Read on Twitter
Pennsylvania data shows that some of the mail in ballots were returned BEFORE they were mailed, on the same day or next day. I marked such ballots as "Suspicious" (RED) in the plot. There are 109421 (3.53%) "Suspicious" ballots. @DatElefan @Peoples_Pundit @shylockh
Data is taken from here: https://data.pa.gov/Government-Efficiency-Citizen-Engagement/2020-General-Election-Mail-Ballot-Requests-Departm/mcba-yywm
If only look at those "suspicious" mail-in ballot - most of them come from PHILADELPHIA and LEHIGH - around 44%
Now let's look at Mail Application Type. OLMAILV-This is a mail ballot application that was submitted online. MAILIN-This is a mail-in ballot application. "Suspicious" ballots have higher ratio of Mail-In compared to Online than normal ballots.
You would expect that under normal circumstances (and if there is a perfectly legal explanation for "suspicious ballots") the turnaround time should not depend on how the ballot was requested? Also, table above is colored by %age for "% of Total(Y)" - Y for "suspicious"
Actually, if we look at distribution of mail-in vs online requests, you can see that anomaly starts for ballots that were received whole FIVE days after they were sent. This increases number of "suspicious" ballots.
Another anomaly. There is a group of around ~1000 ballots that stands out. I called them "Fast" ballots. Suddenly a group of ballots started arriving much faster than before. They are highlighted in the plot.
This is mostly due to: CVO - absentee ballot application for an overseas civilian voter. F - absentee application for an individual who qualifies to vote for federal offices in federal election years. M - absentee ballot application for a military voter.
It seems strange that military and oversees ballots suddenly started arriving THREE days faster than general population. Median turnaround for normal ballots (excluding "suspicious") is 14 days, for "fast" it is 11 days. Means are 16.3 and 12.96 days respectively.
One more observation. This time I overlaid initial plot with data density plot. You can see that the main injection of the suspicious ballots was around 10/26-10/27 (Mon-Tue) for Mailed Date and it was mostly the same day ballots. I estimate around 43k ballots in that group.
Age difference between normal and "Suspicious" ballots. In some counties "suspicious" ballots are much younger, in some - much older. What's the difference between CENTRE and UNION (younger) vs CUMBERLAND, GREENE, LEHIGH (older)? PA folks, let me know!
These FIVE counties mostly dropped their ballots as SAME DAY or close on 09/27-10/08 and then on 10/19. The 10/26-27 dump of same day ballots comes from a different place.
Again, these are the counties where "suspicious" - i.e. SAME DAY ballots, have significant demographic deviation from the rest. Specifically - much younger (Students from Penn and Bucknell?) and much older people. And it's only a subgroup of all the "suspicious" ballots.
In comments ppl say same day ballots might be due to satellite locations. Plot shows when most of same day votes was done in ALLEGHENY county (red cloud) vs when those locations were open per news link (green lines)
https://twitter.com/caityspaghetti/status/1325914977614639105
https://twitter.com/caityspaghetti/status/1325914977614639105
Folks, I just realized that my plots with data density are incorrect. I overlaid it over actually two plots: Mail vs Return and Mail vs Mail (straight diagonal line). Well, that line also gave data density. So I'm re-plotting 3 plots without this line. Update below.
First, counties where median age for "suspicious" ballots is either young or old. Here in legend first letter is if ballots are "suspicious" and then what county it belongs to: one of the normal counties, "old" county or "young" county.
Suspicious ballots in both "old" and "young" groups (red and blue colors) are tightly grouped together. And if we look at the median age for those groups, it's OLD people from "old" counties that have "suspicious" ballots and same for YOUNG people from "young" counties. Weird.
Digging further. Counties that have younger "suspicious" ballots. Lots of ballots mailed out on 10/03 and normally received reasonable time later (green). Then two "suspicious" peaks, most prominent around 10/26-27, returned SAME DAY or NEXT DAY (blue)
LEHIGH. Most of 88706 ballots were mailed on 10/19 (clouds are small, meaning all data is concentrated in those points). Most "suspicious" ballots from this group returned next day, 10/20 or two days EARLIER 10/17
CUMBERLAND. Most ballots mailed out around 10/06-08. Most of them received 10/13-17 (green cloud). But a lot also received same day (red cloud). These (red) are also older people.
ALLEGHENY. Mostly sent out ballots on 09/28 and then normally received it from two days later to month later. (green clouds) and then had three SAME or NEXT DAY drops around 10/11 (Sun), 10/18 (Sun) and 10/25-28 (Sun-Wed)
Final plot for today. Percent of "suspicious" ballots vs age by county. Red are the three counties where older people get more suspicious, blue is for young. Green and LEHIGH are record breaking for percentage of suspicious ballots, both skewed towards older people.
I apologize for such small scale on the rest - these two counties are pretty high. You can see uptick for young people on Union and Centre, also some in PHILADELPHIA. Remember, this is percentage, not absolute numbers. PHILY is pretty smooth due to large number of ballots overall
Probably this way is going to be more telling. I put a smoother instead of bar chart and marked peaks that show as suspicious in my analysis.
