Sunday, June 21, 2015

The Big 5 Orchestras: a statistician's quest

Which tango orchestras are the most important?

Nobody would disagree that the recordings of Carlos Di Sarli and Juan D'Arienzo, the musical antipodes of their glory days, are the backbone of the XXI century milonga playlists. (Interestingly, there is a persistent but unproven rumor that the two grandest orchestra leaders played together for a short period of time in 1934, when Di Sarli bounced between a succession of temporary jobs, with Cambareri in Rosario, with his former Sextet musicians at confitería "Novelty", with Juan Canaro, at Radio El Mundo, and maybe even with D'Arienzo?)

Ron Weigel with his wife Susana and the statue immortalizing
them in Urbana IL, where they have taught Argentine tango
since 1999. The 11-foot sculpture by Larry Young has been
installed in Urbana's Wandell Sculpture Garden in 2001
But the tango music aficionados always prefer to speak of "the Big 4" or "the Big 5", and they make a lot of fuss about who the "next 2 or 3" orchestras might be, once we place Di Sarli and D'Arienzo at the top of the list. "How did you dare to rank Troilo below his due 3rd position??" - "How could you even consider Donato among the big-five??".  In this post, we shall consider the orchestra rankings and preferences through the prism of surveys and statistics.

Enter Ron Weigel, likely the only statistician ever to have a tango monument erected in his honor. Dr. Weigel teaches biostatistics in Urbana-Champaign. During one of his visits to BsAs, in summer 2008, he decided to tally all tandas at all the milongas they attended, trying to get an authoritative answer about the Most Important Orchestras. Ron notes that the surveyed milongas were all in the downtown, all popular with the experienced milongueros, so the results ought to be biased in favor of the subjective tastes there (and indeed, Biagi ranked high in all genre categories, which is exactly what people remember about the Central BsAs milongueros). Still, Ron writes, his subjective experience in other BsAs venues is that they play broadly similar selections. We shall review the data shortly, but first let me set the mood by playing a video from Club Gricel, the home of 2 out of the 14 surveyed milongas:

An average milonga from Ron Weigel's dataset had 13 tango tandas, including approximaely two each of D'Arienzo and Di Sarli, and one of Troilo. None of the other orchestras ranked as a must-play:

As a fellow statistician, I must warn against literal acceptance of these rankings. There clearly is a good deal of variation in milonga setlists, and if one keep observing milongas, then this variability must result in slightly fluctuating tallies each time. In fact we can use the toolkit of statistics to check if the differences between, say, 11 Puglieses vs. 13 Tanturi in this dataset are "statistically significant" (as opposed to falling within the expected range of the random fluctuation).

One-tailed Fisher's test performed on Ron's tallies confirms that Troilo was not significantly preferred over Biagi or Tanturi ( p-value = 0.423086) or Pugliese or D'Agostino ( p-value = 0.273212) or Calo or even Rodriguez ( p-value = 0.147721). Only Di Sarli and D'Arienzo were significantly favored over the runners-up.

So after all the data-collection and analysis, we are left exactly where we started: the study confirmed that Di Sarli and D'Arienzo are The Big Two, but couldn't tell with statistical confidence who the "other 2 or 3" core orchestras ought to be. It doesn't mean that there is no objective orchestra ranking - it just means that it will take many more observations before one can confidently rank them.

And even then we'd be left with selection biases to ponder. Like, are salon milongas different? Are differences cropping up from year to year, both because the availability of the recordings changes and because the fashions and trends shift? Could there be differences between earlier-evening and late-night milonga styles? Between the more youthful and more old-dancer communities? Even between the steamy BsAs summer and its gloomy winter?

There turns out to be a much bigger survey which attempts to parse out some of these influences. But before we move on to it, let me mention some of the other results of Ron Weigel's study.

