Google I/O 2021 Eurovision 2021 Google I/O swag China's Mars rover Apple iMac M1 review Child tax credit calculator

iTunes randomiser: methodology explained

Author David Braue responds to reader feedback on his testing of iTunes' random "shuffling" feature.

Commentary What a firestorm! From the reader feedback, my article iTunes: Just how random is random? clearly touched a lot of nerves. If I may respond to a few points raised:

  1. I am not, and never claimed to be, a professional statistician. I did, in separate and unpublished work, attempt to apply my limited statistical learning to the numbers but failed to reach any useful conclusions either way -- so these numbers were excluded. For the record, a 2-tailed t-test comparing the relative frequency with which songs from certain labels were played, did show a significant difference (p=0.03). However, I agree that a full statistical analysis would be more than welcome -- anybody with the knowledge to do this who is willing to take it on?
  2. One person questioned the sample size. The test involved a pool of 200 songs and creation of 40 different playlists, offering a total of 1300 song slots just waiting to be filled. Someone more statistically minded than me may be able to discredit this with some hard analysis, but on the surface it seemed like a pretty good sample size. Happy to do larger and more playlists in a follow-up analysis, if I could have some guidance on how many times it should be done.
  3. I would not agree that the statistics show a bell curve; if I were Sony, for example, I would love to know why my artists' songs made up 34.2 percent of the iTunes Music Library but were only represented on 18.8 percent of the 40 playlists. Conversely, EMI would rightly be celebrating, since its songs only comprised 16 percent of the library but were chosen for 23.6 percent of the playlists. Maybe this is all randomness rearing its strange head, but it's certainly interesting to note.
  4. One person asked why the record companies would want people to listen to music they already own, more than other music? Well, why does Coca-Cola advertise itself when most people already drink Coke? Why does Channel 7 advertise its coming shows to people that are already watching Channel 7?

    The answer is simple: to prevent them from switching to other brands, shows, or musicians -- and encourage them to spend more on related brands, shows, or musicians. This is a core tenet of the up-sell: if a person buys one song from an artist, they've indicated a preference -- and any record label, which is at its core a marketing organisation, would love to have that person also buy other songs, or a full album, from the same artist. If I were a marketer, I would love for iTunes to give me a way to increase the frequency with which the songs are heard. And if I were Apple, I would resist doing this with all my might, so as to not alienate my customers.

  5. Whether the labels are actually doing this is another question, and one which merits separate exploration. I would tend to believe Apple wouldn't risk doing something like this, and would certainly agree that there may be other explanations. This article was not meant to be pointing a finger at anybody in particular -- just trying to put some numbers around a phenomenon that many of us have observed and wondered about.
  6. The issue of when the song was added was mentioned -- but the songs were all added to the library within two days of each other, so could this really make such a big difference? Also, as to JackAttack's suggestion, the order of adding songs was: grouped iTunes songs first, then individual songs, then grouped songs from CD. Yet these songs from CD were generally played less frequently than the ones from iTunes.

    This is why I started with a clean system -- so the system would have no preconceived notion about my music preferences. When the playlists were generated, none of the songs had been played, not even once.

  7. One person said Apple's randomness generator goes through the playlist and goes until all songs have been played, then repeats. Could you please explain why, then, four of my songs were never chosen at all, out of a possible 650 playlist positions, and that these were all songs previously ripped from CD and not purchased from iTunes? The four songs represented 2 percent of the test music collection; all things being equal, you would expect to have seen each of these songs chosen an average 3.25 times -- equalling 13 times in total. But none of them was chosen, not even once.
  8. Forget commercial or whatever other motivations; as a music listener who likes variety, I would be interested to know why there are five songs by Lionel Richie and many other artists in the playlist -- but Lionel Richie songs were played an average of 11.8 times through this experiment while Christina Aguilera and Oasis songs were only played twice each, on average. I'd also love to know what it is about the Akon featuring Eminem song "Smack That" that made iTunes choose it 17 times.
  9. Another person mentioned that it would be even more valuable to try this project with normal MP3 versions of the songs as well. I considered running the same tests using Windows Media Player under Windows, but time and money were both limiting factors. If anybody has a spare AU$200 or so to cover the cost of buying the songs in MP3 format, I'd be happy to repeat the experiment and see if we can't arrive at some firmer conclusions.