3
$\begingroup$

Something I've noticed occasionally is that posters will include the phrase "One (book/solver/package/library/etc.) per answer please". Questions asking for a list of resources tend to include this phrase. For example, see Where can one obtain good data sets/test problems for testing algorithms/routines? and What are the best Python packages/interfaces to sparse direct solvers?.

How should we treat this phrase? Should we allow it? Should we use it sparingly? Should we blacklist it? Should every question using the phrase be wikified (assuring that it will be used sparingly)? Should all of the answers be wikified instead?

Points for discussion:

  • The phrase prevents answers with both good and bad recommendations (see my answer here recommending C++ libraries for matrices, which has been deservedly criticized in the past for mingling both good and bad recommendations)
  • The phrase encourages answer bloat. If there are a lot of potential recommendations, and each recommendation gets a separate answer, there could be a lot of answers. If the question gets more than 15 answers, according to Meta Stack Overflow, the question and answer will be wikified.
  • It's not clear if votes are on the user's answer, or the package recommended. A user could be a subject expert and potentially offer many recommendations, but if they're limited to one per answer, they need to post multiple answers to demonstrate their expertise; one user posting multiple answers is typically discouraged. Votes are supposed to reflect the quality of a user's answer, not the quality of the thing being recommended. Of course, if a user recommends something bad, and only something bad, the votes should reflect a bad answer.
$\endgroup$
  • $\begingroup$ It seems strange to me that a poster would specify only one suggestion per answer to begin with... If it is a list answer to begin with, why limit it? I just don't understand the asker's motivation... $\endgroup$ – Paul Apr 14 '12 at 21:21
  • $\begingroup$ Is it so bad that a question become community wiki-fied? Are there any disadvantages to this? $\endgroup$ – Paul Apr 14 '12 at 21:34
  • $\begingroup$ @Paul: Yes. Question wikification is allowed, but discouraged. I remember Grace Note saying something like "at most, 1 in 100 questions should be community wiki". Community wiki tends to work best with "big list of resources" questions, but at the same time, Stack Exchange tends not to work well with "big list of resources" questions, even though they would be incredibly useful and popular here. And we've had a big argument about lists on Meta.SciComp that I really never want to get sucked into again... $\endgroup$ – Geoff Oxberry Apr 14 '12 at 21:50
  • 2
    $\begingroup$ "Votes are supposed to reflect user expertise, not the quality of the thing being recommended." What? I thought votes were supposed to rate the quality of the answer. I'm not thrilled with a hard limit of one per answer because many good answers indicate the relationship between list items (i.e. they provide edges in the knowledge graph rather than new vertices). $\endgroup$ – Jed Brown Apr 14 '12 at 22:42
  • $\begingroup$ @JedBrown: You're right; I meant answer instead of expertise. I'll change that. $\endgroup$ – Geoff Oxberry Apr 14 '12 at 22:58
  • 1
    $\begingroup$ Okay, continuing the discussion. What is the value of answers that recommend "bad" things? Is the only reason for having them at all to be complete or politically correct? There are lots of lists on the internet, a significant part of the value of sites like this is to cut out the noise rather than to tabulate the noise. $\endgroup$ – Jed Brown Apr 14 '12 at 23:17
  • $\begingroup$ @JedBrown: As a general rule, heavily down voted answers are deleted. "Bad" answers may recommend something held in low-esteem by experts. It may be well-intentioned ignorance by a non-expert with some experience that leads to such answers. The restriction of one recommendation per answer makes it easier to filter, but harder to make associations. $\endgroup$ – Geoff Oxberry Apr 14 '12 at 23:43
3
$\begingroup$

If a user elicits a list-like response, responses shouldn't be limited to one recommendation per answer. As Geoff suggests, such requests places unnecessary restrictions on the responder. It is much more natural to put all recommendations in a single response, and I think a "one recommendation per answer" approach discourages people from posting more answers.

$\endgroup$
2
$\begingroup$

I think that both the question and answers should be made community wiki at the minimum. I don't like the "one (blank) per answer" requirement the question's author imposes. Personally, I like the answers to make contributions however the answerer wants, and the list to be synthesized into the original question. See, for example, the The Definitive C++ Book Guide and List.