1. Milongas. There were, on average, 2.6 milonga tandas per event (fewer then expected 3.2 tandas if TTVTTM tanda flow was the rule). Almost all milonga tandas were of Canaro, D'Arienzo, Di Sarli, and Donato, (plus a sole Biagi tanda and a sole Tanturi tanda), with 25% milonga tandas being mixed ( which also included Caceres (Tango negro), Calo, Color Tango (La luciernaga), Laurenz, Tanturi, Troilo, and Villasboas). The numbers are too small for meaningful statistical comparisons, but mixed milonga tandas seem to be a widely accepted norm

2. Valses. There were, on average, 2.7 vals tandas per milonga, again fewer than expected under the TTVTTM rule. Fully one third vals tandas were D'Arienzo's (!), and nearly as many tandas were mixed (including Biagi, Calo, De Angelis, Demare, Di Sarli, Firpo (tipica), Laurenz, Tanturi, and Troilo). Biagi and Troilo were distant 2nd and 3rd behind D'Arienzo's, and all other orchestras were represented by just a tanda or two ( De Angelis, Calo, Quinteto Pirincho, D’agostino, Donato, and Tanturi).
Felipe Martinez DJing in Canada

3. Other genres. An average milonga had 2 to 4 tandas of non-tango music ( Chacarera, Cumbia, Merengue, Jazz, Pasodoble, Rock & Roll, Salsa), which may partly explain why there were fewer than expected milonga and vals sets.

While we were discussing Ron Weigel's survey (and its limitations) on facebook, Felipe Martinez pointed my attention to a much bigger annual Tango Tecnia survey which tried to measure not what DJs played, but what the dancers liked. The 2014 report is the latest one available. I took the 2015 survey to familiarize myself with its methodology, and I have to assume that it didn't change much year to year.

Tango Tecnia doesn't probe the opinions of the English-speaking tangosphere well (its North American respondents are overwhelmingly from Mexico, and its European respondents are mostly from Spain, although France and Italy contribute too), and the majority of the survey-takers are young (in their 20s and 30s) .... but it still cut impressively across the cultural and age divides with nearly 1300 responses. D'Arienzo and Di Sarli came on top here, too, with nearly 80% "approval rating":

It may be impossible to evaluate statistical significance from the 2014 survey results, because almost 30% of the respondents skipped the question about orchestras ... and it's impossible to tell if this fraction differed from Europe to the Americas. Those who did like some orchestras liked, on average, 10 of them .... but many must have picked just one or two, otherwise how could one explain the observation that nearly 1 in 5 survey-takers didn't like D'Arienzo or Di Sarli

xkcd: "Significant!" :)
Once one splits already-thin data multiple ways,
then all sorts of improbable spurious
"associations" can be "found" there
("More study recommended", deadpans
xkcd's famous mouseover)
Biases of selection and recall ought to be a huge problem, too - people may not remember unfamiliar or rare titles, and may remember better the music they heard from CDs or online videos than the tandas they loved at the actual milongas. The contemporary and the undanceable definitely gets a very strong favorable bias in the survey data, with Color Tango outpolling Laurenz and OTV, Piazzolla beating Lomuto and Malerba, and even Ojos de Tango getting 10 times the votes of Garcia, Firpo, or Carabelli. 

Popularity breakdown by region and by age looks intuitively right, although it may be impossible to tell apart significant differences from the flukes. Say, Sexteto Milonguero rules in South America and with the under-30 crowd, while Enrique Rodriguez and OTV seem to have more fans in Europe.

Of the specific record titles, the highest ranking is (yes) Poema with 18% approval rating. Many Pugliese records, Bajofondo and Otros Aires, Esteban Morgado, Caceres and Salgan complete the top-20 list of the most poplar record titles, Regional and age differences look mind-boggling sometimes - like, apparently Otros's "Lo Vino" is especially loved ... in Mexico, and Hotel Victoria is preferred by the 60+ age group? - but one can't tell if it is a mere fluke. 

Sometimes it's just so disappointing to look at the world through the statistician's eyes :) :) ... a picture which sort of made good intuitive sense no longer looks trustworthy once you go into the gory details, and start seeing meaningless coincidences and confounds where you used to see patterns.

UPDATE: analysis of SuperSabino's DJ survey (October 2015)

No comments:

Post a Comment