Page 2 of 6 FirstFirst 123456 LastLast
Results 16 to 30 of 83

Thread: Composer Polls: some results

  1. #16
    Senior Member Pugg's Avatar
    Join Date
    Aug 2014
    Location
    The Netherlands
    Posts
    30,976
    Post Thanks / Like

    Default

    Quote Originally Posted by Jacred View Post
    I'm conservative then? Not surprised in the least...
    Nothing wrong with that, don't let anyone bully you in a corner.
    First they ignore you, then they laugh at you, then they fight you, then you win.
    "Mahatma Gandhi"

  2. Likes Bettina liked this post
  3. #17
    Senior Member dzc4627's Avatar
    Join Date
    Apr 2015
    Location
    California, U.S.
    Posts
    522
    Post Thanks / Like

    Default

    This looks good. Nice job! Excited to see more of this kind of stuff.

  4. #18
    Member
    Join Date
    Dec 2016
    Posts
    54
    Post Thanks / Like

    Default

    Quote Originally Posted by Nereffid View Post
    Though I'd like more and more people to vote in all my A la carte polls, I thought I'd start a thread discussing what might be gleaned from the results.

    The full list of polls, list of composers, and the "leaderboard" can all be seen at this page: https://sites.google.com/site/nereffid/
    I've had a look through your tables and I can't see what I thought you had agreed to do, which has caused the long delay since the last poll No 43 was set up, namely produce a matrix showing the voting pattern of each voter against each of the composers.

    Possibly I misunderstood what you said previously, or maybe I'm not looking in the right place? All I can see under the heading "Composers: Leaderboard" is a table showing the percentage achieved by each composer, ranked by size. This kind of table is very simple to produce as it involves no more than reading off the results from the various tables.

    It would be far more interesting to see these percentages supplemented by data on who voted for which composer. I'm pretty sure you said that you intended to produce this info. I recall "mmsbls" saying that only by doing the kind of poll you have set up, based on "likes", enables this kind of comparison. I pointed out that this is not correct as the same can most certainly be derived from conventional composer polls where people are asked to list their favourites.

    Anyway, on the limited presentation you have provided, for the top 50 composers I got the numbers below, showing rank, name, percent. I didn't bother going below rank 50 because the results at these depths are so unreliable that they're hardly worth reporting.

    1 Beethoven 93.5
    2 Bach J S 88.1
    3 Mahler 87.5
    4 Mozart 85.2
    5 Dvorak 84.5
    6 Sibelius 83.9
    7 Schubert 82.8
    8 Tchaikovsky 82.0
    9 Debussy 81.4
    10 Ravel 81.3
    11 Prokofiev 81.1
    12 Stravinsky 80.8
    13 Saint-Saens 80.6
    14 Rachmaninoff 80.2
    15 Strauss R 80.0
    16 Brahms 79.8
    17 Haydn, J 79.8
    18 Shostakovich 78.7
    19 Bartok 78.0
    20 Mendelsshon 77.7
    21 Grieg 77.5
    22 Chopin 76.6
    23 Schumann 76.0
    24 Berlioz 75.7
    25 Vaughan Williams 73.0
    26 Vivaldi 72.4
    27 Mussorgsky 72.1
    28 Handel 71.4
    29 Bruckner 70.9
    30 Wagner 70.2
    31 Liszt 70.2
    32 Faure 69.5
    33 Barber 68.9
    34 Elgar 68.6
    35 Rimsky-Korsakov 67.2
    36 Copland 66.3
    37 Scriabin 65.7
    38 Smetana 65.3
    39 Nielsen 65.2
    40 Holst 64.8
    41 Bizet 64.8
    42 Janacek 63.6
    43 Britten 62.6
    44 Ives 62.6
    45 Weber 60.9
    46 Borodin 59.7
    47 Schoenberg 56.7
    48 Gershwin 55.6
    49 Monteverdi 55.0
    50 Verdi 54.9
    ....

    I accept that many of these individual ranks are not significant in the usual statistical sense. Nevertheless, there are some strange looking results here, in comparison with those one normally sees from conventional poll standards based on lists of members' favourites, as opposed to your "likes" system.

    The main oddities seem to be:

    - ranked too low: Mozart, Brahms, Haydn, Schumann, Handel, Wagner.

    - ranked too high: Mahler, Dvorak, Sibelius, Ravel, Prokofiev, Saint-Saens, Rachmaninoff.

    The most weird result is probably Wagner at rank No 30. What on earth is going on here? Also, seeing Brahms and Haydn outside the top 10 is very peculiar. Mozart outside of the top 3 is also exceptional.

