NPR has an interesting new online gadget that allows readers to test what seems to be the central working theory of Donald Trump's campaign: that their man can win by getting more white voters to show up and vote Republican than in recent presidential elections.
The headline result is that Hillary Clinton would dominate Trump if demographics are indeed destiny: Adjusting 2012 results for differences in the population four years later, she would be expected to capture 345 electoral votes to Trump's 193 -- not as large a victory as Barack Obama won in 2008, but larger than his win in 2012. What would it take for Trump to win? One scenario: Trump increases the GOP's share of the white vote of at least 2 percentage points, with white men also increasing their turnout by 1 point -- while every other demographic group votes the same way as four years ago. That's not impossible; nor is it the only scenario that generates a Trump win, just the one that requires the least tinkering. But it doesn't seem particularly likely, either (mostly because of the assumption other groups would remain static).
For me, the most interesting thing about the NPR model is how likely it deems a Clinton win in Georgia. With partisan shares of the vote and turnout the same as in 2012, adjusted only for 2016 demographics, NPR's model predicts a 2.5-point win for Clinton in Georgia. That would be a 10-point swing from 2012, when Mitt Romney beat Obama by almost 8 points. Is that really possible in just four years, based solely on demographics?
Even in the above-described scenario that gives Trump the presidency, Georgia would be the last state in his column, providing the entire margin of victory in the Electoral College. We would go Republican by a smaller margin than some long-time purple states: Florida, Colorado, Ohio, even Pennsylvania.
Frankly, I have a hard time believing Trump will win Ohio by almost 4 points, Pennsylvania by more than 3, and Georgia by less than 0.2. Diving down into their data, I think I know why the model has it wrong.
For one thing, the model assumes a huge jump in black voting, well beyond even what Obama inspired in 2008. The underlying data for the model shows black voters making up 34.5 percent of the electorate. In 2008, they hit a record of 30.1 percent, falling to 29.9 percent in 2012 and 29.3 percent among current registered voters.
What's more, the model's numbers assume 87.5 percent of currently registered white voters will show up in November, which isn't necessarily absurd if you consider registration will increase between now and then. But it also projects 101.5 percent of currently registered black voters and 121.2 percent of currently registered Hispanics will turn out. Based on the turnout level in 2008, that would require the new registration of almost half a million black and Hispanic voters. Recall that the much-hyped New Georgia Project two years ago sought to increase minority registration by just 120,000, and wound up with just 46,000 . The group's goal this year is 170,000 -- very ambitious compared with their 2014 results, and yet only one-third of that half a million figure.
The one wild card here is that the fastest-growing group of voters in Georgia aren't black or Hispanic, but "unknown." As recently as 2004, "unknown" made up 1.1 percent of voters. Among current registrants, it's up to 7.8 percent. That's not just a faster rate of growth than various minority groups; it represents nearly as large an increase in the absolute number. If we assume a disproportionate share of "unknown" voters are actually black and Hispanic, those groups might reach the share of the vote NPR's model assumes. Maybe. Even if it does, would it really affect the outcome of the election here?
Despite pretty dramatic demographic shifts here already, the Republican share of the presidential vote has been fairly consistent. In 2000, George W. Bush won 54.7 percent in Georgia. The average share of the vote for Bush, John McCain and Romney in the three elections that followed: 54.5 percent. In 2012, Romney improved on McCain's showing and came within 1.5 points of Bush's result in 2000 -- even though the white share of Georgia's electorate fell by almost 14 points during that time. Democratic candidates performed a bit better from 2004-2012 than in 2000, but most of their increase came at the expense of third-party candidates.
If Clinton is to have a chance in Georgia this year, it will most likely be because third-party candidates do well at Trump's expense. It's worked out that way for a Clinton before: We saw that happen in 1992 and 1996, when Bill Clinton won Georgia by 0.6 points and lost by 1.2 points, respectively, while Ross Perot took 13.6 percent and 7.2 percent. If the Libertarian and Green candidates take a Perot-like share of the vote, and if most of their votes come at Trump's expense, and if black and Hispanic turnout rises sharply, then we could see Clinton win the state.
But NPR's model doesn't account for third-party candidates, and yet it shows Hillary winning Georgia by a larger margin than any Democrat since Jimmy Carter. Georgia may be more competitive this year, and it's certainly possible Trump could fritter away his built-in advantages here. But beware predictions that Democrats will turn Georgia blue with little more than demographic increases. The numbers just aren't there -- yet.