Breaking For The Challenger?

Time was, conventional wisdom held that undecideds broke to the incumbent, the devil they knew. Lately, though, the media have been saying it's the other way around, that incumbents break for the challenger. The new rationale is that, if the undecideds had a favorable opinion of the incumbent to begin with, well, they wouldn't have been undecideds, would they? They were self-selected to break the other way.

The media cite a few stats in that regard that on their surface look good. The stats generally show that, indeed, challengers poll better on election day than they did in the nearest polls.

But their samplings are very limited; a handful of recent presidential elections, at most. So I grabbed the data from the 2002 elections, available at Survey USA and ran the numbers myself.

Indeed it's true. Of 18 elections for Governor, the Senate, or the House, where an incumbent ran, the incumbents lost ground to the challenger as compared to the nearest previous poll. It wasn't a lot on average - a fraction of a point - but it was there. Score one for the media.

Except ... I then ran the numbers by party, which tells a different story.

On average, Dem incumbents lost 3.9 points, whereas GOP incumbents, in their own election races, gained an average of 3.2 points.

If we just look at the direction of the move, ignoring its size, we see a similar pattern. Dem incumbents had an overall record of 2-5-1; i.e. they gained ground (basis the nearest previous poll) in 2 elections, lost in 5, and broke even on the other. GOP incumbents, for their part, had an overall record of 8-2.

The "Undecideds Break To The Challenger" theory does not explain this. A better explanation is that pollsters have, by design or by error, underpolled Republican support and overpolled Democratic. It is very hard to avoid this conclusion.

The recent historical era has been one where Republican challengers have won control of the House, Senate and Presidency. If undecided voters have been supporting the challenger lately, it appears to be - at least, based on the most recent results from 2002 - because that challenger was Republican.

UPDATE: Here's the data used in this analysis.
Abs=Absolute change for the candidate relative to last prior poll;
R-Net=Net Chg for the Republican relative to opponent;
DI-Net=Net Chg For Dem Incumbent relative to opponent;
RI-Ne=Net Chg For Rep Incumbent relative to opponent.

Race	Candidate	Party	Before	After	Abs	R-Net	DI-Net	RI-Net
CA Gov	Simon	R	39	42	3	0
CA Gov	Davis *	D	45	48	3		0
CO Senate	Allard *	R	50	51	1	1		1
CO Senate	Strickland	D	46	46	0
FL Gov	Bush *	R	51	56	5	8		8
FL Gov	McBride	D	46	43	-3
IL Senate	Durkin	R	36	38	2	3
IL Senate	Durbin *	D	61	60	-1		-3
IA Gov	Dross	R	42	45	3	6
IA Gov	Vilsack *	D	56	53	-3		-6
IA Senate	Ganske	R	38	44	6	12
IA Senate	Harkin *	D	60	54	-6		-12
KY Senate	McConnell *	R	66	64	-2	-9		-9
KY Senate	Wienberg	D	29	36	7
KY Congress	Northup *	R	49	52	3	3		3
KY Congress	Conway	D	48	48	0
KY Congress	Davis	R	46	48	2	2
KY Congress	Lucas *	D	51	51	0		-2
MO Senate	Talent	R	49	49	0	-3
MO Senate	Carnahan *	D	47	50	3		3
NV Gov	Guinn *	R	63	68	5	11		11
NV Gov	Neal	D	28	22	-6
NY Gov	Pataki *	R	47	49	2	1		1
NY Gov	McCall	D	32	33	1
NY Gov	Golisano	I	17	15	-2
OH Gov	Taft *	R	57	58	1	2		2
OH Gov	Hagan	D	39	38	-1
OK Senate	Inhofe *	R	53	57	4	3		3
OK Senate	Walter	D	36	37	1
OK Congress	Sullivan *	R	56	56	0	-1		-1
OK Congress	Dood	D	41	42	1
OK Congress	Pharaoh	R	25	26	1	-1
OK Congress	Carson *	D	72	74	2		1
SC Gov	Sanford	R	46	53	7	12
SC Gov	Hodges *	D	52	47	-5		-12
TX Gov	Perry *	R	53	58	5	8		8
TX Gov	Sanchez	D	43	40	-3
Avg						3.2	-3.9	2.7