Stradivari and three new violins yielding nine new-old pairs. Listeners
were not asked to judge projection and preference at the same time: such a task would be
cognitively challenging; moreover, the answer to the first question might influence the answer to
the second, creating artificial correlations. Instead, listeners were randomly divided in two
groups; one being asked first which violin of a pair they preferred, and why, and group 2 which
violin projected better. After Player 1 (Soloist 8, see SI) had played the nine pairs, the questions
were switched for Player 2 (Soloist 9), so that the first group now judged projection, and the
second group, preference. Both projection and preference were reported on continuous scale, as
in Paris. Unlike Paris, listeners were asked to define what projection meant to them by means of a
multiple-choice questionnaire (SI Text). As in Paris, the soloists were asked to estimate
projection on a simpler scale: A, B or equal. All tests were done without orchestral
accompaniment.
b) Selection of test violins
We would have liked to use the same old and new test violins as in Paris, but just two of them
were available: the new violin N5 and the Stradivari O6. Two additional new violins were chosen
by a pre-selection process (SI Text) from 15 violins submitted by violin-makers. The two
additional Stradivari were the only ones made available to us at the time.
Results
In this study, some Bayesian statistical procedures supplement traditional frequentist inferences,
using LePAC software [10,11]. They assume an uninformative prior distribution (i.e. no
information is used other than that contained in the data, and no particular hypothesis is favored a
priori). Hereafter, Bayesian statements are notated Pr*. They are naturally conditional on the data
at hand, but conditional notation is omitted for the sake of brevity.
7
Paris experiment
a) Relative projection of three new violins and three Stradivarius (Part 1)
Figure 1 shows the relative projection of the nine new/old pairs tested in Paris. On the left are
results for unaccompanied violin, and on the right for violin with orchestra. N10 was about equal
to O1 and O4 in terms of projection, but in all other pairings, the new violins out-projected the
old. The spaces between the blue and red error bars (which are especially wide for N11) suggest
that projection depends somewhat on the player. While this may well be the case, no firm
conclusions can be drawn, since the two players used different excerpts.
Figure 1: Relative projection of the nine new/old pairs tested in Paris. On the left are results for
unaccompanied violin, and on the right for violin with orchestra. Listeners scored projection on
a continuous scale between 0 and 1. The charts show the average of their scores for each of two
players – P1 in red, and P2 in blue. Error bars correspond to the standard error of the mean.
The two players were asked to choose one of the following for each pair: (1) violin A projects
best (indicated by a mark at the left edge of the graph; (2) violin B projects best (mark at right
edge), or (3) the two violins project equally well (mark at center line). P1’s judgements are
shown in red and P2’s in blue.
Do violins that project well when played alone also project well over an orchestra? We find a
striking consistency across the two conditions. Averaging the evaluations of all listeners and both
players, the observed effect is -0.002 (solo - orchestra), t(52) =-0.25, p = 0.40. Furthermore, we
can state that Pr*(|true effect| < 0.02) = 0.95 (on a 0-2 scale). This means, given the present data,
a 95% probability that the true effect (i.e. the effect in the whole population of experienced
listeners) is between -0.02 and 0.02. While non-null, the 0-1 scale means its value is very small,
and a conclusion of negligibility can be drawn. In other words, very similar results are obtained
whether testing projection with or without orchestra.
Do listeners from different professional backgrounds evaluate projection differently? By design,
the entire audience consisted of experienced listeners, but they came from various professional
backgrounds, and included musicians, violin-makers, and acousticians. All groups produced
rather similar results. The observed root mean squared of the differences between groups is 0.04,
F(2,52) = 1.75, p = 0.18 and Pr*(|true effect| < 0.08) = 0.95 (on a 0-2 scale).
8
Given the consistencies described above, we felt justified in averaging the data from all listeners,
with and without orchestra. Figure 2 shows the differences in projection between each new violin
and all three old violins, and then between each old violin and all three new. Again we see a large
new/old effect, with new violins judged to project better than old. The observed difference in
projection is 0.19, t(52) = 15.14, p < 10-4, 95% CI: [0.16; 0.21]. Pr*(true effect > 0) > 1-10-4) and
Pr*(true effect > 0.17) = 0.95 (on a 0-1 scale). The apparent effect of the player is moderate
(0.13) but clear as t(52) = -5.82, p < 10-4, 95% CI: [0.09; 0.18] (in this case of a one-degree-of-
freedom comparison with exactly the same assumptions as in the frequentist theory, the
frequentist CI and the standard Bayesian CI coincide so that we can state Pr*(0.09 < true effect <
0.18) = 0.95. Therefore, for similar cases in the following, we report only the CI limits.). Again,
we cannot know whether this is due to the player or to the use of different excerpts.
