data4democracy / just-politics Goto Github PK

Identifying vulnerable house and senate seats in the 2018 midterm elections

just-politics's Introduction

Just Politics

Identifying vulnerable house and senate seats in the 2018 midterm elections

If the last presidential election was any indication, we need new metrics for understanding potential voting behavior. Friends close to the problem say their "experience with [traditional political wonks] has been that the answer is always that somebody tried that once and nobody paid attention and they wandered off" -- so there's plenty of room for experimentation and improvement. The Election Transparency project had success finding new and novel explanations for voter behavior in national elections, and now we'd like to try it out for senate and congressional races.

Working from a foundation of historical results and census data, we'll pull in social media data (and whatever else we come up with) to get a sense of which congressional and senate seats could flip from one party to the other in the 2018 elections.

New methodologies for identifying vulnerable races are valuable to campaigns on both sides of the aisle, and as an open, transparent community, Data for Democracy is in a great position to inform the debate. We might get it wrong, and we'll definitely be told "that's not how you do it" -- but we'll be creative, try new things, and produce something useful in the process.

Specific objectives

create a dataset of historical congressional/senate results
create a dataset of district-level census demographic/economic data
create a dataset of district-level social media comments
build models that assigns a vulnerability score to each congressional and senate race

Timeline

We'll collect data in May, and build models in June. The goal is to have a ranked list by the end of June.

For the time being, we're storing data on data.world.

Resources

Election Transparency datasets on data.world
2016 voting data for each state by district on data.world
115th Congress Member guide from Daily Kos
Dave Leip's election maps
National precinct map

Notes

Advice we've received that might be helpful:

Two reasons a seat might be vulnerable are 1) it has a lot of high-propensity moderates, 2) it has a high-concentraction of low-propensity voters of the power that isn't currently in power. It's important to understand what type of state/district you have. In scenario 1, the opposition campaign would be about persuading moderates, and in scenario 2, it's a "get out the vote" campaign.

Some states/districts (northeast for Ds, southeast for Rs), are unlikely to flip. Conventional wisdom says the best place to look is the midwest.

How to get involved

For now, join the #p-just-politics channel on Slack and ping @jonathon.

just-politics's People

Stargazers

Watchers

Forkers

tejasrd tejasdessai scotttarlow acompa nfreeda

just-politics's Issues

Create dataset of representative voting records

This dataset will list how each congressional representative and senator voted in the house and senate, respectively.

Ultimately, the goal is to see how each candidate's voting record syncs with public opinion, and whether or not that's a useful signal in determining the vulnerability of a house or senate seat in midterm elections.

List of county-specific Twitter users

This is an unglamorous but important task. With a list of county-level Twitter users we can build networks around those accounts, and then inspect the language in the tweets from those networks using some techniques that are pretty good at revealing sentiment on key campaign issues.

Our hypothesis is that understanding this "hyper-local" sentiment will be important in predicting which congressional districts could make surprising switches from one party to another.

Collect FB page comments for each Senator and Member of Congress

This issue is the first step for using social media sentiment to assess the popular support (or lack thereof) for a given Member of Congress and Senator, and subsequently testing whether that support score is a useful signal in assessing the vulnerability of their seat.

List of county-specific Facebook pages

This is an unglamorous but important task. With a list of county-level Facebook pages we can collect the comments form those pages, and then inspect the language in those comments using some techniques that are pretty good at revealing sentiment on key campaign issues.

Our hypothesis is that understanding this "hyper-local" sentiment will be important in predicting which congressional districts could make surprising switches from one party to another.

Add SDT (demographics) component to CountyCharacteristics Dataset

There is an "SDT" component described in this paper: http://www.psc.isr.umich.edu/pubs/pdf/rr17-880.pdf

We'd like to add that characteristic to the CountyCharacteristics dataset created during the Election Transparency project, found here: https://data.world/data4democracy/election-transparency/file/CountyCharacteristics.csv

Please ping @scottcame with questions.

Dataset of county-level migration trends

This dataset, probably from the American Community Survey, will show resident and non-citizen migration patterns at the county level.

These county demographics also look like they could be useful: https://datadrivendiscovery.org/wiki/display/gov/Omnibus+List+of+Machine+Learning+Training+Resources

Combine historical congressional election results with Daily Kos 2016

The Election Transparency project compiled a list of county-level congressional results dating back to 1980, up to 2014. https://data.world/data4democracy/election-transparency/file/HouseElectionResults1980to2014.csv

Daily Kos has election congressional election results for 2016 https://docs.google.com/spreadsheets/d/1oRl7vxEJUUDWJCyrjo62cELJD2ONIVl-D9TSUKiK9jk/edit#gid=1163624377

We'd like to combine these two datasets so we have consistent county-level results through 2016.

Dataset of prior + upcoming gubernatorial elections

There are 36 [state] governorships up for grabs in Nov 2018, and two others (VA and NJ) later this year. We would like to compile a dataset of historical gubernatorial election data in order to identify potentially "vulnerable" governorships.

Current status:

@scottcame is checking with @dwillis whether Open Elections has these data, or if they know where to get it.
We may be able to purchase these data from http://uselectionatlas.org/BOTTOM/store_data.php

Some notes to keep in mind for further analyses:

Certain states impose term limits on governors. This will be important to make note of.
There are also governors of US territories. Current plan is not to track those, but we could if someone feels strongly!

Collect Twitter content for each Senator and Member of Congress

Initial "seat vulnerability" model

This is a first pass at a model that will eventually include more demographic, lifestyle, and social media sentiment data. But to get us started:

Create a table of seats that changed hands from one party to another using historical election results: https://data.world/data4democracy/just-politics
Combine the table from #1 with county-level data collected in the #election-transparency project (https://data.world/data4democracy/election-transparency), including CountyCharacteristics.csv, demo_ACS2015.csv, employment_2015ACS.csv, PartyRegistration.csv
Experiment with modeling that predicts when seats with switch from one party to another, push the model in a Notebook to this GH repo, and report back to the group about what additional datasets might be useful.