    On the basis of these oddities, I reckon your approach has failed on this occasion to produce a set of sensible results. Moreover, in the absence of the further data I referred to above - i.e. a cross tabulation of votes by each member against each composer - there is no redeeming feature that might have produced something of interest to look at and consider.
    Last edited by Martin D; Mar-29-2017 at 07:39.

  5. #19
    Senior Member
    Join Date
    Sep 2016
    Location
    California
    Posts
    2,363
    Post Thanks / Like

    Default

    Quote Originally Posted by Martin D View Post
    ...I accept that many of these individual ranks are not significant in the usual statistical sense. Neverthless, there are some strange looking results here. Those composers who seem to stand out significantly from the normal pattern of results from more conventional poll standards (i.e, when based on lists of members' favourites, as opposed to your "likes" system) are Mozart, Brahms, Haydn, Schumann, Handel, Wagner. Those who seem to be ranked too high are Mahler, Dvorak, Sibelius, Ravel, Prokofiev, Saint-Saens, Rachmaninoff...
    I don't think that there's necessarily a problem with Nereffid's methodology. It seems to me that the wording of Nereffid's polls might explain some of the discrepancies that you've observed. Unlike many TC polls, Nereffid asks about likes rather than favorites. This of course produces a different set of results from the other polls, because "like" is a milder word than "favorite." I personally have voted for many composers in Nereffid's polls whom I would never include in a list of my favorites, and I assume that other TC members have done the same.

  6. Likes Nereffid, TurnaboutVox, Art Rock and 2 others liked this post
  7. #20
    Member
    Join Date
    Dec 2016
    Posts
    54
    Post Thanks / Like

    Default

    Quote Originally Posted by Bettina View Post
    I don't think that there's necessarily a problem with Nereffid's methodology. It seems to me that the wording of Nereffid's polls might explain some of the discrepancies that you've observed. Unlike many TC polls, Nereffid asks about likes rather than favorites. This of course produces a different set of results from the other polls, because "like" is a milder word than "favorite." I personally have voted for many composers in Nereffid's polls whom I would never include in a list of my favorites, and I assume that other TC members have done the same.
    I am fully aware that Nereffid's polls are based on "likes", not favourites. I understand that he has chosen to use "likes" rather than request members to list their favourites because he felt that asking members to rank their preferences is often a diffcult thing for some people to do. In fact, all of his polls, not just composer polls, have been based on "likes".

    However, he has maintained throughout the 43 polls that his system based on "likes" should produce results for composer rankings that are broadly consistent with those one normally sees from conventional polls. Soon after the final poll No 43 was set up the results were not looking very good, with some odd looking results, e.g Strauss in the No 3 spot. He decided to extend the period of voting hoping to get some additional votes that might rectify such oddities. More votes were cast and some of the oddities were removed or reduced, but there are still some that stick out like sore thumbs, and are not likely to be changed by any further voting.

    Besides, a claimed benefit of this long-winded way of estimating composer preferences based on "likes" was that it would shed light on the various composers that members liked, i.e. who voted for which composer. This type of information can be obtained from conventional polls and there is no need to go through this circuitous and hazardous procedure, based on an ill-defined "like" system, in which there is no commonality of approach, with different people applying different standards. In any case, so far as I can see, this analysis hasn't been presented, unless it's buried somewhere I haven't spotted. That, I thought, was one of the main aims of this whole exercise, as originally conceived last Autumn.

    I gather that these polls attracted over 200 voters in total across all the composers. It means that there are many people who did not vote in all the polls, even to register their "not liking" or "not knowing" any of the candidates on offer. On that reckoning, I can't see why that higher number (200 or whatever) was not taken as the relevant denominator in each of the composer percentage, rather than the number who happened to vote in each poll. If the results are based only on the numbers who voted in each poll, since they under-record the total number of voters across all polls, it would seem that the procedure would give biased results.
    Last edited by Martin D; Mar-29-2017 at 08:16.

  8. #21
    Senior Member Woodduck's Avatar
    Join Date
    Mar 2014
    Posts
    8,777
    Post Thanks / Like

    Default

    Dvorak's ranking as the 5th most "liked" composer here seems to tell us that few people actually dislike Dvorak. Well, really, what's not to like, after all? Nice guy, lovely music, and that New World goin' home thing on the cor anglais... A lot easier to "like" than Love and Death in Bayreuth, yes?

