Georgia has a surprisingly (to some) close Senate race.  Businessman David Perdue leads Michelle Nunn by 0.5 percentage points in the RCP Average. If neither candidate reaches 50 percent, the race will go to a runoff on Jan. 6, where most analysts suspect Perdue will be favored. But how did things get to this point in the first place?  Is it demographic change in Georgia -- where only about 55 percent of the population is non-Hispanic white today?

As is often the case when studying American politics, it is useful to start with a brief geography lesson, followed by an overview of history, to see how we got here.  Much of the East Coast’s geography is dictated by a fall line -- basically a 900-mile-long cliff that runs in an arc from New Jersey around to Alabama.  Many of the major cities in that region (Baltimore, Richmond, Va., and Augusta, Ga.) owe their existence to this break, as it frequently marked the end of navigable waters in major East Coast rivers.

In Georgia, the fall line basically runs in a diagonal across the middle of the state.

Southeast of the line lies the coastal plain, where plantation culture flourished prior to the Civil War.  Northwest of the line is the Piedmont, which is basically the foothills of the Appalachian Mountains; this is where Georgia’s manufacturers set up after the Civil War.  In the far north is the southernmost reach of the Appalachian Mountains. 

We might think of Georgia’s electorate as divided into five basic groups, which overlap somewhat.

Perhaps the most important groups are the urban centers.  Chief among them is Greater Atlanta -- a mass of counties to the north and west of the fall line -- but important roles are also played by other metropolitan areas:  Augusta, lying about halfway along the Georgia/South Carolina border; Savannah, on the South Carolina border along the Atlantic Coast; Columbus, at the northernmost “indent” along the Alabama/Georgia border; and Macon, near the fall line at the geographic center of the state.

The remaining groups are based upon race.  We can separate out whites roughly along the fall line; factionalism often played out within the Democratic Party along the diverging interests of these groups. A very small vestigial Republican contingent was located in the northern mountains.

African-Americans have historically been concentrated in the “black belt,” a band of unusually fertile soil that runs across the state, and in the cities, especially after World War II.

The final group is the newest one, which we might call “other minorities.” The state’s Hispanic and Asian-American populations are growing fast, but registration and voter turnout rates are low.  They are fairly spread out across the state, with a larger presence in urban areas.

With all that said, we begin our story in 1932, when Georgia showed little regional variation.  Georgia was . . . Democratic.

Herbert Hoover received under 8 percent of the Peach State’s vote that year; Franklin Delano Roosevelt won over 90 percent of the vote in almost 85 percent of Georgia’s counties.  FDR won 726 of the 726 votes cast in Wilkinson County. He lost 22 of the 1,936 votes cast in Washington County.  Hoover’s vote was heavily concentrated in the southern edge of the Blue Ridge, which had voted Republican since the Civil War.

By 1960, the state looked substantially less Democratic:

Because I’ve capped the maps at partisan indexes of greater or less than 10 percent (that is, counties that are more than 10 percent away from the national average in either direction are depicted as having a partisan index of 10 percent away from the national average), the magnitude of the change here is understated.  Georgia moved from having an overall partisan index of +33D to having one of +12D. 

Most of this movement was concentrated in the urban areas.  This is typical of the pre-Civil Rights Act South, where northerners moving to newly industrializing areas brought their Republican voting habits with them, and where capital moving southward began to create a middle class that adopted those habits.

There’s a break in our narrative here, due to unusual circumstances.  The state was overwhelmingly red in 1964 -- Barry Goldwater’s candidacy really was instrumental in changing the Deep South.  In 1972, the state gave Richard Nixon 75 percent of the vote.  In 1976 and 1980, former Gov. Jimmy Carter helped return the state nominally to deep blue status. In 1980, it looked like this:

But in 1984, it was clear that a new Georgia had emerged.

Suburban Georgia was deeply red, as were the counties north of the fall line.  The rise in black voting had helped to turn Fulton County blue, and the black belt is likewise blue.  But what kept Democrats in the game was whites south of the fall line, who continued to vote against Gen. Sherman.  This was true in 1992, when Bill Clinton carried the state:

It was also true in tight Senate contests in 1992,

as well as in 1996:

But in the 2000s, those whites from south of the fall line suddenly flipped, resulting in surprising GOP wins for Senate and governor:

This new division showed up in the three succeeding presidential races:

.

