Undecided? Join the club.

Not me, of course, but they're out there. Nate Silver at 538.com has an interesting write up though. Have a look.

Tuesday, October 28, 2008
The Undecideds

Earlier tonight, I suggested that there were in fact proportionately more undecideds among white voters than black voters, as the McCain campaign claimed today. This was based mostly on a recollection of numbers that I'd looked at several weeks ago when the fraction of undecideds was higher. However, this claim is more debatable now.

Below, among seven current polls which have released racial breakdowns, are the numbers of voters who did not declare their support for either major-party candidate. These totals include undecideds as well as third-party votes -- a group that I refer to as "uncommitted" voters.

Undecided + Other

Pollster White Black Hispanic
===========================================
Gallup 6 5 8
Research 2K 8 2 4
Rasmussen 3 6 7
Battleground 7 5 7
Economist 10 5 11
Pew 12 9 --
Zogby 6 3 --
===========================================
AVERAGE 7.2 5.4 7.4

A somewhat higher proportion of whites (and Latinos) are uncommitted, but the differences are not overwhelming. Suppose that McCain were to win 2/3 of white uncommiteds -- which I'd probably consider optimistic, although perhaps not since the uncommitted whites appear to be fairly downscale. Suppose also that Obama wins 90 percent of black uncommitteds and 60 percent of Latino uncommitteds. Suppose furthermore that the breakdown of white/black/Latino voters is 74/12/10, reflecting slight increases in the latter two groups from 2004's 77/11/8. Under these assumptions, McCain would pick up 3.9 points from uncommitteds and Obama 2.8 points, a net gain of 1.1 points for McCain:

Uncommitted Allocation I

Group % of Electorate to McCain to Obama
White 5.33 --> 3.55 1.78
Black 0.65 --> 0.06 0.58
Latino 0.74 --> 0.30 0.44
===================================================
Total 6.72 --> 3.91 2.81

As I said, however, these assumptions are arguably optimistic for McCain. What if, instead, the distribution of the uncommitteds roughly resembles that of committed voters within each racial group, which means that whites go about 57/43 for McCain, blacks go 95/5 for Obama, and Hispanics go 65/35 for Obama? Under this set of assumptions, the undecideds split essentially evenly:

Uncommitted Allocation II

Group % of Electorate to McCain to Obama
White 5.33 --> 3.04 2.29
Black 0.65 --> 0.03 0.62
Latino 0.74 --> 0.26 0.48
===================================================
Total 6.72 --> 3.33 3.39

Another problem is that we haven't been distinguishing undecided voters from third-party voters. There is an argument that third-party voters should be treated as quasi-undecided voters, since third party support tends to collapse at the voting booth. Nevertheless, Bob Barr and Ralph Nader will probably pick up a collective 1-2 percent of the electorate, and third party support tends to be overwhelmingly concentrated among white voters.

Four of the seven polls distinguished undecided voters from third-party voters. Looking only at those four polls:

"True" Undecideds

Pollster White Black Hispanic
===========================================
Research 2K 2 2 1
Rasmussen 2 5 4
Battleground 4 3 4
Economist 6 3 11
===========================================
AVERAGE 3.5 3.3 5.0

With the third-party votes removed, there is essentially no difference between the number of white and black undecideds, though Latinos perhaps are undecided at somewhat higher rates.

Long story short ... given optimistic assumptions (McCain wins 2/3 of white undecideds, 100 percent of third-party support collapses), the undecided vote is worth a net of about a point for McCain. Given what I'd consider to be more neutral assumptions, there's no particular reason to think that the undecided vote favors him.

My guess is that the truth is somewhere in between and that this is worth, say, half a point for McCain. Even give him the full point if you like. This effect is probably smaller than that of the cellphone problem, from which there may be 1-2 points of cushion in Obama's direction. If on top of that the polls are being overly conservative with their likely voter modeling, the numbers are more likely to be underestimating Obama's standing than overestimating it.