Figure 2: Differences in projection between each new violin and all three old (top three rows)
and between each old violin and all three new (bottom three rows) tested in Paris. Results are
averaged for all listeners, with and without orchestra. Error bars correspond to the standard
error of the mean.
b) Relative projection of two preferred and two rejected violins (Part 2)
In Part 2 of the Paris experiment, the two most-preferred violins (N5 & O1) were compared with
the two least-preferred (N2 & O12). Figure 3 shows that N5 clearly out-projected N2 & O12, and
that O1 was about equal in projection to N2 & O12. There is good agreement between player and
listener evaluations.
Figure 3: Difference in projection between the two most preferred violins and the two most
rejected. Error bars correspond to the standard error of the mean. The players' evaluations are
single points at either 0 or 1 (left or right), with the same color code as for the audience. As the
9
soloists gave the same answer with and without orchestra, there is no distinction between the two
conditions, as in figure 1.
c) Comparison between soloist and listener evaluations of projection
Violinists generally believe that loudness under the ear is not directly related to projection, and
that a violin‟s projection cannot be reliably estimated simply by playing it. During individual
sessions that took place prior to the experiment [8], we asked each soloist to estimate projection
and loudness-under-ear for three violins: their own, their favorite, and their next favorite of the
opposite new/old category. This was done in a rehearsal room and again in the concert hall (see
Table 2). Because of the selection method, only a few violins – N5, O1, and N10 – were rated by
five or more soloists.
Violin
Projection
Loudness-under-ear
Soloists‟ own
8.2 (1.2)
7.9 (1.2)
Old
7.9 (1.3)
7.9 (1.2)
New
8.6 (1.3)
8.6 (1.2)
Table 2: Mean (with standard deviation) of the loudness and projection ratings for soloists’ own,
old, and new violins, averaged across the two conditions – rehearsal room and concert hall.
While the soloists gave their own violins (seven of which were old) a slightly higher rating for
projection than for loudness (8.2 versus 7.9), when it came to the test instruments, the old got the
same averaged rating (7.9) for both loudness and projection, as did the new (8.6). Since the new
violins did in fact out-project the old, loudness-under-ear may be more closely related to actual
projection than is generally believed.
Violin
Averaged estimated
projection
Average loudness-
under-ear
Number of estimates
N5
9,4
9,2
11
O1
8,8
7,6
6
N10
7,5
8,4
5
Table 3: Averaged estimates for projection and loudness-under-ear for the three violins rated by
at least five soloists during their private sessions in the Paris experiment
Table 3 shows the averaged projection and loudness ratings for the three most-preferred violins,
N5, O1, and N10. The projection-estimates wrongly suggest that O1 projects better than N10.
The loudness ratings, however, at least get the ordering correct: N5, N10, O1 (in order of
decreasing projection). The spread between projection and loudness seen in O1‟s ratings (8.8 v‟s
7.6) can be attributed to the four players who chose it as their favorite. Their ratings were (for
projection and loudness respectively) 10 & 6; 9 & 7; 9 & 8, and 9.5 & 9. By contrast, the two
soloists for whom O1 was not the favorite violin rated it very similarly for the two criteria (8 & 8
and 7.5 & 7). Given the small sample size, not too much can be made of this, though it does
10
suggest that the widely-held belief that a Stradivari can project well despite sounding quiet under
the ear may lead some players to over-estimate projection for instruments they like to play but
find relatively quiet under the ear. There is indeed considerable anecdotal evidence that players
and makers consider projection and loudness as two well separated concepts. This is illustrated in
the analysis of the questionnaires filled in by participants in the New York experiment (73% of
whom were makers and musicians). They were asked to select which of seven possible
definitions best captured what projection meant to them (SI Text). Table 4 summarizes the
results. The question referring to loudness was selected far less often than all but one of the
others. More research is needed to investigate whether listeners do indeed evaluate projection and
loudness differently, and if so, what the underlying acoustical parameters for each might be.
