On Fri, 20 Sep 2002 07:42:08 +1000, "Ken Reed" <[EMAIL PROTECTED]> wrote:
> I have data from a survey carried out in several countries. The countries > themselves are not randomly sampled, and have different sample sizes. > > I want to be able to pool the data to compare, eg gender differences within > the pooled sample. > > Is there any problem with re-weighting the data so that each country sample > is the same proportion of the pooled sample? > Problem? - how will you describe it. I don't think you have much hope for useful statistical tests, regardless of how you do weighting. In effect, you can have equal weighting by country when you report that the average of the national rates -- or better, the median. Do you have something similar across all countries, or different? If different, HOW different? One main purpose of averages, when they are intellectually informative, is to convey the central value of a bunch of values that are, indeed, similar. When the differences are what is remarkable, then you need to find how to discuss the differences. One style of report that I have seen is where the weighting gives you a <crude> estimate of how many people in the world are AAA, versus BBB. Extreme 'extremes' really do have to be noted. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
