Date: Fri, 26 Jun 1998 14:06:28 -0400
Reply-To: "Susan C. Underhill" <susan.underhill@UTORONTO.CA>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Susan C. Underhill" <susan.underhill@UTORONTO.CA>
Subject: Re: T-tests and small samples?
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
> I am currently under pressure to compare two groups (white
and nonwhite)
> using independent samples t-tests which I am comfortable
with. However, the
> nonwhite group comprises just 11 to 12% of the total
response group (N=452).
> This percentage is representative of the population. Am I
wrong to be
> uncomfortable comparing these two groups?
>
Ideally, when you are designing a study, you make sure that
you would sample enough people to to make this comparison.
Check your sampling procedure to see if you are okay. Do you
have the minimum number of people to make this comparison?
My gut feeling is that you should have at least 50 people in
the non-white group so you might be okay but check your
sampling.
If you do not have enough people in the non-white group (as
we realize real-life data sometimes doesn't work as nicely
as we hope), you could weight but you would have to
recognize the weakness in your design.
The best resource I've read in recent years for sampling and
weighting is Kervin (1992) Harper Collins Publishers.
To weight you would weight them so that they each
represented 50% of the target population. The weighting
formula would be:
Weight factor=.5/achieved proportion of non-whites
and then repeat this for whites
Weight factor=.5/achieved proportion of whites
To apply this weight simply compute a variable:
compute weight=0.
if (race=1) weight=(put results of calculation).
if (race=2) weight=(put results of calculation).
execute.
Then,
weight by weight.
This would effectively upweight the non-whites and
downweight the whites. Your total N should remain the same
(check this). Before you weight you should be sure that
there was no sampling bias (e.g, whites were more likely to
answer than non-whites. Since you said that the proportion
of non-whites was represenative of the population, I'm
assuming this was not problem.
Good luck.
Susan
************************************************************
Susan C. Underhill, Senior Research Officer
Institute for Human Development, Life Course & Aging
222 College St., Ste.106
Toronto, ON M5T 3J1
PH: (416) 978-5968 FAX: (416) 978-4771
************************************************************
