Cheers John! My only pertinent observation on your analysis is that Apple’s 75.6% market share is precisely what makes them achingly unattractive to contrary sorts like myself.

First, this is not an uncommon perception, and while some simply put it down to the human tendency to seek order/patterns even where there are none, I think it also comes from a misconception about what constitutes “random.”

The word “random” means, of course, that all outcomes are equally likely. A 6-sided die (or 20-sided, or crazy die) is random: any face (4, say) could come up on any given throw. Crucially, the same face could come up on a second throw — and even a third. Does that mean the dice manufacturer has a secret preference for 4s? No, just that random refers to the long run. By the same token, a track played twice (”Riders on the Storm”, in your example) in a day is perfectly plausible in a truly random system.

Apple’s iPods* use the word “shuffle” instead of random, and it’s important to contrast this term because it connotes a different method. The shuffle metaphor obviously invokes a deck of cards (emphasis added):

The [iTunes] shuffle algorithm chooses songs “without replacement.” In other words, much like going through a shuffled deck of cards, you will hear each song only once until you have heard them all… or until you have stopped the player or selected a different playlist.

That last portion adds another piece to the puzzle. Let’s say you switch your Zen to the shuffle equivalent, what Jim refers to as “random but played only once.” Might you still hear “Riders” twice before The Clientele? On an iPod, at least, the answer is ‘yes’ because the player will re-shuffle when turned off. This might be counter-intuitive; why not save the shuffled list once made, then play through to the end? One answer would be to pick up new tracks: without re-shuffling, you’d have to wait up to 150 hours** after new additions before your new track had a 1/3001 chance of being played. And therein lies the rub. Because your collection is a solid week’s worth of music, you’ll never leave it on and play it enough times to see if the algorithm truly is random.

So why not try a synthetic approximation? I took a 1 second sine wave and wrote a script to duplicate it to 100 different files, named test00…test99.mp3. Then I loaded them into an iTunes playlist (I couldn’t even find my iPod Shuffle, which tells you how much I listen to music), selected Party Shuffle, and played exactly 100 tracks.

Here’s the resulting histogram:

A continuous shuffle play of my “collection” resulted in exactly what you’d expect: each song played a single time.

Next I wanted to simulate the playlist-jumping/power cycling/short sessions that would be typical with casual use. Using Party Shuffle with the same 100 tracks, I played 10 sets of 10 tracks. Between each set, I switched the playlist from the test set and then back, to achieve a re-shuffle.

Result:

Still 100 plays, but obvious differences. Some songs never made it through, others made repeat appearances. As you might guess, that spike of 4 plays was not anything by Bob Dylan; it was the same 440Hz tone as the rest. (That copy happened to be test33.mp3.)

Now think of your collection. Which chart would it resemble? Have you set the “random/play only once” option? Do you always keep your Zen on, pause songs when you leave, and immediately resume? For at least 150 hours of listening? Of course not.

So, before you start wondering about the ghost in the machine, try this test. Download my 100 test tracks [711K] and add them to your Zen via a “test” playlist. Let your Zen play the shuffled playlist for as long as you can (overnight would be great, but I don’t know about battery life) then sync up and check your play count. After thousands of plays (achievable in just a few hours), a truly random shuffle would result in a distribution very close to equal amongst the whole set.

If you see different results, come back and we’ll get into entropy and random seeds.

Yours in Science,

J

* By the way, if you wanted to create “a hugely popular electronic medium for playing music” you’d be better off creating the iPod. Apple’s share of the (US?) MP3 player market is 75.6%; Creative’s, 4.3%.
** Assumes 3,000 songs at 3 minutes apiece.

My other theory is that Jesus did it. If all logical thought fails…then Jesus will prevail….

My suggestion is that say you have 500 tracks listed 1-500, the machine will randomly pick tracks. It just so happens that the ‘favourite tracks’ happen to be towards the middle of the listings.

I’m guessing it all about the mode, mean and median (some maths phrases there) I suggested that its more likely to pick, say track 235 rather than number 2, as randomly its more likely to be around there. Just a suggestion.

Also, if your not happy about hearing the same tracks all the time, on the Zen you can choose to have random but played only once. That way you’re guaranteed to hear every song!