Possible definition of 'Which violin projects
better?'
Number of times it was chosen
Which violin has the best carrying power?
29
Which sounds loudest?
14
Which sounds most full?
30
Which seems most “present” to your ears?
30
Which seems to fill the hall best?
30
Which comes through to your ears more
clearly?
22
Which comes through to your ears more easily?
10
Table 4: Number of times each possible definition of projection was chosen during the New York
experiment.
d) Distinguishing old from new (Part 3)
We asked listeners to guess whether each of seven violins was old or new. Thirty-nine of the 46
subjects made a guess about all seven violins. (Incomplete evaluation sheets were discarded from
the analysis). The distribution of their correct guesses is shown in figure 4. It shows a unimodal
distribution centered between three and four. In all, just 122 of 273 (or 44.7%) of the guesses
were correct.
11
Figure 4: Distribution of correct guesses about the age of seven violins by 39 subjects.
It is interesting to see how the correct guesses were distributed among violins. For all but two
instruments (one old, one new) the confidence intervals (SI Text) include 50%, so the chance
hypothesis cannot be ruled out. For the other two violins, there is a systematic error: listeners
guessed that the old violin was new, and the new one was old. A supplementary analysis built on
the hypothesis that the guesses were entirely governed by chance is available in SI.
New York experiment
The Paris Experiment left at least one important question unanswered: How do listeners‟
evaluations of projection relate to their evaluations of preference? It is easy to imagine, for
example, that violin A is found to project better than violin B, but violin B is preferred for its tone
quality. Exploring the relationship between projection and preference was the principal objective
of the New York Experiment.
Figure 5: Relative projection (left side) and preference (right) for each of the nine new/old pairs
tested in New York. Listener scores are on a continuous 0 to 1 scale. The charts show the average
of their scores for each of two players – P1 in red, and P2 in blue. Error bars correspond to the
standard error of the mean. The two players were asked to choose one of the following for each
pair: (1) I prefer violin A (x at the left edge of the graph; (2) I prefer violin B (x at right edge), or
(3) I like them equally (x at center line). P1’s judgements are shown in red and P2’s in blue.
12
Figure 5 shows relative projection (left side) and preference (right) for each of the nine new/old
pairs. Looking first at projection, results are very similar to those of the Paris Experiment. New
violins are found to project better than old, though in this case the difference happens to be even
larger: observed difference = 0.27, t(80) =14.55, p < 10-6, 95% CI: [0.23; 0.31], Pr*(true effect >
0) > 1- 10-6 and Pr*(true effect > 0.24) = 0.95 (on a 0-1 scale). The better projection of the new
compared to the old therefore seems robust across a change in halls, listeners, players, and
violins. The observed effect of the player/order/excerpt is 0.03 (on a 0-2 scale), thus smaller than
the effect of the player/excerpt in Paris (0.13 on a 0-2 scale). However, little can be inferred about
the player effect as it cannot be separated from the effect of the excerpt (in Paris and New York)
and the order (in New York only).
Looking at preference (Figure 5, right side), listeners clearly preferred new violins over old:
observed difference = 0.27, t(80) = 14.67, p < 10-6, 95% CI: [0.23; 0.30], Pr*( true effect > 0) >
1-10-6) and Pr*( true effect > 0.24) = 0.95 (0-1 scale). The figure shows slightly different results
for the two players, but we cannot conclude there is a player effect, as it is inseparable from the
possible effects of different excerpts and testing order (i.e., testing preference first rather than
projection first). We can however conclude that if there is a player/order/excerpt effect on
preference, it is rather limited: the observed value is 0.07 and t(80) = 1.87, p = 0.07,
Pr*(|true effect| < 0.13) = 0.95 (on a 0-2 scale).
There is an almost 90% agreement between soloists and listeners about preference, at least when
testing them together in a hall. Consider that player/listener agreement was just 70% when the
soloists were asked to estimate projection under similar conditions in Paris.
Given that listeners generally preferred new violins over old and found them better-projecting,
one may ask whether the new were preferred precisely because they projected better. As the
listeners were asked to explain their preference choices, we look to the explanations for insight.
