While I certainly endorse what Sides said, I should note a new article by Tom Brunell and David Smith forthcoming in LSQ. As the authors note:
We examine the predictive power of the results of special elections on the general election outcomes for the U.S. House of Representatives from 1900-2008 and find that those special elections that result in a change in partisan control do have predictive power for the general election.As the scatterplot below shows, there is some significant relationship between special election victories and future general election victories. It's a noisy relationship, but it ain't nothin'.I still think the facets of the NY-23 race are too peculiar to lead us to expect Democratic victories for 2010. Still, the research is suggestive. Maybe we should cut the pundits some slack.