[lbo-talk] Race & Opinions

[Wojtek Sokolowski]

Kelley:
> 
> Did you run a chi-square test of significance?! Your sample is pretty
small!
> 
> speaking of which, I wonder how many of the opinion polls showing a
> significant difference of opinion re: Iraq between blacks and whites
would
> stand up to a significance test. As I recall, whenever I ran
regression
> analyses on n=10000 surveys and tried to examine differences by race,
it
> was difficult to get statistically significant results because the n
was
> too small. That's because, even if 18% of the N surveyed is black,
that's
> still only 180 people. When you try to disaggregate further by
religion,
> income, gender, etc. the numbers get even smaller and the results less
> reliable. statistically speaking, o'course.

18% of 10,000 is 1,800 and that should produce significant results for
even minute differences.  One of the problems working with large samples
is that almost every difference is significant.
A common way of handling relatively small sub-populations is
oversampling (and then weighting) - which allows you to further
disaggregate those population (GSS does that routinely).  
The main point I argued, however, had little to do with the description
of populations but with some views I encountered  without claiming that
they represent those of a larger population.  

Wojtek