Fifty-eight of 82 listeners used words related to projection in at least one case (SI Text), meaning
that just 158 of the total of 548 collected statements refer to projection. (Listeners evaluated nine
pairs, but many listeners didn't provide any explanations, or else provided them for just a few of
the pairs). Among these 158 statements, some were of the kind 'I preferred violin A because of its
tone quality, despite it being less loud/projecting than violin B' rather than 'I preferred violin A
because it projects better'. This goes well with the extra definitions of projection given by two
participants: “Doesn't always mean the best sound but one that cuts through noise, fills a room
and allows the musician to be heard” and “full and loud, but not necessarily beautiful; sometimes
harsh.” All of which to say, listener-explanations do not suggest any obvious link between
preference and projection.
Conclusions and discussion
This study compares Stradivari and new violins principally from the listener‟s perspective.
Projection and preference are taken as the broadest criteria by which listeners might meaningfully
separate violins in a hall. Which violins are heard better, and which are preferred? Also explored
is the relationship between evaluations made by listeners and those made by players. Two
13
separate experiments were conducted, one in the suburbs of Paris, the other in New York City. In
Paris, violins were played with and without orchestra by blindfolded soloists behind an
acoustically transparent screen. The same was done in New York, but without an orchestra.
The results are unambiguous: listeners found that new violins projected significantly better than
those by Stradivari. Moreover, listeners preferred new violins over old by a significant margin.
Though the listeners came from various professional backgrounds (they included musicians,
violin-makers, and acousticians), very similar results were obtained from all backgrounds.
We find a strong correlation between projection with and without orchestra. This seems fortunate
for both players and researchers, in that an orchestra is evidently not required to meaningfully test
projection.
Projection estimates made by players during private sessions [8] did not agree very well with
judgments made later by listeners in the hall. When playing for an audience in a hall, however,
about 70% of player-estimates of projection agreed with listener-judgments. Put another way,
players were wrong about 30% of the time, confirming the wisdom of bringing a trusted listener
along when trying out instruments. On the other hand, when players in the New York experiment
were asked, after playing each new/old pair, which violin they preferred, their preferences were
in strong agreement with those of the audience.
During private sessions in Paris [8], four of ten soloists chose a Stradivari as the instrument “best
suited to replace their own violin for an upcoming tour.” We find here that these four soloists
chose violins with significantly less projection than the two best-projecting new instruments. The
most-preferred Stradivari barely out-projected the least-preferred Stradivari and the least-
preferred new violin. The New York results suggest that the most-preferred Stradivari in Paris
would also be less-preferred by listeners than the better-projecting new violins.
The most-preferred new violin in the Paris Experiment was considered louder under the ear than
the most-preferred old – and indeed than every other violin that was rated for loudness. Soloists
who chose a Stradivari may have preferred a violin that was less loud under the ear, at the same
time believing it would project well, for the three players who chose the most-preferred Stradivari
rated it notably higher for projection than for loudness.
Previous studies [2,8] indicate that under blind conditions violinists readily separate violins they
like from those they do not, but are unable to reliably distinguish new from old. We show here
that listeners (including violin-makers and musicians) are no better at telling whether a soloist‟s
violin is new or old.
The test violins were necessarily few in number, and there is no way of knowing how
representative they are of the larger population. While there is every reason to believe there are
Stradivari violins with outstanding projection, we find little reason to suppose players would find
them quiet under the ear. Violins (old or new) that are (1) relatively quiet under the ear, and (2)
out-project apparently louder violins would be worthy subjects for future research. To date, we
have found no evidence for their existence.
A player‟s requirements for projection are clearly dependent on musical context. Even a
relatively quiet violin will be heard when played unaccompanied in a small hall, and a good
14
conductor will do everything possible to ensure that a soloist with a relatively quiet violin can
still be heard over an orchestra. In auditions and playing competitions, however, jurors listen to
one player after another, making apparent the relative projection of their instruments. While it is
common for gifted young players to borrow Old Italian violins for such occasions, our results
suggest they would be better served by good new violins – at least so long as the identity of the
instruments remains unknown to those judging.
Note that this study relied on soloists, whose primary need is to be heard over an ensemble. The
vast majority of violinists, however, are required to blend into ensembles. Research is needed to
clarify the relationship between projection and “blendability.” Less-projecting violins could have
the advantage in this regard.