    I always like Dvorak when I hear him. I don't always "like" Wagner, but - I revere Wagner, and consider him an earth-shaking genius who can rip my guts out and take me to unimagined and sometimes scary corners of my psyche. Not an experience I always "like"! But when I do want that experience, the likable Mr. Dvorak can just wait at the gate till I've come back from Valhalla.

    It's a bit like "liking" people's posts on the forum. It means something, but it seems best not to assume too much.

  9. Likes Nereffid, SiegendesLicht, Ingélou and 5 others liked this post
  10. #22
    Senior Member Nereffid's Avatar
    Join Date
    Feb 2013
    Location
    Co. Dublin, Ireland
    Posts
    4,522
    Post Thanks / Like
    Blog Entries
    7

    Default

    Quote Originally Posted by TwoFlutesOneTrumpet View Post
    Perhaps a little surprised that Mahler is more liked than Haydn or Brahms but not that much.
    Really, I'd like to be able to present all the results with the statistical confidence intervals attached so people don't put quite as much store in the ranking. If we're talking about a 99% confidence interval (which is very strict) then technically Mahler scored 79.7-95.3%, Haydn got 70.3-89.4%, and Brahms got 70.7%-90%. So yeah, Mahler gets higher on the actual percentage of votes but there's quite an overlap between them.
    See the results of my polls: most-liked composers and most-liked works.

  11. Likes Ingélou, TwoFlutesOneTrumpet liked this post
  12. #23
    Assistant Administrator
    Join Date
    Mar 2011
    Location
    California, USA
    Posts
    8,702
    Post Thanks / Like

    Default

    Quote Originally Posted by Martin D View Post
    On the basis of these oddities, I reckon your approach has failed on this occasion to produce a set of sensible results. Moreover, in the absence of the further data I referred to above - i.e. a cross tabulation of votes by each member against each composer - there is no redeeming feature that might have produced something of interest to look at and consider.
    Quote Originally Posted by Nereffid View Post
    I've also started a new Group http://www.talkclassical.com/group.php?groupid=120 for those who might want to discuss in greater detail than might be of interest to the general TC member.
    In the group post:

    At the bottom of that page is a .csv file containing all the poll data - every participant's vote in every poll - up to date as of March 23.
    The data are actually quite interesting. I don't see any oddities. I can't imagine why someone would think that actual stated preferences are somehow not sensible results.

  13. #24
    Senior Member Art Rock's Avatar
    Join Date
    Nov 2009
    Location
    Kampen (NL)
    Posts
    5,132
    Post Thanks / Like

    Default

    Quote Originally Posted by Nereffid View Post
    If we're talking about a 99% confidence interval (which is very strict)...
    In my experience with statistics a 95% confidence interval is usually applied, which would narrow down the potential spread considerably compared to 99%.

    And of course the "like" system will lead to some differences from the "favourite" system. Wagner is a clear example. In compilation of favourites he usually ends up top10, because the people who like him tend to love him and put him high in their list of favourites. But his music is certainly not universally liked. On the other hand we have composers like Saint-Saens who is rather easy to like, but would probably not show up as often in lists of say 10 most favourite composers. The general trend though is not that far off previous lists based on favourites.
    Und Morgen wird die Sonne wieder scheinen.....

  14. #25
    Senior Member KenOC's Avatar
    Join Date
    Mar 2011
    Location
    SoCal, USA
    Posts
    14,538
    Post Thanks / Like

    Default

    Actually 95% is the confidence level. The confidence interval is expressed otherwise. E.g.,

    We are 95% confident that the preference is 47%, plus or minus 3%...

    Both the confidence level and interval are required for a statistical result to be weighed against the likely characteristics of a population being sampled.
    Last edited by KenOC; Mar-29-2017 at 08:47.


  15. #26
    Senior Member Nereffid's Avatar
    Join Date
    Feb 2013
    Location
    Co. Dublin, Ireland
    Posts
    4,522
    Post Thanks / Like
    Blog Entries
    7

    Default

    Quote Originally Posted by Martin D View Post
    I accept that many of these individual ranks are not significant in the usual statistical sense. Nevertheless, there are some strange looking results here, in comparison with those one normally sees from conventional poll standards based on lists of members' favourites, as opposed to your "likes" system.

    The main oddities seem to be:

    - ranked too low: Mozart, Brahms, Haydn, Schumann, Handel, Wagner.

    - ranked too high: Mahler, Dvorak, Sibelius, Ravel, Prokofiev, Saint-Saens, Rachmaninoff.

