My early model for Indiana is projecting: Sanders 50.6 / Clinton 49.4

This is Sanders down 2.8% from the previous 53.4% projection. It also is in disagreement with the polls, 538, and PredictIt - who all say Clinton will win.

I added a PredictIt variable to the model. I use the PredictIt close price for Clinton (percent chance of winning) the day before the election. Then I convert it into standard deviations (which makes it linearly correlated with the predicted percentage). Turns out it is a fairly minor factor, except for caucuses.

I also updated my poll numbers and 7 day search trend data.

Take this with a huge grain of salt. Hopefully my real-time county swing model will do better.

My county model (or more accurately one of the models - as the situation is more of a continuum) of primary only states came up with a prediction of Clinton 62.0 / Sanders 38.0 for NY.

That said, the main goal was to predict county values that would let me use my real-time county swing analyzer to predict the state swing.

I decided to create a state level model for predicting the 2016 Democratic nomination race.

So I've aggregated my county level data into state level variables. I also added election results for WY (actual votes - from http://www.crowdsourcingdemocracy.org), KS, and AK (state delegates).

Note the election results for WA are legislative district delegates, and for ME they are state delegates - not popular vote.

My various state models give different results for NY. I've got Sanders at anywhere from 33% to 46%.

Download

I think that given the number of polls, it is likely that a purely polling based forecast will prove the most accurate for NY.

Currently Pollster has it at 56.2% Clinton / 43.8% Sanders.

My demographics models are significantly more pro-Clinton, with Clinton being at around 66%. It is possible that the truth will like somewhere in between in which case we could see a 60/40 split.

If you want to make your own county-level model for the 2016 Democratic nomination race, you can use my data set.

Download the Data Set

I would LOVE to hear from anyone who is using this. What does your model look like? What variables are you including? What are your predictions? What additional factors are you adding to the model that I don't have?

Using my county-level model and a caucus model (excludes the primaries), which has a much smaller standard error than when I combine primaries with caucuses - I am predicting WY - Sanders: 70.5% / Clinton 29.5% (of the Clinton + Sanders vote).

I am hoping that this will be within 2-3 percent of the final result, but no guarantees. My WI forecast was off by 5.85%.

This is based on race, income, age, sex, old FB like data (could be an issue), education, density, and past election results. It also includes Google Search trends for the last 7 days (the latest 7 days possible).

I've been updating my county-level model to predict the outcome of the democratic primary.

I recently added a Google search trend variable that uses the last seven days before the election (but not including the actual election date) and is equal to Sanders / (Sanders + Clinton). While Sanders dominates the search engine trends (typically 2:1), there is a strong positive correlation between the percent of searches that he gets and the outcome. As of this past hour, Sanders is getting 73% of the searches in Wisconsin - which is a strong showing.

Previously I was relying upon Obama presidential 2008 vote to be solely predictive of turnout. However I've now created a turnout model for primaries with additional variables.

When I apply this to my predicted Sanders vote percentages (county-level), I get some small to medium sized changes in my predictions. Old values in parentheses.

Sanders Predicted Vote Share:
CT: 45.2% (45.3)
DE: 39.6% (35.5) <--- biggest change
IN: 53.4% - no change
KY: 44.3% (44.0)
MD: 27.6% (28.1)
NY: 33.1% (33.9)
OR: 71.8% (71.7)
PA: 42.7% (42.5)

I'm learning a lot and have made significant improvements to my model.

Notably I've added a turnout variable - and am assuming that turnout will be proportional to Obama's presidential vote in 2012. While this is likely flawed, I don't have a better idea on how to predict turnout.

I added FB likes by county. Interestingly the FB likes by state are still significant.

I created a caucus-only model which has a much smaller confidence interval for its estimates (40% of the general model's interval).

Predictions