https://github.com/washingtonpost/2020-election-night-model

2020-election-night-model
https://github.com/washingtonpost/2020-election-night-model

Last synced: 5 days ago
JSON representation

2020-election-night-model

• Host: GitHub
• URL: https://github.com/washingtonpost/2020-election-night-model
• Owner: washingtonpost
• License: mit
• Created: 2020-10-19T17:28:16.000Z (about 4 years ago)
• Default Branch: main
• Last Pushed: 2021-01-04T01:06:01.000Z (almost 4 years ago)
• Last Synced: 2024-10-27T14:30:58.724Z (18 days ago)
• Language: R
• Size: 104 KB
• Stars: 58
• Watchers: 4
• Forks: 11
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Election-Night Model - 2020 General Election

This is The Washington Post's general election night model. The model was created in conjunction with Decision Desk HQ/0ptimus.

## Model

The model uses quantile regression for point prediction and quantile regression + conformal prediction to generate prediction intervals. You can read about the methods employed [here](https://elex-models-prod.s3.amazonaws.com/2020-general/write-up/election_model_writeup.pdf) or for a less technical read see [here](https://washpost.engineering/2020/10/22/how-the-washington-post-estimates-outstanding-votes-for-the-2020-presidential-election/). 

## Data

The model in production uses data collected by Decision Desk HQ/0ptimus and The Washington Post. 

To see how the model works we have included county-level 2012 and 2016 election results for Georgia, Kansas, Kentucky, Missouri and Texas. The data was taken from the [MIT Elections Lab](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/VOQCHQ).

We were unable to include the county level data we use in the actual model. Instead this model uses a state level fixed effect and a national fixed effect *only*. You will see, however, that it still performs well.