Originally Posted by
sgopal2
You have completely misunderstood what I've tried to do. I'm not trying to compare $$ spend for GS qualifiers vs non-qualifiers. This would take a large sample of people who did not qualify for GS to submit data. Instead I compared new GS qualifiers vs requalifiers. Similarly I compared GS from hub cities vs non-hub cities. The parameters I compared were drawn from a convenience sample: PQD,PQM, UAPQD, 3 year PQD, BIS, CPM, PQS and others. The sample sizes were sufficiently large, and the distributions were close to normal. I also took a close look at the residuals and q-q plots. So because of this, I felt comfortable using a parametric test such as the t-test. More sophisticated modeling could be done, but for the purposes of this simple question, I didn't think it was necessary.
Actually I do understand what you tried to do and my point still applies. A comparison of average PQD for GS qualifiers vs. re-qualifiers only compares the relative spending behavior of these two pools (and that's only based on an extremely tiny sample). It tells you absolutely nothing about the qualifying requirements for these two categories and it tells you nothing about why the behaviors might be different.
Suppose the average PQD for re-qualifiers is 10X the average for initial qualifiers. Does that mean it's harder to re-qualify? Or easier? You really can't tell, because the averages tell you nothing about what the minimum is for either group.
All you can tell is out of the small sample who answered the poll, the re-qualifiers were bigger spenders. That could just be because re-qualifiers who aren't huge spenders didn't bother to answer because it's not that big a part of their life. LOTS of possible reasons. And you have absolutely no statistical basis for selecting one over the others.
Unfortunately, confidence in the mathematical minutia of statistics tends to go hand-in-hand with poor modeling technique.