nonrandom sampling


Suppose I have nonrandom samples of sizes n1 and n2 over two groups. I
know the population proportions, so I can compute weights w1 and w2,
which I'd use, eg, for running regressions.

Now suppose I delete m1 and m2 observations from my sample, because of
(say) missing data. I have no reason to suppose that the occurrence of
missing data is anything but random. So I now have two samples, of sizes
n1-m1 and n2-m2.

Question: if I want to run a regression using the smaller data set, do I
need to re-compute the weights? One theory suggesting "no" is that if
the original data were random (ie weights = 1), then eliminating
missing-data wouldn't change that. Is this correct for my case?


--


Phil Viton
Ohio State University
.

Relevant Pages

  • Re: T-shirts for family
    ... Sort of OT-we get T-shirts for our employees and interns from ... a wide variety of weights, sizes and colors. ...
    (rec.arts.disney.parks)
  • Re: Sharp knives
    ... Melba's Jammin' wrote: ... rest with 3 cleavers in different sizes and weights for that as well ...
    (rec.food.cooking)
  • Re: Exercise sites please - Tom?
    ... I have dumbell type weights in different sizes. ... finding that some days it is much harder to get out and cycle, ... I would be happy to help you and while the site Susan has given you looks ...
    (alt.support.diabetes)