A belief in the near-miraculous qualities of Old Italian violins has preoccupied the violin world
for centuries. It may be that recent generations of violin-makers have closed the gap between old
and new, or it may be that the gap was never so wide as commonly believed. Either way, the
debate about old versus new can perhaps be laid aside now in favor of potentially more fruitful
questions. What, for example, are the physical parameters determining the playing qualities of
any violin, regardless of its age or country of origin?
Acknowledgements
We thank all dealers, makers, players, and collectors for their kindness and trust in making
available these valuable instruments. Special thanks go to the soloists for their participation and
enthusiasm, to the numerous audience members for their dedication and patience. Special thanks
as well to the COSU players, their conductor Vincent Barthe and their administrators, Agnès
Pussilieux and Lucie Paladino, for their kindness and professionalism! We acknowledge Philip
De La Croix and Stéphane Agasse for the use of the Vincennes auditorium, and for their
logistical help, and Kojiro Yamada and NHK for their amazing support in New York. We would
also like to thank the LAM team (particularly Indiana Wollman and Laurent Quartier), the
Borsarello family, David Griesinger, Stefan Avalos, and Suzanne Ortmeier for their kindness and
help throughout the Paris experiment. And, finally, we are grateful to the Centre National de la
Recherche Scientifique and Université Pierre et Marie Curie for funding this experiment and to
the Violin Society of America for additional financial support.
References
1. Wade N (January 2 2012) In classic vs. modern violins, beauty is in the ear of the
beholder. NY Times. Available at http://www.nytimes.com/2012/01/03/science/inplay-off-
between-old-and-new-violins-stradivarius-lags.html. Accessed April 10, 2017.
2. .Fritz C., Curtin J., Poitevineau J., Morrel-Samuels P. and Tao F.-C. (2012) Players
preferences among new and old violins. Proceedings of the National Academy of Sciences
of the USA. 109: 760-763.
15
3. Frank Almond (2012) “They blinded me with Science”. Personal blog.
http://www.insidethearts.com/nondivisi/they-blinded-me-with-science/
4. Steven Isserlis (January3, 2012) “Stradivarius v modern violins: why this latest study
strikes a discordant note”. The Guardian Available at
https://www.theguardian.com/music/musicblog/2012/jan/03/stradivarius-v-modern-
violins-study. Accessed April 10, 2017.
5. Moya, H. and Piper, T. (1916). Violin tone and violin makers (Chatto & Windus, London).
6. Windust E (2009) From the archive: 100 years ago, a Paris competition sets out to
compare old and new violins. The Strad 120: 95 .
7. Coggins, A. (2007). “Blind faith”, The Strad 118: 52-55.
8. Fritz C., Curtin J., Poitevineau J., Borsarello H., Wollman I., Tao F.-C. et Ghasarossian T.
(2014) Soloist evaluations of six Old Italian and six new violins. Proceedings of the
National Academy of Sciences 111: 7224-7229.
9. Dubois D. and Fritz C. (2016) “La projection du violon : analyse sémantique” (Violin
projection : semantic analysis), In Le violon en France du XIXe siècle à nos jours, ed C.
Fritz and S. Moraly (in press).
10. Lecoutre, B. and Poitevineau, J. (2005) Le logiciel “Le PAC” dédié aux procédures
bayésiennes. La Revue de Modulad 33. Available at https://www.rocq.inria.fr/
axis/modulad/logiciels.htm#lepac. Accessed April 10, 2017.
11. Lecoutre, B. and Poitevineau, J. (2014) The Significance Test Controversy Revisited. The
Fiducial Bayesian Alternative. New York: Springer.
