Friday, December 16, 2011

Is the Game Over? (concluded)

Rove's map appears to be based on "uniform swing", the model used by pundits in the UK to translate national polls to projected seats in Parliament (polls in each constituency are impractical) assuming that the parties' vote-shares will change from one election to the next by the same amount everywhere, any mis-predictions in one direction due to local variations cancelling with those in the other. Since Obama won the national vote by 7% in 2008, a swing in the -8% range, if exactly uniform (enough to flip Virginia, which had +6.3% for Obama, but not Colorado at 8.6), would create the situation of the Republican winning the popular vote but Obama getting 273 electoral votes (with all the uproar that would cause). Rove's list of targetted states are the ones that would flip with a swing about -10 for Obama, and presumably he hopes that if the national swing is about -8, local variations in at least one of the targets would flip it (but Virginia must not vary in the opposite direction). Obama's approval-minus-disapproval is at -4 recently, a swing of -11, but "disapprovers" might still turn in to "lesser-of-two-evils" voters: Obama is still at +2 over Romney (a -5 swing) or +11 over Gingrich (a +4 swing). The "approval" question is inherently volatile (often heard by the audience as "do I approve of what he has done most recently?") but that is not the whole problem: as with polls about the health-care bill, "disapproval" messily combines those who think the President has gone too far with those who think he has not gone far enough. Real Clear Politics shows some sub-tabs indicating that much of the "swing" away from Obama is among Democrats, blacks, self-identified liberals, the young; in short, among those who would never consider voting Republican; whether they will show up at all is the question, but a sufficiently scary opposing candidate might provide the motivation if Obama cannot. Obama is holding his own among retirees, Hispanics, and Jews, all groups the Republicans wish they could turn.

Axelrod by contrast looks at the map in a very "granular" fashion, thinking of which particular local swings would suffice: making the effort to peel off Omaha's 1 electoral vote is archetypal of what Axelrod did in 2008. It can be assumed that Obama will not do as well this time, but in some places the negative can be kept down low enough to hold, or a positive generated for some "offense". His five alternatives are: going all-in on Florida while playing "Pure D" with the hard-Blues; holding Ohio and one or two related soft-Blues; concentrating on holding in the New South; or on the Rocky Mountains, with either some Baby Blue defense or Pink offense. So, this raises the question of whether "uniform swing" is a reasonable model in the US (in the UK, Nate Silver and others attempted to find a more predictive model, with poor results) or whether the regional variations are crucial. There is the famous "net-Red" map in which counties are colored Blue or Red not based on Obama's margin, but on his margin minus Kerry's; so almost everywhere is Blue (where Obama lost, he generally lost less badly than Kerry) but a band from the Ozarks to the Appalachians was actually net-Red, stronger for McCain than for Bush (or rather, more strongly against Obama). On the other hand, the change from the 1980 Reagan-Carter map to 1984 Reagan-Mondale is a disastrous case of uniform swing: Mondale did not do very much worse than Carter, but he did so consistently, not-quite-carrying a lot of states for a memorable electoral-college wipeout. So I have analyzed not just the state-by-state swings, but the variations among them.

The case-studies were the swings from 2000 Bush-Gore to 2004 Bush-Kerry, and from 2004 to 2008 Obama-McCain. The net-Blue swing 2000-to-2004 had a mean of -2.1, standard deviation 4.5, while 2004-to-2008 had mean +9.9, standard deviation 7.3. It is a little odd that the mean swing +9.9, averaging over the 51 cases (including DC) treating small and large states as of the same weight, is rather higher than the national swing +9.2 (from Kerry's negative margin to Obama's positive); one outlier, the +26.6 in Obama's birthplace of Hawaii, accounts for about 0.4 of it (DC, consistently a run-away for the Democrats, is an outlier in raw margin, but not atypical in terms of the swing) but the rest reflects the campaign's strategic focus on turning out votes where it mattered. The higher standard deviation is also significant: there was wider variation from state to state in how people reacted to the very novel candidacy of Obama; the Republican ticket and platform were also changed, which could contribute to wider variability in how McCain-Palin, likewise, were received in one state or another. By contrast, 2000-to-2004 was more uniform as it shared features with 1980-to-1984: same candidate on the Republican side (true, incumbent-Bush or incumbent-Reagan, with some track record, was not perceived the same as candidate-Bush or candidate-Reagan; and likewise, incumbent-Obama in 2012 will not be identical to candidate-Obama of 2008), and rather similar candidates on the Democratic side (Kerry was not identified with Gore, as Mondale was with Carter; but both were perceived as bland personalities lacking in firmness). I suggest therefore that the degree of uniformity depends upon how "familiar" the electoral choices are: if we get not only the same candidate on the Democratic side, but also a Republican ticket that feels similar to McCain-Palin, with Romney still suspect to the evangelicals and hard-line conservatives but choosing a firebrand running-mate to compensate, we can expect the swing to be fairly small and fairly uniform. A very different ticket could shake things up, but cut both ways from one part of the country to another.

Looking at the raw Blue margin, some states are fairly "rigid" and disregard national swings (in Massachusetts, the margins for Gore, Kerry, and Obama were +27, +25, and +26; in Oklahoma, Kerry and Obama at -31.2 were indistinguishable even to the tenth of a percent) while others are "national microcosms" whose swings are within a percent or two of the average (Iowa and Pennsylvania, notably). Moving to the net-Blue shifts in 2004-to-2008, opposite from the five net-Red states (Oklahoma, Arkansas, Lousiana, Tennessee, West Virginia) where the shift was actually negative against the national current, the six states with highest net-Blue are an odd collection: Hawaii (outlier), Indiana (a surprise victory), Nebraska (the Omaha pick-off), Montana (McCain by plurality only), North Dakota and Utah (neither were any threat to go for Obama, but he did surprisingly well). It is instructive to look at "net-net-Blue", that is, the difference between the local swing and the national swing (where a candidate over- or under-performed compared to his average performance). To see if there are any real underlying regional trends, divide states by whether the net-net-Blue was positive two times in a row (in 2000-to-2004 and again 2004-2008) or negative two times in a row; each should happen one-fourth of the time, so finding 15 double-positives and 13 double-negatives looks like random chance.

Except: there is distinct clumping. The 13 double-negatives include a contiguous block of the 5 net-Reds (OK-AR-LA-TN-WV) and 3 neighbors (Missouri, Alabama, Kentucky) tenuously reaching through Pennsylvania (but its net-net-Blue is negative only by <1% and then 2%) to New York; disconnected are Florida, Rhode Island, and Arizona. The 15 double-positives include a contiguous block Oregon, California, Nevada, Colorado, New Mexico, Texas, Idaho, Montana, North Dakota, South Dakota; and a little block of Wisconsin, Illinois, Michigan, not touching North Carolina or the strongest double-positive, Vermont. In the Northeast, then, it does appear to be just statistical noise. The continued rightward drift of the Interior South (weakly reaching into The T in Pennsylvania and upstate New York, though not enough to tip either state) is of course very well known. But the leftward drift of the West (not just the Left Coast, where any increase in the Blue margin is irrelevant to the electoral college, but deep into Pink territory) is a less-noted trend. The Western branch of the Republican Party does not seem to be getting any attention from other branches: three former governors from the West are running, with Huntsman of Utah unable to break out of single digits, Johnson of New Mexico only getting enough attention to be invited to a debate just once, and Roemer of Colorado not even a blip. It is hard to say why this disconnection is happening, but my sense is that the Republicans have little chance of retaking any Western states, and a good chance of dropping one or two.

No comments:

Post a Comment