    The most weird result is probably Wagner at rank No 30. What on earth is going on here? Also, seeing Brahms and Haydn outside the top 10 is very peculiar. Mozart outside of the top 3 is also exceptional.

    On the basis of these oddities, I reckon your approach has failed on this occasion to produce a set of sensible results.
    Where should Mozart, Brahms, Mahler, Dvorak et al be ranked? What should the % of people who like these composers be? Should that % be consistent across time and space, or should it just reflect the people who answered the question? Do you have a list somewhere that you're confident is a correct assessment of every composer's ranking, and if so, can you point me to it? If it's based on some sort of scoring system that would be really handy because we could correlate its results with my polls and see how they match up.

    As to what's going on with Wagner being at no.30. Let's just break it down. 121 people offered an opinion on Wagner. 85 said they liked him. The other 36 didn't. This translated into a percentage score of 70.25%. 29 other composers got a higher percentage score. Your issue shouldn't be with me but with the 36 people who didn't say they like Wagner.
    I wanted to know how many people like Wagner, not whether he's their favorite or a great composer or anything else. My answer is 70%, and that figure applies specifically to TC voters circa 2016-17.
    One response is to decry this finding as outrageous and wrong. Another is to wonder if it might reflect a broader reality: maybe a composer can be acknowledged as among the greatest, and a favourite of many, and yet still not be liked by a relatively large number of classical fans.
    See the results of my polls: most-liked composers and most-liked works.

  16. Likes Art Rock, Headphone Hermit, hpowders and 3 others liked this post
  17. #27
    Senior Member Nereffid's Avatar
    Join Date
    Feb 2013
    Location
    Co. Dublin, Ireland
    Posts
    4,522
    Post Thanks / Like
    Blog Entries
    7

    Default

    Quote Originally Posted by Art Rock View Post
    In my experience with statistics a 95% confidence interval is usually applied, which would narrow down the potential spread considerably compared to 99%.

    And of course the "like" system will lead to some differences from the "favourite" system. Wagner is a clear example. In compilation of favourites he usually ends up top10, because the people who like him tend to love him and put him high in their list of favourites. But his music is certainly not universally liked. On the other hand we have composers like Saint-Saens who is rather easy to like, but would probably not show up as often in lists of say 10 most favourite composers. The general trend though is not that far off previous lists based on favourites.
    Yeah, I know 95% is more usual. 99% is futile ***-covering.
    See the results of my polls: most-liked composers and most-liked works.

  18. Likes Art Rock, Pugg liked this post
  19. #28
    Member
    Join Date
    Dec 2016
    Posts
    54
    Post Thanks / Like

    Default

    Quote Originally Posted by Nereffid View Post
    Where should

    As to what's going on with Wagner being at no.30. Let's just break it down. 121 people offered an opinion on Wagner. 85 said they liked him. The other 36 didn't. This translated into a percentage score of 70.25%. 29 other composers got a higher percentage score. Your issue shouldn't be with me but with the 36 people who didn't say they like Wagner.
    I wanted to know how many people like Wagner, not whether he's their favorite or a great composer or anything else. My answer is 70%, and that figure applies specifically to TC voters circa 2016-17.
    One response is to decry this finding as outrageous and wrong. Another is to wonder if it might reflect a broader reality: maybe a composer can be acknowledged as among the greatest, and a favourite of many, and yet still not be liked by a relatively large number of classical fans.
    OK, let's take Wagner purely as an example. As you say, 121 members voted for him in the relevant poll (No 11), and 85 said they like him. That gave Wagner a percentage score of 70.25% (85/121), which placed him in position No 30 in your table.

    Leave aside for the moment the matter of statistical significance, you stated recently in another thread that there were 212 voters in total across all the 43 polls, and that only 72 voted in all 212. I fully accept that some of those 212 who didn't vote in poll No 11 may have left the forum, or were unable to do so for other valid reasons. But I suspect that some of the 212 who didn't vote in Poll 11 could have done so but chose not to do so either because they didn't like any of the composers on offer, or for some other reason that one can only speculate upon. If you disagree, how do you account for the missing 91 voters?

    This kind of thing (which affects many of the polls, not just Poll 11) suggests that the number of voters in Poll 11 is probably under-recorded, so Wagner's rating would otherwise be different. The same applies to other composers, in that they would all shift around depending on the precise denominator you use. One way around this problem is to base the calculations on the number of votes cast for each composer. If you do so, a different picture emerges: Wagner's position jumps to No 23.

