Overview

Announcements

Guest lecture!

• Homework 9 due Thursday.
• Survey 9 will be released on Friday.
• Fill out the official course evaluation!
• If 80% of the class fills them out, everyone who filled them out will get an extra 1% added to their overall grade.
• 63% as of yesterday.
• Overall Score on Gradescope updated to include Quiz 3, Homework 8.
• See Practice Problems by Topic.
• No QC this week.

Elections and Data

Lakshya Jain

Terms to know

• Partisanship: The net support for a party in a county
• Swing: The change in partisan vote share/margin between two elections.
• Swing voters: Voters whose vote could tip either way for the upcoming election
• Electoral College: The number of votes each state gets in the presidential election
• Elasticity: How much do voters change their vote per election?
• Early Voting: The period before election day during which voters can cast their vote. Typically with high Democratic advantage.

Elasticity

• Do Swing Voters Exist? Where does ticket-splitting happen the most?
• Use Euclidean Distance between electoral margins to quantify the elasticity of voters in an area
• Elastic states: New Hampshire, Arizona
• Persuasion-based, lots of swing voters
• Inelastic states: North Carolina/Georgia
• Turnout-based
• Keys to winning in states differ -- swing states are not swing voters!

Some quick notes on this…

• When initially (mis)used, this metric actually didn’t work nearly as well as hoped.
• Elasticity is a great example of a concept that should make sense in theory…but in practice, it’s not good as a predictive metric.
• Main value is in explaining what happened, not what will happen.
• Key part of data analysis: Know the limitations of your data and technique. Don’t over-extrapolate.
• Just because something has past value doesn’t mean it necessarily has future predictive value. But that doesn’t make it useless!
• Elasticity can flag areas of interest. Just don’t over-extrapolate about what *will* happen.

​

​

​

​

Modeling Georgia Runoffs: Inelasticity and Suburban Swing

• Margins largely static in this state
• Turnout highly variable
• How do you win the state?
• Turn out more of your voters than the other side.
• Persuasion is not the main way to winning the state, but...important swing pockets exist
• Use elasticity to identify pockets of the state that have been swinging towards Democrats. Use them to flag areas of interest.

Suburban Swings

• Educational Polarization -- key story of 2020
• Suburban areas swung very far to the left
• Rurals swung right
• Educated suburbs powered Biden’s path to the White House
• Let’s look at case studies!

Back to Georgia…how does this help?

• Question: could suburban swing, combined with inelasticity, help Democrats win Georgia in the runoffs?
• Educated suburbs have high *margin* elasticity – they have swung heavily towards Democrats
• Democrats now bank a lot of the vote in educated suburbs.
• Rural white areas have high *turnout* elasticity in Georgia – they’re heavily Republican, but the amount of people showing up to vote varies greatly
• Could this mean that more Democrats show up in the runoffs?

Georgia Modeling

• Goal: Model Runoff Electorate
• Problem: US Polling never really did what it was supposed to, missed horribly
• How do we build a model without polls?
• What is a poll really trying to do?
• Capture the sentiment of the electorate at a given moment in time
• Georgia: Highly inelastic state; few swing voters
• Decided to rely on November margins -- as accurate as polls can get!
• Geographic Reprojection did not capture variance accurately
• Race and Precinct based adjustment, made with help of Nate Cohn of NYT
• No way of predicting election day turnout, so...
• Friends helped present breakeven points instead to communicate possible outcomes

Chart by Jeff Williamowsky (@JDubWub)

KEY POINT: Make things easy for viewers to digest quickly. Highlight, shade, make neat tables

Only county missed by party: Dooly

National Election Modeling: Explaining 2020

• Narrative: 2020 saw some crazy results!
• Florida got redder
• Texas got bluer than Ohio and Iowa (Obama won Iowa by 6 and lost Texas by 17 in 2012…)
• Georgia (!) went blue before North Carolina, Florida, Ohio, Iowa
• Can you explain the variance in 2020 results with just the underlying demographics and past partisanship?
• How much of the 2020 election can be explained with the national environment and demographics?
• Surprisingly, a lot! Variance in results almost entirely explained by demographics.
• R^2 = 0.99. Average County Error: ~2.4%
• Be wary of what narratives say.
• Demographics explain a lot of what went on in the 2020 election, and the results are far more explainable than the initial media narrative suggested.

Results

Plotting Exercise!

• Jupyter notebook time!