I think the list questions are good as long as they are broadly useful to the community.

$\endgroup$
  • $\begingroup$ I just dropped a community wiki bomb on the post under discussion. I would like to resurrect Greg Wilson's question on the necessary skills for a computational scientist if this one works out. $\endgroup$ – Aron Ahmadia Apr 14 '12 at 22:32
  • $\begingroup$ I agree with this strategy, and I'm happy to support this within scicomp if the community likes it. $\endgroup$ – Aron Ahmadia Apr 14 '12 at 22:34
  • $\begingroup$ @AronAhmadia: The issue of "one (noun) per answer" is separate from Greg's question, and I would rather a discussion of reopening Greg's question be done in a separate thread, especially because Greg's question is a complicated issue. $\endgroup$ – Geoff Oxberry Apr 15 '12 at 3:04
  • 1
    $\begingroup$ @JamesCuster: I have two concerns. As discussed in the comment thread below the question, questions should be wikified sparingly. What limiting principle do you suggest we (as a community) apply to ensure that 1% of our questions or fewer are wikified? The list questions we tend to discourage are ones that are unmanageable. Your link to the "The Definitive C++ Book Guide and List" is a perfect example of an unmanageable list question, which is why there's a text box suggesting not to use it as an example of a good question. $\endgroup$ – Geoff Oxberry Apr 15 '12 at 3:15
  • $\begingroup$ I think the types of list questions should be made sparingly. As I said in my answer, I think list questions are fine as long as they are broadly useful to the community as a whole (therefore there should be very few list type questions). I'm not sure how to strictly determine that, but I believe the current question being discussed does not fall into that category. Maybe it could be rephrased in terms of a list of linear algebra packages for python and their capabilities (including sparse matrix solving capabilities). Does that make sense? $\endgroup$ – James Custer Apr 15 '12 at 4:52
2
$\begingroup$

I've thought about this issue a lot over the past couple days, and I'd like to propose the following solution for comment:

  1. We should essentially never use the phrase "One (book/solver/package/library/etc.) per answer please". I've seen it used well maybe once: Where can one obtain good data sets/test problems for testing algorithms/routines? and even there, it could be argued that a few of the data sets overlap (for example, there are a couple data sets that cover optimization test problems), and it would be better to condense the list, especially since the question has been wikified.

  2. Let's not wikify these questions. It's a slippery slope to bad things, as I alluded to in the comment thread to the question, but to repeat myself anyway, wikifying questions is discouraged. Unwikifying questions that have been wikified is a separate issue to visit another time.

  3. If you see an answer that is "out of control" in that it recommends both "good" and "bad" resources, you have a bunch of options:

    • Flag it for moderator attention; ask for the answer to be wikified
    • Comment; ask for the answer to be wikified (this option is less effective than flagging!)
    • If you're a 2K+ rep user, you can wikify it yourself
    • Edit it before or after it's wikified

The motivation for this proposal is simple: let's do the simplest, least disruptive things to address the real problems. The real reason people suggest "One package per answer" is to make filtering package recommendations easier, which solves the filtering problem, but does a whole bunch of bad things:

  • It turns the question into a poll on packages, which is discouraged
  • The number of answers gets larger and less manageable
  • It prevents making connections among software packages, as pointed out by JedBrown

If we just want to make filtering easier, let's just wikify answers that pose problems for filtering, like answers that recommend a mix of good packages and bad packages. It could be worthwhile to point out when a package is not of good quality, or no longer maintained, so that newer or less expert users will avoid it. Wikifying those answers empowers users who are topical experts (regardless of their current reputation, as long as it is at least 100) to edit the question to correct or clarify misconceptions. We trust users to write good answers, so trusting them to step in and improve answers (which people already do) is a less intrusive approach more faithful to the Stack Exchange concept.

Wikifying the question causes other issues:

  • It's frowned upon
  • It means every answer will be community wiki, so no one can earn rep from the question, lessening the incentive for user participation

Let's avoid causing more problems than we solve.

$\endgroup$

You must log in to answer this question.

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