Thomas Holbrook put together this chart at his blog:
Andrew Gelman points out that we should round to the nearest integer, as a movement of anything less that 0.5% in the polls is virtually meaningless (because of random sampling error).
Also, this is ONLY the last six election cycles – so keep that in mind.
And it shows ONLY the incumbent party candidate in the polls, it doesn’t say anything about how polls moved for the challenger.
And this chart does NOT show whether the polling moved back to pre-debate levels after one week post-debate. It shows only from pre- to post-debate plus one week in either direction.
Also this shows correlational relationships, which does NOT prove causal relationships.
Taken all those caveats, though…
It looks like in 4 out of 6 debates, the incumbent was doing slightly worse after all debates were over, and in 1 (1988) was actually doing better. Only in 2000 was this anti-incumbent movement fairly substantial after the debates (Al Gore’s infamous sighing?).
There doesn’t seem to be a clear pattern as to which debate (i.e. the first, second, or third) produced the most substantial movement. The first one made the largest difference in 2004 but the second one made the most difference in 1992.
In sum, based on this, I think the only reliable conclusion that can (perhaps) be drawn is that the incumbent party candidates tends to (sometimes) go down slightly in the polls after debate time. But again, the correlation/causation rule does not preclude the possibility that these shifts are due to other factors not related to debate performances.
We’ll see what happens this time!