According to http://office.microsoft.com/en-us/excel/HP052038731033.aspx 
degrees of freedom in t-test for two unequal variances (regardless of using
TTEST or t-test from toolpak) uses a formula where
(S1^2+S2^2)^2(((S1^2/m)^2/m-1))+((S2^2/n)^2/(n-1))) = DF (call this formula
"Formula 1").

Normally, DF is simply equal to count of sample1 + count of sample2 - 2
("Formula 2"). Is there any reason why Excel uses Formula 1? I have searched
for Formula 1 in a number of statistics manuals and have been unable to find
an example of it.

Best,
David