Analyzing Georgia Runoff NYT (and other) Data

Mick West

Administrator
Staff member
2021-01-05_17-07-22.jpg

The data used by the New York Times to display their interactive page of results was both useful and problematic in the 2020 election. But analyses of it existed, and were the basis of some conpiracy theories and rebuttals.

It's not clear how close the the Georgia run-off election will be, but in the somewhat inevitable situation that there might be some questions, I'm copying the data at one minute intervals for future analysis.

The data seems rather different to the general election. In particular there is per-county data broken down into various subset of actual vote counts. Previously we simply had a vote total for the entire state, and percentage breakdowns.

I'm hopeful this data will be useful in throwing light both on how this run-off election progressed, but also how the count might have progressed in the general election.

However, for now I'm just going to keep downloading, and see what issues people raise.
 
If you want the official source for the unoffical results, I believe it's at https://results.enr.clarityelections.com/GA/107556/web.264614/#/summary

Article:
For the 2020 US General Election, Scytl provided 4 products and services to city, county, and state clients across the United States. Each of these products and services provide information created and approved by each individual client, to voters or temporary election workers.
  1. Election Night Reporting (ENR): ENR is a platform that provides visual representation of votes that have been tabulated by local elections officials. ENR allows elections offices to provide real-time election results to citizens in a multi-lingual, mobile friendly platform. On Election Night, each election office tabulates the vote using their tabulation vendor’s physical machines and uploads those results to Scytl’s ENR. All Scytl ENR servers being used for the US elections are physically located in the United States. Scytl does NOT tabulate, tally or count any votes.
 
Last edited:
you can just follow AP on google. its clearer to see and automatically updates (you dont have to refresh browser or anything)
1609899861567.png
 
fair enough. thought you might want to see the actual numbers. sorry.
The clarityelections site shows the "actual numbers", shows how many precincts are reporting, and lets you download the data.

I expect that the youtube livestream will be available later, so you can go back to it when you want to quickly see how the reported tally developed over the course of the night. (That said, this stream is now showing an empty stage and no bar graphs.)
 
Thanks Mick,

Great site. I found it while searching for GA Runoff timeseries data, which it appears you’ve managed. Having spent MUCH time analyzing the frustratingly misunderstood November 2020 files, from which sprouted all manner of bad conclusions, I’m elated to read that this time out we’re getting the ACTUAL vote allocations (and at the county level) rather than the total count and absurdly imprecise percentages of the so called “Edison Files.”

BTW, based on your screen grab you’re pulling the JSON from https://Static01.nyt.com/elections-...odel/2021-01-05/senate/ga-s-s-2021-01-05.json. I’m drawing a blank at that URL – privileges issue?

Thanks.
 
There is a vote flipping conspiracy circulating for the Georgia run off. As in the general election, it is based on watching count updates in the media.

At 2:05 in the video, made on election night, you can see the total for Perdue dropping from 774,723 to 742,323.

Brad, the Triumph News Network host asks, "How do you lose 32,400 votes out of thin air?"

The conclusion that is drawn is that it is, "flat out cheating." He then goes on to make some ominous comments about the DC rally. But, that seems like another thread.

Mick, does this drop exist in the underlying data stream you recorded from Georgia? Does it exist in the NYTimes data?




Screen Shot 2021-01-08 at 1.38.15 PM.pngScreen Shot 2021-01-08 at 1.38.58 PM.png
The total number of votes for Perdue drops by 32,400 at 2:05 in the video.
 
There is a vote flipping conspiracy circulating for the Georgia run off. As in the general election, it is based on watching count updates in the media.

At 2:05 in the video, made on election night, you can see the total for Perdue dropping from 774,723 to 742,323.

Brad, the Triumph News Network host asks, "How do you lose 32,400 votes out of thin air?"

The conclusion that is drawn is that it is, "flat out cheating." He then goes on to make some ominous comments about the DC rally. But, that seems like another thread.

Mick, does this drop exist in the underlying data stream you recorded from Georgia? Does it exist in the NYTimes data?
Unfortunately, the insanity in DC last week and subsequent insane responses throughout the nation have conveniently obscured this story, effectively relegating it to the tin-foil-hat-zone with ZERO investigation. I haven’t managed to locate the time-series data IN ANY FORMAT for analysis and will reserve judgment until then. That said -- This “shut up and sit down” climate won’t last long and the truth (AND THE DATA) will eventually out. Pendulum swings.
 
I haven’t managed to locate the time-series data IN ANY FORMAT for analysis
Here:
https://static01.nyt.com/elections-assets/2020/data/liveModel/2021-01-05/senate/summary.json
Attached, reformatted.

It has a data structure with two races, so ["races"][0 or 1]["timeseries"], then ["vote_counted"] - I'm a little unsure then, as it has ["democrat_voteshare"] and ["republican_voteshare"], but they both have five entries, like:
Code:
"republican_voteshare": [
   0.4861,
   0.4885,
   0.4902,
   0.4919,
   0.4942
],
I'm not quite sure what to make of that.

