Since Marion County, IN = Indianapolis, apparently, then I can report that Obama won 50% of the white vote there in 2012, plus or minus a couple of points. It is more accurate in counties where the share of the non-white, non-black vote is lowest. It's safe to say that the white vote was effectively tied in Indianapolis.
Could you elaborate? Indianapolis seems like whites would narrowly vote Romney, simply because it annexed almsot all of its suburbs.
I hope this makes sense. I feel like I started repeating myself after a while.
I built the national map out by using formulas that combined voter registration/voter turnout data by race (where available), and where not, Census 2010/2012 data that compared the difference/discrepancy between the share of each state's electorate by race (2008/2012 exit polling) with the share of the population of each state. Essentially, the white share of the vote was calculated by reverse-engineering the totals using non-white support and turnout levels, which are far easier to project as a whole. With all of this information, I then had available the basis for a "national formula" that plugged in all of these variables (turnout as a share of their adult population + support for Obama), and proceeded to build custom state-by-state formulas based on the information I was able to find. For instance, the formulas had to be customized to a greater degree in states like Texas (Latinos), where I had to do three separate rounds of tweaking not only statewide, but at the county level in order to produce something that is still probably quite inaccurate at the county level, but definitely less so than before.
In states like IN - where whites swung by some of the biggest margins of any state - I was able to use the formula with a combination of 2008 exit polling, the racial makeup of the state/likely electorate in both cycles, and the amount of the overall swing statewide from 2008-2012 to further refine the white share of the vote. States like IN were a bit easier to do due to their relative whiteness, and on top of that, analyzing that statewide swing to calculate the white share of the vote was more reliable since whites were such a large share of it. That was then extrapolated to the county level.
I'm quite confident with my projections in Indiana; +/- 2 points or so. However, if it is off, then it is more likely that Obama did even better: I observed - where it was possible for me to do so thanks to exit polling - that my formulas more often under-represented Obama's share of the white vote than it did over-represent it. It's also worth noting that my statewide formula for IN spat out the exact same number as the NYT exit poll ("38.0%" in my formula; "38%" for NYT).
Indianapolis may be to a greater extent like ATL. It seems to me at a glance that the (real) city comprises a greater share of the county's population than ATL proper does of Fulton County; it's possible that just like in Fulton/ATL, your city whites were overwhelmingly Democratic (65-70%) and the suburban folks were 30-40%. Those are just approximations/examples; I haven't actually broken down the county's population proportions et al to see what the most likely ratio would be in that case.