16
Supporting information
Information about the soloists and the violins they played during the experiment
Venue
Soloist
Name
Own violin
Favorite violin
Paris
1
Ilya Kaler
new
N5
2
Yi-Jia Susanne Hou
old
N5
3
Tatsuki Narita
old
O4
4
Giora Schmidt
new
N10
5
Solenne Paidessi
old
O1
6
Elmar Oliveira
new
N9
7
Marie-Annick Nicolas
new
New
York
8
Elizabeth Pitcairn
Strad
9
Giora Schmidt
new
Table S1: Information about the players who took part in each experiment
Information about the listeners
Listeners who turned in incomplete data-sheets were excluded from the analysis. The distribution
of the remaining (55 in Paris, 82 in New York) is described in Table S2:
makers
violinists
other
N/A
Mean age
Paris
15
19
21
0
N/A
New-York / Group 1
18
11
13
3
49.2 years
New-York / group 2
11
10
13
3
46.8 years
Table S2: Information about the listeners who participated in each experiment
Information about the orchestra
The Sorbonne Universités student orchestra (called COSU) was supplemented by several
professional players and conducted by Vincent Barthe. In all there were 10 violins, 5 violas, 3
cellos, 1 double bass, 2 flutes, 2 clarinets, 1 oboe, 1 bassoon, 2 horns, and timpani. To allow
players to read their parts on the darkened stage, small directional lamps were attached to the
music stands.
Acoustical measurements
The acoustical output (sound pressure per unit force at bridge) for each violin measured using an
impulse hammer and microphone. The bridge was tapped both horizontally and vertically, and
the sound pressure measured at 12 equatorial microphone positions 20 cm from the center of the
violin in the plane of the bridge [Curtin J (2009) Measuring violin sound using an impact
hammer. J Violin Soc Am: VSAPapers XXII:186–209]. This yielded 24 frequency response
17
functions (FRFs) for each violin. The real average for all 24 FRFs was calculated and then
reduced to a single average (in dB) for the frequency band spanning 200 – 6400 Hz.
Figure S1: Difference in averaged sound output level (per-unit-force at bridge) for each test
violin relative to violin O3, for the 200 – 6400 Hz frequency band (represented in dB).
The three new violins with highest levels are, in descending order, N5, N11, and N10, and the
three old are O6, O4, and O1.
Selection of the violins for the New York experiment
New violins were chosen from a pool of 15 violins (including N5) submitted by their makers. A
pre-selection took place in the same venue as the experiment. Author JC and the distinguished
violin maker Sam Zygmuntovicz played the violins under blind conditions and selected what they
considered the best eight. The selection of three new instruments was made by a blind test similar
to the pre-selection in Paris [8]: Three blindfolded soloists (the two participating in the
experiment, along with Karen Gyomo) each played the eight violins behind an acoustically
transparent screen for a small audience (JC, F-C T, SZ and the two soloists who were not
playing). Three violins (including N5) were chosen via an informal discussion of the player- and
listener-preferences. There was good agreement among all parties.
Score sheet used in Paris
PART 1 Name:
Email: Seat Number in Hall
Expertise: (violin maker, professional etc.)
In Parts 1 and 2, you will hear a series of comparisons between pairs of violins. The instruments
in each pair will be referred to as Violin A and Violin B. A violinist will play a short solo excerpt
on violin A, then violin B, then violin A again, then violin B again. Which violin projects better?
Please see below for an example of how to mark your answer.
18
The player will next play an excerpt with orchestra on that same pair of violins in the same
order. Which violin projects better now?
PLAYER 1
A Superior A=B B Superior
Example #1 |------------------------------|---X-------------------------| B slightly superior to A
Example #2 |-----X-----------------------|------------------------------| A very superior to B
A Superior A=B B Superior
Q1a. Pair 1, Solo: |------------------------------|------------------------------|
Q1b. Pair 1, Orchestra |------------------------------|------------------------------|
[…]
Q9a. Pair 9, Solo |------------------------------|------------------------------|
Q9b. Pair 9, Orchestra |------------------------------|------------------------------|
A Superior A=B B Superior
Part 1 Continued:
PLAYER 2
Which violin projects better?
[Same score sheet as for Player 1]
PART 2
The instructions are the same as for part 1. Which violin projects better?
[Same score sheet as for Part 1, with 8 pairs played solo and then with orchestra]
PART 3
Each player will now play about 2:30 minutes from a concerto on just one violin. Do you think
the violin is new or old? Why?
Player
New or old?
Why?
1
New Old
2
New Old
19
[…]
7
New Old
Questionnaire sent to the listeners after the Paris experiment
1. What is your definition of projection, i.e. the one that you used to evaluate the different
violins?
2. Were there one or more musicians for whose the evaluation was easier/harder? Were there
one or more excerpts for which the evaluation was easier/harder?