I've also got the other data I described above, unfortunately missing a chunk, but still might be useful. Recent events have distracted me though. Here it is if someone wants to take a look. There's two different files, I think ...GEORGIA... is a subset of ...SENATE...

https://www.metabunk.org/f/nyt-periodic.zip
 

Attachments

  • summary-Jan-12-2020.json.zip
    124 KB · Views: 386
effectively relegating it to the tin-foil-hat-zone with ZERO investigation
That's a false claim. I googled "vote drop georgia runoff election fact check" and found several articles addressing this issue. It should become obvious from the data that this was a quick up-and-down vote movement within a few minutes; and that you're only shown half of this movement by those peddling propganda. (The same tactic is used to sell you stocks, by the way. Always look at the data from before the cutoff point they show you!)
Article:
The scrolling statewide vote count at the bottom of the screen on both networks had shown Perdue’s tally at 774,723, before dropping to 742,323 — a difference of 32,400. “MORE ELECTION FRAUD: 32,400 Votes Removed from Senator Perdue’s Vote Tally Live on TV,” a Gateway Pundit headline said.

That change, though, was actually a correction of faulty data for one county.

Information from Bibb County had been logged in wrong by a data entry worker, explained Rob Farbman, executive vice president for Edison Research, which provides vote tabulation data to the networks.

Shortly after 8 p.m. on Election Day, vote totals for Ossoff and Perdue were 27,986 and 12,496, respectively, Farbman wrote in an email to FactCheck.org. Less than 10 minutes later, the feed Edison uses reported that Ossoff’s vote total had remained the same and incorrectly reported that Perdue’s had increased by 32,400 to 44,896.

“This error was obvious to our quality control team and corrected within 5 minutes. The initial error accounts for the brief additional 32k votes for Perdue. The vote drop reflected the correct vote,” Farbman wrote in the email.
 
I haven’t managed to locate the time-series data IN ANY FORMAT for analysis and will reserve judgment until then.

I found this scraped data source: https://github.com/Ne02012/NYT-2021-GA-Sen-Runoff-Vote-Tracker

This JSON file has the vote totals for each candidate and also a breakdown by county, but it does not have time series data.
data.races[1].candidates[0].votes for Ossof, and data.races[1].candidates[1].votes for Perdue

The time series would have to reconstructed from the git history. I would like to have a graph of the spike, if it's in this data. Bonus for a county breakdown, since that would confirm the Bibb County explanation from FactCheck.

It should become obvious from the data that this was a quick up-and-down vote movement within a few minutes

I think that's exactly right, and the explanation on Fact Check seems quite reasonable.
 
I found this scraped data source: https://github.com/Ne02012/NYT-2021-GA-Sen-Runoff-Vote-Tracker
...
I would like to have a graph of the spike, if it's in this data.

Narrator: it was not in the data.
Georgia Senate Runoff Vote Count.png
I've combed through the various NYT data sets, and best I can tell, none have the vote count drop. They may have caught the error before it went out, or it was corrected between data scrapes.

But, since ABC and CNN both showed the drop in votes, it stands to reason that they would show the bump up, before the drop.

Sure enough, at 1:17 in this archive of the ABC election night broadcast, we can see Perdue's number jump up while Ossoff's count stays the same.

At 1:17:02, we see that Perdue has 725,898, and Ossoff has 924,26. In the video, it cycles to the Warnock/Loeffler race, then when it returns at 1:17:16, we see that Perdue has 758,298, and Ossoff still only has 924,26. A jump of 32,400 votes.




Then, at 1:21, in the same video, we see those 32,400 votes taken back off.




So, the feed that ABC uses was corrected in about 4 minutes.

Which is consistent with the FactCheck article which quotes Rob Farbman of Edison Research.
“This error was obvious to our quality control team and corrected within 5 minutes. The initial error accounts for the brief additional 32k votes for Perdue. The vote drop reflected the correct vote,” Farbman wrote in the email.
Content from External Source
 
Interesting and useful data set and visualization here:
https://www.nytimes.com/interactive/2021/upshot/2020-election-map.html

Has precinct-level results on a map, and also as a dataset for 2020
https://int.nyt.com/newsgraphics/elections/map-data/2020/national/precincts-with-results.geojson.gz
(also uploaded to https://www.metabunk.org/f/precincts-with-results.geojson.gz , for posterity)

Properties on each precinct polygon:

GEOID: unique identifier for the precinct, formed from the five-digit county FIPS code followed by the precinct name/ID (eg, 30003-08 or 39091-WEST MANSFIELD)
votes_dem: votes received by Joseph Biden
votes_rep: votes received by Donald Trump
votes_total: total votes in the precinct, including for third-party candidates and write-ins
votes_per_sqkm: total votes divided by the area of the precinct, rounded to one decimal place
pct_dem_lead: (votes_dem - votes_rep) / (votes_dem + votes_rep), rounded to one decimal place (eg, -21.3)

Content from External Source
County FIPS codes can be found here:
https://www.nrcs.usda.gov/wps/portal/nrcs/detail/national/home/?cid=nrcs143_013697

The first two digits is a state identifier, with 13 being Georgia. So you can extract all the 13XXX-????? precincts for Georgia.
 
Last edited:
Back
Top