bq. Vacancies in the House of Representatives are filled using special elections. These elections occur off the usual American electoral cycle, and the results of these elections are routinely portrayed by the American mass media as indications of what to expect in the next general election. 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.
That is from a forthcoming paper by David Smith and Thomas Brunell. Their central finding is this: if you compare a year where all of the special elections that produced seat changes were won by Republicans to a year where all of these elections were won by Democrats, we would expect Democrats to pick up about 15 seats in the next general election, on average.
They conclude:
bq. We are not arguing that this relationship established here is necessarily causal; rather the results of special elections are a barometer of sorts that provide some information about the national political mood, which manifests itself in the general election that follows. One could, however, imagine a causal connection – high quality candidates witness the results of these special elections and interpret the results as an indicator of swing toward their party, which motivates them to run for office. Future research might investigate whether the emergence of high quality candidates is linked to the results of special elections.
The paper is here.