[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: when your sample is the entire population

From   "Austin Nichols" <[email protected]>
To   [email protected]
Subject   Re: st: when your sample is the entire population
Date   Fri, 18 Jan 2008 13:13:32 -0500

Lloyd Dumont <[email protected]>:
If you imagine sampling your "population" data from a superpopulation
of possible populations, via a SRS, the calculated SE refer to that
process; so if you google "superpopulation" you will find more info.
Fisher and Neyman and others famously wrote about this in the
1915-1935 period, and the issue continues to confuse and confound.

Imagine this: testing whether mean earnings are higher among men or
women in some firm.  If you have population data (every worker) there
is no statistical "test" necessary, right?  Simply compare the means
and accept/reject based on that comparison.  But most of us would like
to assume a data generating process where men have a mean m and women
have a mean f and the population data are generated according to some
stochastic process, and we want to test m=f via a t-test, say, not
whether the observed population means are identical.

On Jan 18, 2008 12:59 PM, Lloyd Dumont <[email protected]> wrote:
> Hello, everyone.  I am facing a statistical
> "challenge" that must be commonplace as microdata
> becomes more and more accessible.  I have been
> estimating  models using xtreg, as I have people
> coming and going monthly over about a two year period.
>  Some estimates significant, others not.
> But, if the people in the "sample" are the entire
> population that I am inferring to, conventional
> measures of significance seem inapprpriate.  But, I
> have never read any social science that presents
> regression estimates, and then says something along
> the lines of, "These are what they are.  Significance
> doesn't apply here."
> So...
> -Am I roughly correct?
> -Is there some other measure of "certainty" that might
> be informative in these situations?
> -Is there a name for this type of "sample" or
> estimation issue?  Googling it has been a real
> challenge, though there must be lots of
> writing/commenting on this matter.
> Thanks for your thoughts.  Lloyd Dumont
*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index