10
$\begingroup$

I feel like I'm being a nuisance here as I've posted similar questions on meta before. It's been a few days and there have been no comments or answers to my question. It's just a bit frustrating. I know I can offer a bounty - I did that for this question, but to no avail so I'm reticent to do it again and give up my hard-earned rep !

Can I improve my question ? I spent a long time researching it and I thought I wrote it OK ?

Or maybe it's just that my questions are a bit stupid / naive ? Everyone has to start somewhere, right ? Even a suggestion to point me in the right direction would be great !

$\endgroup$
10
  • 2
    $\begingroup$ Your questions look good to me but it appears that one has to read those papers in order to provide a decent answer. This may limit the interest on the question. Maybe a more specific question would be more successful. $\endgroup$
    – user10525
    Sep 13, 2012 at 21:15
  • 1
    $\begingroup$ No answers doesn't always mean a bad question. Looking at your questions I wouldn't say there is anything you are doing wrong as such. They are high level questions that require a lengthy and considered response to adequately address and no one yet has been prepared to put in the time to formulate such a response. I'd echo Procrastinator in suggestion that tightly focused questions are more likely to be answered. $\endgroup$
    – Bogdanovist
    Sep 13, 2012 at 21:29
  • 1
    $\begingroup$ I am familiar with propensity score matching in case control observational studies and have some experience with it but not when multiple imputation for missing data is involved. I agree with Procrastinator. Even though I would love to help you i am not familiar with those papers and without the proper background I cannot provide a good answer. It is perhaps an investment of too much time to read the papers and study your problem to answer your question. Perhaps it would help to know how the imputation was done and what covariates need to be balanced for proper analysis. $\endgroup$
    – Michael R. Chernick
    Sep 13, 2012 at 21:31
  • $\begingroup$ To do multiple imputation on missing variables require knowledge of the conditional distribution for the missing covariate given the other covariate values. It would seem that the matching in the propensity scoring could involve taking advantage of this knowledge of correlations between covariates to match as well as impute. But that is just off the top of my head guessing. $\endgroup$
    – Michael R. Chernick
    Sep 13, 2012 at 21:35
  • $\begingroup$ @MichaelChernick thank you kindly. I think I understand your point about the conditional distributions, but my impression is that, while it is important for the specification of the imputation model and the propensity score calculations, it is not related to the question of whether (after calculating propensity scores on all the imputed datasets) do you average them, or sample from them, before going on to use them ? I hope I am explaining myself - I am not familiar with some of the terminology. $\endgroup$
    – Joe King
    Sep 13, 2012 at 21:45
  • 3
    $\begingroup$ Dear Joe, I think you are asking fantastic, interesting, challenging questions. I think the lack of answers and comments is a reflection of the challenging nature of them and not a reflection of lack of interest. (Do notice that each of your questions gets well above the median vote count.) Personally, I wish I had more time to think about them and dig into them more. Please keep asking them! :-) $\endgroup$
    – cardinal
    Sep 13, 2012 at 23:20
  • $\begingroup$ @cardinal thank you for your friendly comment, though as a beginner in statistics it surprises me that you and others think my question is challenging ! While it is frustrating that I sometimes don't get answers, on the other hand, writing it does make me think about the question more, so that's good :-) You mentioned the median vote count ? Where can I see that ? $\endgroup$
    – Joe King
    Sep 13, 2012 at 23:29
  • 2
    $\begingroup$ Joe: data.stackexchange.com/stats/query/75712/…. The median is 2. $\endgroup$
    – whuber Mod
    Sep 14, 2012 at 1:45
  • $\begingroup$ I don't understand what is hapenning. Yesterday there was an answer here from @MichaelChernick that had been migrated from CV, and now it is nowhere to be found, except for a "migrated from stats.stackexchange.com 8 hours ago" but nothing following... $\endgroup$
    – Joe King
    Sep 14, 2012 at 6:30
  • 1
    $\begingroup$ That answer was not appropriate here, Joe: it addressed the question you reference, not the question here. Questions about the site are "meta-questions" and are posed and discussed here on our "meta" site. Questions about statistics and data analysis are posed and discussed on the main site. We don't mix them up. Also, @Michael had complained that he couldn't get any reputation points for his answer here on meta. I therefore deleted his answer and invited him to copy it into your thread on the main site. (Nobody has the power unilaterally to migrate answers from one thread to another.) $\endgroup$
    – whuber Mod
    Sep 14, 2012 at 16:09

0

Reset to default

You must log in to answer this question.

Browse other questions tagged .