election-transparency
election-transparency copied to clipboard
Data4Democracy
Build a simple model of the 2016 presidential election
We have a lot of data put together about the 2016 election, let's build some simple models trying to explain the outcomes so that we can discuss further.
Either of these notebooks would be a good start (if you are working in R):
https://github.com/Data4Democracy/election-transparency/blob/master/notebooks/r-notebooks/model_2016_presresults.Rmd
or
https://github.com/Data4Democracy/election-transparency/blob/master/notebooks/r-notebooks/basic-viz.Rmd
Great @jenniferthompson! This is a big one, so others are definitely welcome, want to get several different ideas down and start comparing.
I'd like to jump in and do some experimenting too. Are we looking at all types of models, sticking with more simple models like logistic, or just seeing what works best?
Are there basic goodness of fit stats that we are going to look at to compare? AIC, BIC, adjusted r-squared?
@chrispelkey, I went with basic just to get people started. I would say we can start looking at all types of models. Also, take a look here near the bottom of the document to get some further ideas of what we were thinking about modeling and trying to answer. At the end of the day, we are trying to get a good model fit, but some of the most interesting information (at least for me) is going to be what are the strong predictors of a Trump or Hillary win in a county, and where do we mis-classify?
As for model comparison I am of the mind that you usually need to use more than one measure to explain the pros and cons of each methods fit. But, that is just my view of model selection, so it is always up for discussion.
+1 to @chrisdick14. My plan was to start with the predictors listed here, do a redundancy analysis and start with a logistic model for whether each county "picked the winner."
I'd be happy to work on this for a bit. I have been kind of out of the loop the past couple weeks, but I will try to get some work done on this on the coming weekend.
Great! Thanks @fhollenbach! Impossible to have too many of us working on this.
I did a first stab at building three simple logistic models. I also don't really know how to work GitHub (this is my first project), so I'm not sure if I uploaded the right way.