I do, but you can make a case the other way.
When I'm making picks (with the automated system, that is), I
throw every negative I can at the projected outputs. Pool dilution,
tax effect, statistical variance, corrections for small numbers
of observations, etc. My theory is than anything that can survive
that is a good bet.
So, in that scenario, I add my bets to the pool and recompute
the expected payout.
You could also say that your bet is just part of the great statistical
swamp, especially if you're always in there with lots of dollars,
anyway.
It would be interesting to run some of my historical analyses
both ways.
