T-stat looks too good.
Use robust standard errors--
This hit home, since I just had to figure out how to do
clustered standard errors
for a paper. Lukily, my significance didn't go away.