Predicting Pandemics

This Discussion is now a Draft.

Once it's ready, please submit it for review by our team of Community Moderators. Thank you!


This Discussion now needs to be approved by community moderators.

Question closing, resolving (obsolete)

There's been a ton of discussion about question closing times, retroactive closing, etc., and we really need an authoritative policy that can be part of the FAQ and consistently adopted. Here's a first draft of that, for feedback.

How should the closing time be determined, and what is "retroactive closing"?

Brief guide:

  • If the outcome of question will very likely or assuredly be determined at a fixed known time, then the closing time should be immediately before this time, and the resolution time just after that. (Example: a contest between known competitors)

  • If the outcome of a question will be determined by some process that will occur at an unknown time, but the outcome is likely to be independent of this time, then it should be specified that the question retroactively closes some appropriate time before the process begins. (Example: success of a rocket launch occurring at an unknown time.)

  • If the outcome of a question depends on a discrete event that may or may not happen, the close time should be specified of order 1/2-2/3 of the way between the open and resolution time, and if the event occurs before the close time, the question should be resolved just prior to the event with no retroactive closure. (Example: successful accomplishment of some technical challenge or publication of some type of paper.)

  • If the outcome of a question depends upon a steady stream of events so that the outcome becomes steadily more clear with time, the question's close time may be chosen relatively freely; given the present scoring system a time of order 2/3 or more of the time between opening and resolution time is generally appropriate. (Example: cases of an illness confirmed by a given date.)

Note: points are steadily awarded between question opening and question closing, with an extra 50% apportioned at the time of question closing. Thus if resolution occurs before closing (as in the third case), less overall points are awarded. This is to prevent gaming of the point system.

In more detail:

There is no universal rule for determining a question closing time. Ideally, most questions should close relatively soon before there is definitive "reality input" that resolves or effectively resolves the question. For an election or fixed-time contest, for example, the close time should be immediately prior to the results being determined and made public. (If the closing time is later, people will gain some "free" points from the "absolute" part of the score function; earlier and people will feel frustrated they cannot predict, and salient predictive information will be lost.) The resolution time is generally later than the closing time, in which case its precise time has no bearing upon the points awarded.

However in many cases it is unknown at the time of question launch when the major "reality input" will occur. In this case there are a few salient guidelines:

  1. If "reality input" occurs before a question closes, it should generally be resolved just prior to the "reality input" and therefore before its assigned close time. In this case point are only awarded for the time during which the question was open but not resolved; this truncation is necessary to avoid "gaming" the point system. Since there is an extra set of points assigned to the time of question closure, this can mean that the total points awarded is relatively low for such questions. If there are details regarding the timing or condition of early resolution worth making clear these should be included in the question. For example, a question of "will N events of type X happen by Y" can become more and more close to resolved as date Y approaches. If this may occur then the question could specify retroactive resolution some duration prior to crossing the threshold.

  2. If a question may have "reality input" prior to its close time but the probability of positive resolution is independent of the time then it is appropriate to have the question retroactively close prior to the reality input, then resolve immediately after closing. This then accords full points to the question. This would be appropriate, for example, when predicting the success of a rocket launch that is assumed to be happening, but at an unknown time. Retroactive closure does not happen by default, and its condition and timing should be specified in the question. The degree to which retroactive closing is encouraged or frowned upon depends on the degree to which resolution is independent of the time at which it occurs. It should not, for example, be used when the prediction is of whether an event X occurs when it is expected that X has some roughly time-independent fixed frequency of occurrence, because the probability of positive resolution steadily declines with time. Rather, a good rule of thumb in this case would be to put the closing time ~2/3 way through the time between opening and nominal resolution.

  3. When details are not specified in a question the moderators will have to adjudicate early resolution or (less likely) early closure. When details are specified, these should generally be followed unless circumstances are exceptional.