Voting in a Flawed System
Yes, our current elections system is deeply flawed and transparency sucks. But who benefits when Republicans don’t vote?
This article is part of a pair. See Steve Umfleet’s article on what a Cast Vote Record is and how analysis of it sheds light on possible voting system irregularities.
Having been a strong advocate for hand counting, and having lived through a court battle for election records, this is an article I never thought I’d write. I am not advocating for the current paradigm of universal mail in voting and mandated use of computerized voting systems. There should be change to every aspect of our elections - from voter rolls and mail-in voting, to the voting “machines”, and transparency in election records. However, over time, I have had to reconcile my experience serving in the Nevada County Republican Party with research and theories coming from election integrity circles. My experience has led me to take a more moderate and objective stance on computerized voting systems, and to question election integrity influencers who are broadcasting a message of hopelessness.
X posts from “The Professor” David Clements, an election integrity influencer who broadcasts that voters “can’t outvote the machines.”
As election integrity officer for the Nevada County Republican Party, I spoke with many voters who were jaded and hesitant to vote because they had internalized the narrative that “you can’t outvote the machines”. One thing I can promise a voter - no matter how bad they fear that fraud could be - is that if they do not cast their ballot, their vote will not be counted.
Election Results Should Be Consistent with Voter Registration Numbers
We have reason to continue to optimistically engage in our political system. I have seen firsthand how old fashioned “Get Out The Vote” (GOTV) efforts can and do move the needle for elections. For example:
November 2024: Nevada Union Joint High School District school board - Republican-endorsed candidates won by wide margins in hotly contested high school board races despite opposition by the teachers union.
November 2024: The efforts of the Nevada County Republican Party to promote early and in-person voting had a predictable and measurable impact on the outcome, with election night results indicating Trump leading Harris by 2,375 votes in in-person voting. In other words, Republicans’ in person votes were seen in the results by over a 2:1 margin (election night results include most, if not all, of the in-person voting).
November 2024: Republican candidate in Grass Valley City Council race edged out Democrat incumbent by 120 votes in a voting jurisdiction with a strong Democrat registration advantage. The Republican candidate benefited from the local party endorsement and strong GOTV efforts which had a predictable, measurable effect on the outcome.
March 2024 Primary: Locally-endorsed candidate for Assembly District 1 (Tenessa Audette) edged out other Republican challengers. The local endorsement and GOTV efforts of the Nevada County Republican Party for this candidate had a predictable and measurable effect on the election outcome.
March 2024 Primary: Measure B - City of Grass Valley sales tax increase passed by only 20 votes in the Grass Valley City limits, which has a strong Democrat registration advantage. The measure encountered opposition by voters reluctant to pay higher sales taxes, and the Nevada County Republican Party urged a “No” vote on its voter guide. Voter sentiment and campaign advocacy against tax increases had a measurable effect on an election contest.
Consider these other examples of election outcomes that are consistent with voter registration demographics: In the March 2024 primary election in Nevada County, the county supervisor race in a district that has a strong Democrat registration advantage was predictably won by the liberal democrat candidate (Heidi Hall, District 1). Meanwhile, in District 2, candidate Robb Tucker prevailed thanks to the strong Republican registration advantage. Tucker was also endorsed by the Nevada County Republican Party and led a strong, organized campaign, all factors that led to an outcome that was consistent with “real world” circumstances.
Is the party of a winning candidate consistent with the party that has the voter registration lead? Did the winning candidate (or ballot measure) benefit from a strong campaign? If the answer is yes, then the outcome is consistent with the preference of the voting population, and it was likely not determined by a nefarious software algorithm or vote flipping by an international internet scheme.
Voters know their county best. Applying information about campaigns and candidates (or measures) is an important step towards making conclusions about election results. While data scientists look at a sanitized, numbers-based, “30,000 foot” view of election results, county residents can unite election results with specific, local details. Helpful resources for this research would be county political parties and the elections office.
