« Blog Forum – Significance Testing | Main | Social Science Statistics Blog »

29 January 2007


Tracy Lightcap

Professor Henkel:

You did indeed understand my question and I share your reservations. I think the re-sampling techniques may show more promise. Thanks for taking the time to share your expertise with us.

William Henderson

Ramon, this was quite an interesting and informative primer. Thanks for such a clear exposition. bh.

Tracy Lightcap

Professor Henkel:

What's the current thinking about the use of significance testing with cross-sections of populations? When I was back in school, we normally got the Blalock take on this: that population cross-sections could be seen as a sample of an set of possible population values, presumably extending over time. Thus, normal significance testing could be applied, provided one kept the limitations were kept in mind. Today, if I'm not mistaken, there has been a vogue for non-parametric estimates of error, usually drawn from bootstrap procedures. These new estimates can then be used to conduct the usual significance tests.

Of course, if I showed more industry and had the maths, I could determine the state of opinion on these matters for myself. Lacking both, might I ask you to expound on this? The question is still pressing; one sees studies using usual significance tests on populations regularly. Thanks for participating here.

Professor Lighcap:
Since I retired from teaching over a decade ago and have not kept up with the literature, I can't really say what the current thinking is regarding population cross-sections and I'm not certain I understand your question. I think your question is "Can an existing population be considered a probability sample of the hypothetical population unbound by time or space?" If that is the question, my position is that a yes answer involves making an additional assumption that cannot be demonstrated to be either true or false, but from the perspective of a social scientist (yes Blalock was a social scientist as was Hagood who raised this justification many, many years ago) I'm not at all convinced that existing populations are samples from the hypothetical unbound by space and time population, to which theories usually apply, for many characteristics of interest.
As to the non-parametric approach, if the sample isn't a probability sample of the population to which one wishes to make inferences, one simply cannot use mathematical techniques to make the sample a probability sample, making the use of probability theory (statistical inference) questionable at best.

The comments to this entry are closed.


June 2022

Sun Mon Tue Wed Thu Fri Sat
      1 2 3 4
5 6 7 8 9 10 11
12 13 14 15 16 17 18
19 20 21 22 23 24 25
26 27 28 29 30    

Site Meter

Creative Commons License

  • Creative Commons License
Blog powered by Typepad