A few notable trends emerged while running various scenarios:
Pennsylvania was the tipping point in many of them, both when racial polarization was at a more normal level (Clinton performs worse than Obama with African-Americans but better with whites) and when it was more extreme (Trump outperforms Mitt Romney with whites but nose-dives with Hispanics). This bolsters Wasserman’s case from earlier this year that Pennsylvania might be the tipping point.
Florida, Wisconsin, Iowa, New Hampshire and Pennsylvania were the tipping point state (or close to it) in many of the more racially polarized scenarios. Many of these states make sense — if Trump were to win by running up the score with white voters, some Rust Belt and Midwestern states might fall into his column and push him over 270. New Hampshire, also a highly white state with a number of persuadable voters, could swing to either side in such a scenario. Florida was also a tipping point for some inputs, which agrees with results from the predictive models at FiveThirtyEight and The Upshot.
These results also underscore how small leads in the popular vote can lead to lopsided wins in the Electoral College. To see this, design a scenario where either candidate has a two- or three-point lead and check the margins for states under the tipping point. In most cases, the winning candidate will take a number of swing states by relatively small margins, giving him or her big returns in the Electoral College for a comparatively small popular vote advantage.