The Problem with Randomness

We ran the Vacancy Information Service for many years for TIAPEI. The service provided information about lodging vacancies on PEI by telephone or from a website. The telephone service, which was expertly crafted by Gary Clow (son of Bobby Clow, of Clow’s Red and White in Hampshire), was designed to provide callers with a random selection of five properties with vacancies in their area of interest.

The service was truly random: every time a new call came in, a brand new random selection was culled from the database and presented. Over the season of thousands of telephone calls this meant that every property with vacancies had an equal chance of having their information delivered to callers. We had statistics that showed this clearly.

The problem was that although this was true over time it wasn’t, of course, true for just one or two calls. Because every property had an equally random chance of being selected during a given call, it was quite possible that a specific property would show up in two successive calls. Or that a specific property wouldn’t show up in three or four successive calls.

If you were Bob’s Bed and Breakfast, and you called in twice and heard Jane’s Bed and Breakfast on both calls, you assumed that Jane was somehow getting a “special deal.”

And that’s the problem with randomness.

I had cause to think of this today when I read this Newsweek article that calls into question the randomness on the iPod shuffle. I think the problem is the same: if the iPod shuffle is truly random, then it’s likely that it will appear that certain songs are being preferred over others, even if, over the long term, the exposure of every song is statistically equal.

And if Apple’s experience is anything like ours, a lot of people will never believe that.


Ken's picture
Ken on February 26, 2005 - 21:12 Permalink

Extract each choice from the pool of unchosen choices until the pool is empty, then repeat.

Peter Rukavina's picture
Peter Rukavina on February 26, 2005 - 21:17 Permalink

Ken, that may work on an iPod shuffle, but for the Vacancy Information Service, the “pool” was constantly changing (as vacancies came and went), so it wasn’t possible to maintain accurate “state” information as you describe.

oliver's picture
oliver on February 26, 2005 - 23:53 Permalink

Huh? Did the system recommend places without knowing whether they had a vacancy? If the fact of a vacancy is known, then I’d have thought you had all the state information that needed supplying. The computer system on its own ought to know which places it last recommended and thus to know when it’s cycled through them. Then if the system followed ken’s recipe, the only randomness is in the order or sequence in which the fixed set of hotels get assigned new visitors. If duration of stay is has nothing to do with where the guest ends up, then the amount of paid occupancy each hotel gets from the system should be equal over time. Of course, that’s absolute “occupancy” in number of guests and not in terms of the fraction of a hotel rooms that get filled. If you didn’t add a weighting factor proportionate to the number of rooms a hotel has to fill, then arguably there’s some unfairness to the big hotels, but that unfairness is easy to eliminate with weighting. So what am I not understanding here?

oliver's picture
oliver on February 26, 2005 - 23:56 Permalink

Oh, I see, the set of hotels is not “fixed,” because sometimes some will have no vacancy (due to old family friends who arrived no thanks to the computer) and so be ineligible some cycles.

oliver's picture
oliver on February 27, 2005 - 00:00 Permalink

But if a hotel gets left out for one cycle due to full occupancy, why not just consider that “tough luck” (actually good luck) for them and follow ken’s recipe with whatever hotels among the total set are eligible at any given moment?

Ian Williams's picture
Ian Williams on March 1, 2005 - 05:08 Permalink

There was much discussion on the Apple boards that the larger iPods on “shuffle” would
a) play the songs you rated highly more often
b) play the songs you played a lot more often
Which, of course, is why I never “rated” any of my songs, and I doubt hardly anyone else does either.

But if it’s true that the iPod on “shuffle” would show a preference for songs you liked to play (via the “times played” number at far right on iTunes) then wouldn’t the iPod — if left to its own devices — start playing songs that IT selected over and over? Pretty soon, the iPod would just have a playlist of ten songs that happened to be constantly selected, even if there were thousands in your library.

Oh, I’ve lost my ability to make sense on this. Please continue.

Ken Williams's picture
Ken Williams on March 25, 2005 - 23:41 Permalink

I found the ultimate MP3 player:
MuVo N200

It has shuffle, shuffle once, shuffle folder.
It records using a built in mic, or in stereo using inline plug.
It comes in 256, 512, or 1Gb.
Mine (256Mb) cost $84.99.
It has an FM radio, which works perfectly.
It records from FM in stereo!
It has a display.
It runs 15 hours on a AAA battery.

I love it.

Ken Williams's picture
Ken Williams on March 26, 2005 - 16:29 Permalink

The N200 Manual explains it’s shuffle feature best.

I’ve been testing the fidelity of the MP3 encoder (44.1Khz Stereo 160kbps) which takes a line input. It is clean from my old CD player, however from my Toshiba Satelite Notebook it picks up hum (probably from the CPU clock etc).

I now need a portable two line mic preamp, the closest I can find with Google is the Azden Cam 3 but it is only a mixer — no preamp. Where’s my old Radio Shack mixer now, probably deep in some landfill. So it looks like I need to build one, which I think I can make just bigger than a 9volt battery.

I want to to field recording, so if anyone knows of a small mic preamp please tell me.