    Admittedly, this is not a big jump, but if you apply the same procedure to all the top 50 composers a rather different set of ranks results. Brahms moves up from rank 16 to 7, and Haydn from 17 to 12. Of course, other ranks move around somewhat. Looking at the ranks from a numbers approach, rather than your percentages, seems to me to produce a rather better correlation with ranks you tend to get from typical favourite-based approaches. Here I'm thinking of some of the past composer polls carried out here at T-C.

    [BTW: may I ask who/what is "Jim Shine" as noted at the foot of the csv file?]
    Last edited by Martin D; Mar-29-2017 at 10:46.

  20. #29
    Member
    Join Date
    Dec 2016
    Posts
    54
    Post Thanks / Like

    Default

    Quote Originally Posted by KenOC View Post
    Actually 95% is the confidence level. The confidence interval is expressed otherwise. E.g.,

    We are 95% confident that the preference is 47%, plus or minus 3%...

    Both the confidence level and interval are required for a statistical result to be weighed against the likely characteristics of a population being sampled.
    Agreed that 95% is the usual confidence level.

    In the present context, rather than quoting confidence intervals for individual ranks it would also be useful to know what is the "least significant difference" between ranks, in order to be able to ascertain whether or not the rank for composer N is significantly different from composer N+1, etc, as one goes up/down the table.

    I would guess that if such data was given it would show even more clearly that the ranks are not reliable beyond maybe the top few positions only.

  21. #30
    Member
    Join Date
    Dec 2016
    Posts
    54
    Post Thanks / Like

    Default

    Quote Originally Posted by mmsbls View Post
    In the group post:

    The data are actually quite interesting. I don't see any oddities. I can't imagine why someone would think that actual stated preferences are somehow not sensible results.
    I guess that since you were partly instrumental in encouraging the setting up these 43 polls last Autumn you may feel more inclined to support the method and results. I don't feel so constrained.

    As I have said before, the trouble with all of these "like" based polls, not just the recent 43 composer polls, but the dozens of previous ones as well, is that they're all based on selecting an arbitrary number of items from a list of candidates deemed to be suitable by the OP, and the sole criterion is whether or not the item is "liked" or not.

    This procedure seems to me to be a very poor way of going about sampling voters' opinions, as "like" is such a vague and elastic term. The problem is exacerbated by the fact that no definition of "like" is offered to assist voters in making make consistent selections, either between different voters' selections or across each voter's selections, with the aim of achieving a broadly consistent set of results across the entire set of polls.

    For example, in the composer polls now under discussion it might have been suggested by the OP that a "like" may be given if the voter reckons that he/she has heard sufficient of a composer's work to form an opinion, and if so whether they would rate that composer at least somewhere in their list of favourites, or pay money to acquire more of their material/attend a concert etc.

    As things are, a ragbag of responses has been attracted with all manner of different interpretations placed on the items voted for. It was apparently hoped that such variability will average out in the "wash," but I'm not sure this is justified, as the fundamental question being asked seems far too sloppy and likely to generate optimistic results for some of the composers who would normally be less favoured.

    It has also been necessary to deal with the fact that it has taken 43 separate polls, and over 7 months to complete, to get through a very long list of 602 composers, many of whom most people have evidently never even heard or rate very low. This has involved making arbitrary judgements about which group of composers to include in each poll, and at the end of the long-winded process how to aggregate the results given the sizeably different response rates across all 43 polls. On this latter point, I don't believe this has been done properly, but rather fudged much as I guessed would happen, and probably introducing yet more bias.

    This is hardly an efficient way of conducting polls that are supposed to form a consistent whole. Perhaps it would matter less if it was a one-off duff poll, but there appears to be an "industry" behind this latest bunch of polls, using this place as some kind of test-bed. It might have been better if just one poll had been taken that simply asked people to identify which of the 602 composers (if that's how many it's worth including, although it seems excessive) they happen to "like", and also given the members a clue of what is meant by "liked".
    Last edited by Martin D; Mar-29-2017 at 13:22.

Page 2 of 6 FirstFirst 123456 LastLast

Similar Threads

  1. Why so many polls?
    By Celloman in forum Classical Music Discussion
    Replies: 28
    Last Post: Mar-06-2015, 04:49
  2. Trinity College of Music London Audition results
    By andreea in forum Keyboard Instruments
    Replies: 4
    Last Post: Dec-19-2012, 00:40
  3. Top 100 Symphonies - predict the results
    By GraemeG in forum Orchestral Music
    Replies: 66
    Last Post: Sep-28-2009, 03:03

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •