I was trying to decide which House races were actually close enough that my time or money could make a difference. So I parsed FiveThirtyEight's assessments, parsed Cook's assessments, and wrote some code to join the two datasets together. Jeff Leek also added information about the fundraising data in close races (source: Q2 Fundraising Data), which is useful if you want to know who actually needs money and who has a lot already. (To my knowledge, this combined data doesn't exist in easily sortable/parseable form). This is current as of 8/17/18. Click on combined_data.csv to see the results + download them.
Important update: Wikipedia, bless them, implemented a version of this idea which will likely be better and more current! See their table here.
- combined_data.csv is the combined data file. It is sorted so the races FiveThirtyEight predicts will be closest are at the top. You can load it into Python/R/Excel and parse it however you choose. Column names:
- district: the name of the district
- name: 538's name for the incumbent
- 538_democrat_wins: the percentage chance the Democrat wins, according to the FiveThirtyEight model (0 - 100).
- 538_diff_from_even: the absolute difference between 538_democrat_wins and 50. Table is sorted by this column.
- cook_name: Cook's name -- just used this so people can see if something went wrong with the data parsing.
- cook_pvi_rating: Cook's PVI rating -- negative values indicate Republican is winning.
- cook_category -- the category Cook puts the race in
For example, the top of the table looks like
- cook_original_data.txt and fivethirtyeight_original_data.txt are my versions of the original data.
- goodbye_stranger.ipynb contains the parsing + analysis code (that is just the song I was listening to when I wrote it).
Here's a scatterplot of Cook's PVI's versus FiveThirtyEight's probabilities.
Please contact emmap1@cs.stanford.edu with any comments, questions, or errors.
Please contact jtleek@gmail.com with comments/questions or errors
- fundraise_2018_dailykos.csv - fundraising data for close races from dailykos with columns
- state: state of the raise
- district: district within that state
- name: name of the candidate
- party: original party information from DK - this is tidied in later columns
- q2_raised through transfer: the amounts raised or on hand (I think ) for q2
- committee_id/candidate_id: I think these are DK ids maybe?
- incumbant - inferred incumbant status from the party variable
- party_ind - indicator of independent inferred from the party variable
- party_dem - indicator of democrat inferred from the party variable
- party_rep - indicator of republican inferred from the party variable
- pull_fundraising.Rmd - R markdown file showing the data pull
- pull_fundraising.pdf - compiled version of the data pulling code with session info