This is a script that'll run rcv cruncher on all the elections you have in the cvr folder, and summarize the results in a csv file (assuming you specify an output file, otherwise it'll output a json blob to the console).
You can find descriptions of most of the stats here.
Here's additional fields that this computes:
- competitive_ratio: This is defined as the ratio in first place votes between the 3rd place candidate and the 1st place candidate. It's useful for comparing to results from this paper
This project assumes you download CVRs from here and put them in a folder called cvr (I have Moab 2021 committed to the repo as an example)
https://data.census.gov/table?g=050XX00US06001$1000000&y=2020 https://statewidedatabase.org/d20/g22_geo_conv.html https://mapshaper.org/ https://data.acgov.org/datasets/5c2a208663ec40d8aa18bfe65ed3a32f_0/explore?location=37.679110%2C-121.907993%2C10.75 https://data.acgov.org/datasets/3d3205bb21904c3db4f8597e1c55cc5e_0/explore?location=37.805475%2C-121.657363%2C9.87
Here's information for each of the output files (the date corresponds to the date the CVRs were imported from Dataverse, not the date the simulation was ran).
The outputs have also been posted as a Google spreadsheet
I collected all the single and sequential IRV datafiles and ran single-winner IRV on all of them.
There were 455 elections in the dataset, however several files were bugged, and I added Aspen manually. My final spreadsheet had 448 entries (so it seems like there's 2 other elections that I'm missing, not sure where those went)
Bugged Files
- Vineyard_11052019_CityCouncil_tab1.csv
- Vineyard_11052019_CityCouncil_tab2.csv
- Easthampton_11022021_Mayor.csv
- Payson_11052019_CityCouncil_tab1.csv
- Payson_11052019_CityCouncil_tab2.csv
- Payson_11052019_CityCouncil_tab3.csv
It assumed the following settings. Individual elections require different settings, but these will hopefully be close enough for a starting point:
exhaust_on_duplicate_candidate_marks=False,
exhaust_on_overvote_marks=True,
exhaust_on_N_repeated_skipped_marks=0
This dataset also specifes a number of overrides for fields that aren't computed by rcv_cruncher. The overrides also include the Aspen election. This election was said to use STV, but the specific algorithm they used was closer to block-IRV, so I felt it was fair to include it here.
The FairVote RCV Cruncher tool was very helpful and gave us 99% of the data we needed.
The only other bit I'd like to include is an investigation of skipped and stalled ranks. These are situations where ranks are either skipped because the candidate was already eliminated, or stalled because the ballot had already been allocated toward one of the finalists. It would be super interesting to know what percent of ranks are counted, skipped, or stalled overall, and also know the average number of these ranks per ballot.
I'd also like to make the tool a bit easier to use. Right now it runs for an hour+ and it would be great if it could pick up where it left off if it bugged halfway through.