This has had reverberations down-ticket.  For all the talk of the state’s trend toward Democrats, Georgia’s partisan index has been stable since 2000, notwithstanding the surge in black participation that accompanied the Obama elections.  Republicans took control of the state Senate in 2003, and haven’t had an election since where they lost seats. Republicans took control of the state House in 2004, and haven’t had an election since where they lost seats.  Perhaps most tellingly, in 2010 Republicans swept the statewide offices for the first time.

Which brings us, at long last, to 2014. The tightness of the race can be explained by two scenarios, which are not mutually exclusive: demographic change and candidate issues.  The candidate issues argument is straightforward: Michelle Nunn has run a good campaign, and has a family name associated with a conservative southern Democrat.  Perdue has, by most accounts, not run a good campaign, and has made favorable comments about outsourcing that probably play poorly with the sorts of older, rural whites who have favorable memories of Sam Nunn.  Voters who were 18 in 1990 are only 42 today, so there is still a substantial portion of the electorate that remembers the senator. Of course, we can’t know this with certainty until we get actual results.

What I think we do know, however, is that the growth of the non-white electorate is probably overstated.  In nominal terms, the white share of registered voters is down 4.6 percent from 2008, 3.7 percent from 2010, and 1.1 percent from 2012.

What’s odd, however, is that the black vote is perfectly stable, at 30 percent. The Asian share is up 0.2 percent since 2008, from 1.2 percent to 1.4 percent. The Hispanic share is up 0.4 percent, from 1.4 percent to 1.8 percent. The “other” share is stable.

The change in the electorate is almost entirely due to the “unknown” vote.  Who are the “unknowns”? To be honest, we don’t know! “Unknown” means the question is left blank. If someone marks two races, they are categorized as “other.”

We might hypothesize that “unknowns” are disproportionately nonwhite.  But, we really have no way of knowing.  They aren’t concentrated in the black belt:

In fact, there is a statistically significant relationship between the “unknown” share of the electorate and the black share of the population, but it is negative. As a county gets less black, its unknown share increases.  Even this doesn’t tell us much, though -- it could be that blacks in heavily black counties feel more comfortable checking the “African-American” box than they do in heavily white counties.

Maybe these are Asian/Hispanic voters who don’t identify with a particular race.  Or maybe it is whites who don’t check the box as a sort of “protest” vote.  After all, we have Dade, Walker and Catoosa counties, which have nonwhite shares of 5.8 percent, 8.2 percent and 8.4 percent, respectively, and unknown registration rates of 6.1 percent, 8 percent and 8.1 percent, respectively.  It just seems unlikely this is purely a minority phenomenon there.

A method-of-bounds analysis at the precinct/census block level might help us flesh this out further, but for now, the best we can do with “unknowns” is allocate them proportionally to known racial groups.

When we do this, the black share of the electorate climbs about 2 percent since 2010 and 1.5 percent since 2008, which is closer to our knowledge of demographic change in Georgia.  The other groups climb as well. 

But the white share of registered voters declines much more slowly: 2.1 percent from 2008, 2.3 percent from 2010, and only 0.4 percent from 2012. All of these were solid Republican years in Georgia.

This is actually good news for Michelle Nunn.  Remember, in 2010, a good deal of Saxby Chambliss’ lead was built on an improved showing in heavily white counties; black belt and urban counties shifted toward him, but much less dramatically:

This is probably the result of the much-discussed drop-off in minority voting that people have been talking about for runoff elections.

If Nunn makes it to a runoff as the result of a strong showing among white voters, that strength could very well carry through to the runoff.  It’s far too early to say what a runoff electorate looks like, and Nunn will have other problems besides minority turnout to deal with (the University of Georgia goes back into session the day before that election). But if Nunn is really performing better among rural whites due to Perdue’s status as a wealthy businessman who says favorable things about outsourcing, she might have a shot at upending the conventional wisdom in a runoff, and re-creating the coalition that enabled Democrats to win elections in the 1990s.