I am very interested in a straightforward definition of the concept "positive regression dependency on each one of a subset" (with straightforward example a plus) as described by Benjamini and Yekutieli. Harvey Motulsky seems to share my interest: The meaning of "positive dependency" as a condition to use the usual method for FDR control. There has even been a bounty placed and awarded on this question.

However, none of the answers actually clarifies this this concept in plain language (although one answer with a single upvote does simply quote from Benjamini and Hochberg). The most voted on answer—which won a bounty for the question—does not actually answer the question… a point made abundantly clear in the comments following the answer, and also by the answer's lack of acceptance by the OP.

I have a feeling this kind of situation may not be unique.

Is there a way I can offer a bounty to encourage a solid answer, without it simply being awarded to a much up-voted yet poor answer?


Benjamini, Y. and Yekutieli, D. (2001). The control of the false discovery rate in multiple testing under dependency. Annals of Statistics, 29(4):1165–1188.

  • 3
    $\begingroup$ I'm willing to place a bounty on that Q of any size you like (I have rep to spare) - maximum is 500 points; there can also be a minimum related to previous bounties - and to award or not award it where you see fit, but I can't really avoid the possibility of an auto-award either. [Well, I suppose strictly speaking there is a way - I believe a bounty can be cancelled by the community moderators and the reputation refunded - but it would be a clear abuse of the system to try to do that for this situation.] $\endgroup$
    – Glen_b
    Mar 21, 2017 at 23:49
  • 3
    $\begingroup$ @Glen_b I just placed another 200 point bounty on that question to reopen it and hope to get a clear answer. $\endgroup$ Mar 22, 2017 at 2:09

1 Answer 1


The basics are as follows:

  1. If you place the bounty you can award it as you see fit.

  2. If nobody posts an acceptable answer and so you don't choose to award the bounty, typically the bounty simply lapses unawarded (the reputation already having been lost).

  3. Under particular circumstances, an unawarded bounty may be - fully or partly - awarded automatically. Detailed information here: How does the bounty system work? (search for "What is automatic awarding"):

    If the asker placed the bounty and accepted an answer that was posted during the bounty period but doesn't award the bounty, that accepted answer is automatically awarded the [full] bounty.

    Failing that, half the bounty is awarded to the highest-scored answer out of those which were posted after the bounty was started, have a score of at least +2 and were not written by the bounty starter.

    If none of these conditions apply, the bounty is not awarded.

You can't stop the auto-award, unfortunately, so it's a risk that you take that the bounty may be automatically awarded to an answer that may have been interesting enough to be upvoted but doesn't fully answer the question in the desired fashion. About the best you can do is call for something like a canonical answer in the bounty, downvote posts that really don't do what you seek and encourage others to do likewise -- but you can't make people vote the way you want, so a new answer could still end up with +2 net upvotes and get half the bounty anyway.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .