538 maps Facebook likes for each candidate by county

[Boston Bread]

Top candidate in each state. Carson - I, Sanders - D, Trump - R.

[Boston Bread]

BUT MUH SOLID D VIRGINIA, MUH SOLID D NEVADA

Seriously though, those look a lot like general election maps. Eerie.

Sander's awful recognition in the black belt heavily skews the south. Just look at Maryland/Delaware.

[Boston Bread]

I noticed that Trump is ahead of Sanders in the Bronx.
Overall, I'd say Carson vs Sanders looks more like an actual general than Trump vs Sanders.

[Boston Bread]

I think you can improve the model by looking at the % of democratic likes.
The model as is underestimates Sanders in republican places like Sioux county, Iowa. If there are only 10% likes for Sanders and 2% for Clinton that's going to favour Clinton.
Basically, a county with 10% of likes for Sanders and 2% for Clinton should be treated the same way as a county with 50% of likes for Sanders and 10% for Clinton.

[Boston Bread]

You must be logged in to read this quote.

The way the model works tends to underestimate Sanders in republican counties and overestimates Sanders in democratic ones, which explains most major cities going for Sanders.

Chicago might have 50% of likes for Sanders and 20% Clinton which is treated as a win for Sanders, while Sioux County, IA could have 5% for Sanders and 2% for Clinton and would be treated as Clinton win even though it should really be the same result.

[Boston Bread]

So I tried factoring in %Likes in relation to the total Dems percentage in each state, but it doesn't exactly work because of the phenomenon where lower total Dems percentages are usually more Clinton friendly states. I haven't figured out a way to account for that while also keeping the Sanders friendly states numbers accurate.

I tried this too and wound up with Sanders winning SC. So my idea doesn't work outside of theory