Score sheet used in New-York
Name: Age: Email:
Expertise: (violin maker, professional etc.) Seat and row number in Hall:
PART 1:
You will hear a series of comparisons between pairs of violins. The instruments in each pair will
be referred to as Violin A and Violin B. A first violinist will play a short excerpt on violin A, then
violin B, then violin A again, then violin B again. You then have to answer one question, before
the violinist plays the next pair. After 10 pairs, the violinist will change, as well as the question
(part 1B).
PART 1A: player 1
Question: Which violin do you prefer and why? Please see below for an example of how to
mark your answer.
A preferred A=B B preferred Why?
Example #1 |-----X-----------------------|------------------------------| Because A has this
I much prefer A than B and that while I don’t
like this and that in B.
Example #2 |--------------------------X--|------------------------------|
20
I slightly prefer A than B
Example #3 |----- ------------------------X------------------------------|
I like A as much as B
Example #4 |------------------------------|-----X-----------------------|
I prefer a bit more B than A
Example #5 |------------------------------|--------------------------X--|
I considerably prefer B than A
A preferred A=B B preferred Why?
Pair 1 (practice) |------------------------------|------------------------------|
Pair 2 |------------------------------|------------------------------|
[…]
Pair 10 |------------------------------|------------------------------|
PART 1B: player 2
Which violin projects better?
This question may mean slightly different things to different people, for example:
o Which violin has the best carrying power?
o Which sounds loudest?
o Which sounds most full?
o Which seems most “present” to your ears?
o Which seems to fill the hall best?
o Which comes through to your ears more clearly?
o Which comes through to your ears more easily?
Please check the one/s that best capture what projection means to you.
If you have some other definition, please write it here:
21
Please use the scale the same way as before by replacing “I prefer A more” by “A projects more”
as illustrated in the example below:
A projects best A=B B projects best
Example #1 |-----X-----------------------|------------------------------|
A projects much more than B
Pair 1 (practice) |------------------------------|------------------------------|
Pair 2 |------------------------------|------------------------------|
[…]
Pair 10 |------------------------------|------------------------------|
Statistical analysis for the old/new guesses (Paris part 3)
For each of the 7 violins, the numbers of correct guesses over the total number of guesses are
presented along with two intervals, the first one being the 95% Clopper-Pearson interval and the
second one the Bayesian Credible Interval with Jeffreys‟prior:
violin 1 (old): 7/45 [6.49%; 29.46%],[ 7; 28.13%]
violin 2 (new): 11/46 [12.59%; 38.77%], [13.42%; 37.57%]
violin 3 (N): 27/45 [44.33%; 74.30%], [45.43%; 73.33%]
violin 4 (N): 26/45 [42.15%; 72.34%], [43.24%; 71.35%]
violin 5 (O): 24/43 [39.88%; 70.92%], [41.00%; 69.88%]
violin 6 (N): 22/44 [34.56%; 65.44%], [35.62%; 64.38%]
violin 7 (O): 23/46 [34.90%; 65.10%], [35.92%; 64.08%]
What if the guesses were governed entirely by chance, as if the subjects were in fact tossing
coins? All guesses would then be independent one from each other – both from one subject to the
next, and among each subject‟s guesses. In this case, the number of correct guesses would be
distributed as a binomial distribution with parameter n being the total number of guesses (273),
and parameter π = 0.5 being the probability of a correct guess. Then the 95% Clopper-Pearson
confidence interval would be [38.7%; 50.8%]. It includes 50%, meaning that the hypothesis that
chance entirely governed the guesses cannot be ruled out. Nonetheless, given these data it could
be stated that Pr*(40.9% < true proportion < 48.6%) = 0.80, just as if the subjects were using a
biased coin with less than 50% chance of success! But that rests on the assumption of every
subject performing independent guesses which is far from plausible.
Explanations provided by the listeners in New York to justify their preferences
Only the explanations related to projection and loudness were taken into account here. The terms
that we considered associated with projection are the following:
22
project(ion), loud, quiet, power(ful), muted, big sound, small sound, direct sound, carrying
power, strong, weak, present, full. Clear was added to this list for the listeners who selected the
'"Which comes through to your ears more clearly" as a possible definition for projection.
