I used MI for a couple variables, and just want to be sure I know what SPSS did when it pooled test statistics.
I've read that "pooled" is not "averaged"…so what is the calculation being done when it pools results?
Maybe it depends on the statistical test?
To be specific, I will be running t-tests, bivariate correlations, linear regression (all of which are MI-support analyses in SPSS 22).
Thanks in advance!
Best Answer
Pooled typically refers to a "weighted" average. If you have two samples and estimates of each samples variance is $s_1^2$ and $s_2^2$ you might consider the pooled estimate:
$$ s^2 = \dfrac{ (n_1-1)s_1^2 + (n_2-1)s_2^2}{n_1+n_2-2}$$
Note this is not a simple average which would be $$\dfrac{s_1^2+s_2^2}{2}$$
The idea is that each sample might be based on a different sample size and you want to account for that in your estimate (the estimate that comes from the larger sample size should have more of an impact on your final estimate than the estimate from the smaller sample size).