“Does Your IPod Really Play Favorites?”, 2009-08 (; backlinks; similar):
[cf. Poisson clumping, birthday paradox, “Methods for Studying Coincidences”, Baader-Meinhof effect] Since the introduction of the first iPod portable music player (MP3 player) by Apple, Inc. users have questioned the randomness of the shuffle feature. Most evidence cited by users claiming to show nonrandom behavior in the shuffle feature is anecdotal in nature and not based on any systematic analysis of its randomness. This article reports on our attempt to investigate the shuffle feature on the iPod and to test its randomness through the use of probability and statistical modeling.
We begin by reviewing the research on people’s inability to perceive and understand both random and nonrandom behavior. Probability models are then developed, under the assumption of a random shuffle, for several of the most common types of events cited as evidence of a nonrandom shuffle.
Under this null hypothesis of a random shuffle, several goodness-of-fit tests of one of the probability models are conducted using data collected from real iPods.
No evidence to support user claims of a nonrandom shuffle was found.
Finally, we conclude with some reflections on and ideas for incorporating these examples into undergraduate probability and statistics courses.
…5. Classroom Uses: Several of these examples have been used in the undergraduate probability and statistics courses at Iowa State University.
These examples were well received by students, and several of them stayed after class or visited office hours to discuss ideas related to testing the shuffle feature or their personal impressions of the randomness of the shuffle. After teaching these courses for many years, the first author can state without reservation that no other examples or material has elicited this kind of response from students in these courses
[This is an example of why standard white-noise-style randomness is often the wrong kind of randomness; users don’t want “a random song” when shuffle goes to the next one, they want a song which is not like the previous one. (Similarly, in game design, i.i.d. sampling-with-replacement is unfun and feels deeply unfair, so game designers often quietly implement some sort of biasing procedure, like the magic of sampling-without-replacement.)
This is entirely different, and raises interesting questions of design—if it should be something like pink noise or low-discrepancy sequences, which one? Should you try to “unsort” lists to maximize the difference/distance between each adjacent entry (as opposed to the usual sort which minimizes the difference)? And eventually, Apple introduced “Smart Shuffle” to eliminate the complaints.]
View PDF: