5
$\begingroup$

Some of you may have heard about the Polymath projects in research mathematics. Polymath is an approach that was conceptiualized by Timothy Gowers to enable "massively collaborative research" in mathematics. Since the initial conception there are several polymath projects that the math community has started (some of which can be found at this wiki for polymath projects.)

Based on my cursory outsider perspective of how these projects have evolved, it seems to me that SE software is not suitable for polymath projects for the following reasons:

  • Multiple attempts to solve a math problem are often tried and a threaded discussion often becomes necessary to make sense of the different proposals.
  • Voting up/down solution ideas may be counter-productive as what may seem as a dead-end initially can be the 'winner' once the difficulties in the suggested solution are resolved after long back-and-forth discussions.
  • Votes 'destroys' the threaded nature of comments/answers and in any case there is no downvoting possible for comments.

However, I feel that the above disadvantages may not be necessarily hold for applied statistics projects. An applied statistics project typically has the following stages:

  1. Identify a research question and collect suitable data.

  2. Exploratory data analysis which drives model specification and further analysis.

  3. Choose several competing models that provide alternative answers to the research question.

  4. Perform analysis and select 'best' model

  5. Data visualizations to highlight aspects of the raw data and to highlight model fits etc.

It seems that the SE software can be used for a Polystats project in the following manner. Since each phase of the typical statistics project is modular, we can ask a 'question' for each phase and let the community come up with different answers. For example, suppose that we want to address whether there is global warming (a potentially bad idea for our first polystats project) and that we have temperature data from some external source.

The first question would be:

  1. Here is the data reg global temperatures for the past xx years. What kinds of plots would help us in specifying a model? An actual plot would be an answer along with a brief explanation why that helps. (perhaps, it suggests a functional form, a data transformation etc).

    Once we have collectively exhausted the answers to the above question and reached consensus (e.g., via the two highest voted answers) we lock the question and move on to the next one.

  2. Given what we know from qn 1, what would be some appropriate model specifications.

    Again once we have consensus, we lock this question and move on to analysis.

We need to obviously establish some ground rules in order for this process to work well but first things first:

What do you think about using the SE software to start a "Polystats Project"?

$\endgroup$
10
  • $\begingroup$ +1 I've been thinking of the same thing. Great idea! $\endgroup$
    – Shane
    Sep 7, 2010 at 17:26
  • $\begingroup$ Like this idea. But can we avoid the emphasis on statistics through the name "polystats"? $\endgroup$
    – ars
    Sep 12, 2010 at 16:39
  • $\begingroup$ @ars Sure, we can. What would you suggest? Polydata seems kind of weak, Polyanalysis seems kind of generic. Any ideas? $\endgroup$
    – svadali
    Sep 13, 2010 at 0:18
  • $\begingroup$ Besides, it seems from the current CW question that suggestions are more in the direction of a data driven project. polydata reflects that, rather than potentially misleading assumptions that we'll be proving statistics theorems with analogy to polymath. $\endgroup$
    – ars
    Sep 13, 2010 at 0:58
  • $\begingroup$ @ars I do not like polydata because it is not informative about what we are doing. The focus of the projects is on analysis of a particular dataset rather than some specific dataset per se. I agree that we are not necessarily proving theorems but 'stats' is not necessarily synonymous with proving theorems. Shall we wait for others to chip in with some suggestions? $\endgroup$
    – svadali
    Sep 13, 2010 at 2:00
  • $\begingroup$ Generic, uninformative? Heresy. It's all about the data, dude! :) Sure, let's wait, though it occurs to me it might be too late to change now. Unfortunate that we have a statistics-centric tendency, but so it goes. $\endgroup$
    – ars
    Sep 13, 2010 at 2:45
  • $\begingroup$ @Srikant: by the way, I think this is a great idea and thanks for taking the initiative, so hope I didn't come out sounding negative about that aspect. Deleted an earlier comment because think it might have sounded inadvertently snippy! $\endgroup$
    – ars
    Sep 13, 2010 at 20:01
  • $\begingroup$ I should add that my broad goal is to get more machine learning types involved here and I worry site names and project names prevent that. That was really behind my concern. $\endgroup$
    – ars
    Sep 14, 2010 at 4:36
  • $\begingroup$ @ars I understand the issue and I am open to new ideas as long as they relate to what we are trying to do. Polyanalysis seems to be the one that is neutral in tone and is related to what we are doing. $\endgroup$
    – svadali
    Sep 14, 2010 at 11:29
  • $\begingroup$ @Srikant: on reflection, I think it's a nice and short name that works, or at least that there's no urgent need to "fix" it. If this project gains notoriety, it should draw people to the site regardless. And since our site name should change soon, it won't sound exclusive to people visiting from outside statistical disciplines. Thanks. :) $\endgroup$
    – ars
    Sep 15, 2010 at 22:21

1 Answer 1

3
$\begingroup$

I am very interested to see how we can use this site for a greater purpose beyond individual Q&A.

Regarding your proposal: I would be more interested in seeing a "polystats project" which tackles big open questions in statistics, rather than specific statistical projects. Although I think the site could be used for both.

There should be a higher-level aspect to this to make it cohesive (such as a group blog which would catalog the progress as a single thread).

See this related discussion the CS Theory stackexchange site. I had started a thread on that site to identify a list of important questions, and similar thread was started on here recently.

We can also use the idea of a data analysis competition in conjunction with this.

Edit:

After re-reading your question, I think that the question as you phrased it is a great idea, and keeping it grounded in a real problem is the right approach. We can build something like this in stages (as you are doing here) and open a meta thread that would ask "what should our first project be" where people can list proposals. This kind of incremental solution to a data analysis project could be managed completely on SE without relying on any other third party software. I propose that we tag these questions with [polystats-project].

All being said, there is of course nothing to stop someone from doing this currently without any meta-discussion.

$\endgroup$
8
  • 2
    $\begingroup$ I am inclined to start small so that we do not get discouraged too easily and in order to better understand what sort of ground rules we need to have. As an example, Terrence Tao recently moderated a mini-polymath project on a problem that was posed in the math olympiad. Small projects will enable us to achieve success relatively easily and we could then move on the big ones. $\endgroup$
    – svadali
    Sep 7, 2010 at 17:54
  • $\begingroup$ That's a fair point. Well, I think that this requires two things beyond the existing site to work effectively: a group blog and a wiki (similar to Terrence Tao's approach). I'm happy to set these up unless someone else has a burning desire to do so. $\endgroup$
    – Shane
    Sep 7, 2010 at 18:02
  • $\begingroup$ In response to your edit- I am inclined to wait for another day and ask a question on the main site seeking potential ideas for our first polystats project with some suggested guidelines for our first project. Asking on the meta may result in a sub-optimal choice as the voice of the community would be mis-represented. Note that not everyone visits the meta site. $\endgroup$
    – svadali
    Sep 8, 2010 at 3:00
  • $\begingroup$ Good ideas ... I remember that it is written somewhere that if the subject is already known in advance by participating people, the starting of the project is too chaotic and the principle of collaborative thinking cannot rise.... how to deal with that ? $\endgroup$ Sep 8, 2010 at 13:44
  • $\begingroup$ @robin I have no ideas on that issue. Perhaps, we can just wait and see how a polystats project will evolve. Of course, there is the possibility that it will not take off at all. $\endgroup$
    – svadali
    Sep 8, 2010 at 17:17
  • $\begingroup$ @Srikant: Would a discussion of a polystats project be "on-topic" on the main site? $\endgroup$
    – Shane
    Sep 8, 2010 at 17:45
  • $\begingroup$ @shane I would not pose it as as a discussion. Rather, the way I would phrase it is: "What should be our first polystats project?" with a link provided to this meta thread. I would of course suggest some guidelines as to what which type of projects could qualify for our first project. The answers and the associated votes would determine our first project. A clear question in need of specific answers. I have been thinking of one issue though: I think polystats project related questions should not get rep but answers should get rep. Is this possible? $\endgroup$
    – svadali
    Sep 8, 2010 at 18:30
  • $\begingroup$ @shane In fact on second thoughts, for the first question mentioned above the answers should also be CW so that answers can be downvoted/upvoted without rep issues. However, for other phases of the project (e.g., model spec, analysis etc) it would be ideal if the qn itself got no rep but the answers do get rep. This is required because asking the qn itself is trivial and is more of a moderator type of work but providing good answers may not be easy and hence needs to be rewarded by rep. $\endgroup$
    – svadali
    Sep 8, 2010 at 18:50

You must log in to answer this question.