Hacker News Reader

Comment by mgrund | Hacker News Reader

It’s a good lesson in the importance of picking the right things to measure/optimize for. Queue length is not a good thing to optimize for by itself.

View on HN

I don't even see how it does that. Unless there is some factor that prevents you from leaving and immediately re-entering the line at the front.

I feel like I missed something.

Yes! I feel this is the main point that is totally missed. The first time I read Tim Hartford years ago he was talking about this idea uncritically and I thought it was stupid then and I don't have much respect for him.

The thing we are trying to optimise is not length of que, but how long and how reliable it is to get served.

Under the old system you may have had to wait 10 minutes to get served BUT you knew before you started queuing it would be about 10 minutes (and thus could make an informed choice) and you can clearly see your progress in the queue and if you are patient you will be served.

Under the new system it now may take you 1 minute to get served or 1 hour. Or you will try multiple times and never get served. And you will have no indication before you start queing how successful you will be.

Any UX designer would easily tell you which of these systems is going to piss people off more.

Exactly, I don't get how that would increase throughput.

It's not about increasing throughout, it's about increasing user satisfaction. A system where 60 customers are served an hour, and none of them has queued, is theoretically better than one where the same amount are served, but each has queued for 10 boring minutes. You achieve this by disincentivising queueing.

So people would only go to a stand if there's nobody being served, because being behind the person currently being served is discouraged. Essentially, you'd just move the queue: people would wait for a "spot" to open up (a stand with no customer) and then go there.

If it's orderly, they kind of queue in the middle. If it's not, you'll have a large brawl as people try to get in front of each other.

I think you misunderstood the proposed queueing system.... The stand will never be 'open'.

People who arrive join the front of the queue, and are either served next, or someone else joins in front of them, pushing them back. When they're position 2 in the queue, they will only be served if 2 people are served and 0 arrive (low probably). If another person arrives, they're now position 3 in the queue, so 3 need to be served before any arrivals (even lower probably).

At some point, they decide the probability of being served drops low enough that they give up and leave - this will usually happen quickly.

The only way to 'game' this system is to look up the street and join the queue when it looks like nobody else is coming. If many people try to do this, then those people will be waiting forever (both want to join last - stalemate), and anyone who doesn't employ the tactic gets served.

In practice, you would just replace the queue with a disorganized rabble of people standing around, where as soon as the person currently being served leaves, there is a mad dash at the one secure spot, which is at the very front.

Those who really want your product now have to mob up outside your stall and hope they're the next lucky random fucker. They're still waiting.

Those who don't really want your product don't wait in a line for it anyways - the length of the line itself is a natural filter for availability of the good.

Basically - has satisfaction actually gone up if those who would be extremely satisfied by your product no longer receive it, and random folks who may or may not want your product do? I'd bet money the answer is a solid "NO".

Another way to look at it: with the queuing system, you spend 10 minutes, and you get what you want. How many minutes will it take with the no-queue system? It’s not like the thoroughput has increased. And how likely will it be for you to get what you want after that time? Is it definite?

Reply on news.ycombinator.com