Limitations of High-level Analyses : After the Dr Shiva presentation, there has been a lot of effort dedicated to to understanding the pattern he showed in his graph. Many tweets speculating about the reason such a pattern might be true / reasonable.
2. Some have compared the graphs across elections (2016 and 2020) to understand how the trends compare and what the differences may mean in terms of the voters. While these analyses are interesting from an academic perspective, we should ask what is actually being compared?
3. Let's take a step back from voting results analyses, and consider what these results are and what they represent. Using a random county as an example, an analysis of this county's precincts will show us the outlines / broad pattern of the votes tabulated for the county.
4. Tabulated! Not how they were cast (intended). This is a critical aspect that is must be considered in light of the shifting analysis I shared. We don't truly know the votes on the ballots, we only know how the ballots were tabulated.
5. This is why Dr Shiva uses a rough approximation using straight ticket voters. The concern of manipulation requires a substitute / alternate guide for what might be reasonable. His analysis showed the impact of this manipulation by making an axis the % of Rs in a precinct.
6. The trend in the graph only shows the outlines of the impact of the manipulation. While this is interesting, it's not possible to determine the overall / cumulative impact of the manipulation. Since our hypothetical county has different methods for voting, we needto understand
7. the differences of the manipulation for each type of method to accurately calculate the impact. The granularity of data necessary to to do this is rarely publicly available (e.g vote type by machine counter by batch). In VA, we can derive it to some extent using the Change log
8. With these individual batches, we clearly see a systematic pattern of manipulation. This can be potentially reverse engineered to understand how the process works. My working theory is it's self contained within the machine, using two variable :
9. 1st Var = number of ballots scanned. 2nd Var = lopsidedness of those votes (% of Ds vs Rs). This second variable would implicitly reflect the same axis Dr Shiva used to plot the pattern of manipulation he detected (%R in a precinct). We will need to test this theory.
10. Regardless, if we can determine a formula, we can use this to extrapolate the effect of the manipulation votes, which could give us an estimate of the total. But without the granular data, determining the impact can only be estimated with rough substitutes. One option would
11. be to compare unaffected and affected counties to get an approximation. It's less precise, but is the only alternative without the batch-level data. The problem is that a county could have 50% of it's votes affected and the other 50% not be. It could be 20% / 80%.
12. So, when doing comparisons between counties, the inferences that are drawn from the differences are themselves unclear without the knowledge of the processes that feed the overall totals that are being analyzed. Well, maybe we should compare the same county btwn elections?
13. If both elections were affected, a comparison will only show the different outlines of the manipulation. But questions linger : what is the same formula for the manipulation? what the % of votes affected the same? Are differences between the elections the result of
14. different turnout, diff voter preference, diff voting methods, etc. These are irrelevant to the core question of how much manipulation occurred. This is the crux of the problem, and with out the proper data, it's all speculation. This is why these high-level analyses are
15. limited in their usefulness in understanding the impact. In fact, they distract from the core question / issue, by accepting that the data is a true reflection of how the ballots were cast, rather than how they were tabulated. The most powerful aspect of these analyses is to
16. show there is an unnatural result that is systematically occurring. Anything further is speculative and of little use. Further scrutiny of the results of the election is needed, but unless they are used to determine an estimated impact, these only serve tobe discussion pieces
17. This weekend I'll be checking for the symmetric pattern in the 2016 data. This will help us establish if this existed before 2020. If I have time, I'll do 2012 as well. If all 3 elections have the pattern, I think understanding / developing a formula for the manipulation is
18. about as good as we will get to real answers. As for split ticket Rs in this election, and why these voters are the reason for this anomalous behavior. This is doesn't pass some sanity checks. ~95%s support from Rs, but they split off 10-20% in the batches? No way.
19. Or how about that many Rs don't like the Prez on a personal level. Sure, this is probably true. But that doesn't mean they wouldn't vote the man, especially considering the alternative. This argument mixes personal feeling with how the electorate would vote. Not reasonable.
20. These positions really are just trying to sidestep the core issue. The manipulation. These serves to explain / defend / justify the manipulation rather than address the problem of election interference. Don't fall for it!
@threadreaderapp unroll
• • •
Missing some Tweet in this thread? You can try to
force a refresh