If people choose to “throw up their hands” in resignation or spend their time learning how to hand count ballots at seminars in the critical days before an election instead of doing GOTV (Get Out the Vote) efforts, it could mean the difference of just enough votes for a good local candidate, or even whether a county “flips” from one party to another.
“Cast Vote Record” Analyses Yield Troubling Results
The Cast Vote Record, or CVR, is a digital record of all choices on every ballot. In the event of a recount or audit, the clerk could compare the ballot-specific information on the CVR to the paper ballot and its corresponding digital image.
Many government officials denied public records requests for this data after 2020, even though it contained no information that could identify a voter or compromise the security of the voting system (per California certification standards).
For clarity, the CVR does not reflect the order in which ballots were received at the elections office, and the sequential batch numbers do not reflect the “organic” behavior of the voting population. Vote-by-mail ballots are sorted by precinct and grouped into batches before they are ever introduced into the voting system. “Batch 1” will be the first batch scanned in and reflects a higher rate of return for that particular precinct. “Batch 1” will not be the first ballots received at the elections office, in the order in which they were received.
After the 2020 election, many people started looking into the data behind election results. Data analysts started to look at “Cast Vote Record” files and found patterns that seem suspicious. Steve Umfleet, a data analyst, makes an argument for how such patterns could indicate fraud in elections.
Data analysts looking at the CVRs point to patterns that are similar, regardless of voting system software, county, or election contest. They note that the behavior of the graph of CVR data does not follow the “Law of Large Numbers”. This is a compelling argument and any intellectually honest county clerk or Secretary of State should take it seriously and investigate further. Does the voting system count match what is on the paper ballots? That would require auditing a full election, or at least one contest, which is almost never done.
In counties where mail-in ballots are pre-sorted by precinct, most batches will be homogeneous, that is, a batch is made up of ballots from just one precinct (this is due to the fact that ballots are sorted by precinct as they arrive). However, batches that are scanned towards the end of the election will be made up of a mix of precincts, because the ballots are arriving in smaller numbers, and they are being “cured” by the voter (a process where the voter needs to verify their signature before the ballot can be counted). For convenience, these late ballots from different precincts are often grouped together in one batch and scanned into the voting system. This phenomenon could factor into the “end behavior” of a CVR plot.
Fraudulent algorithm or bogus ballot box stuffing? Choose one.
If voting system software or remote Internet-based interference can alter any outcome algorithmically then why bother with fraudulent ballots received by mail or drop boxes? In other words, if the software is just “manufacturing” votes and not tabulating what is on the paper ballots, then why bother injecting fraudulent paper ballots with the desired votes? These arguments are conflicting.
Consider this example of a CVR plot of data from San Francisco county for the 2020 election.
Cumulative Count of Ballots vs. Biden-to-Trump Ratio. Graph reproduced with permission from Steve Umfleet.
On the horizontal axis, we see the “normalized” count of ballots. This is the fraction of ballots counted over the entire election. On the vertical axis, we see the “Biden to Trump” ratio. This means that for a set number of consecutive ballots, as represented on the raw CVR data, the number of Biden votes is divided by the number of Trump votes. For example, a ratio of “10” on the vertical axis means that there are 10 Biden votes to every 1 Trump vote.
Near the middle of the graph, corresponding to the point when about 50% of all ballots are counted, we see a large spike in the vertical direction, meaning that the Biden-to-Trump ratio increases dramatically for that series of ballots - up to about 23 Biden votes to 1 Trump vote.
Some people will see this spike as an indicator of fraud - either by nefarious algorithms, remote interference, or a “ballot dump” of dubiously sourced paper ballots all with votes for Biden.
There are other real world factors to consider that could explain this peak. The first is that this is San Francisco county! Can we reasonably conclude that there are some areas of this county where there are 20 Democrats to every 1 Republican? In 2020, Democrat registrations in San Francisco county outnumbered Republican ones by a factor of 10 to 1.
Second, this peak occurs at the 50% mark. On the Thursday after election day, the elections office receives a huge number of last minute ballots that were mailed on election day (the Tuesday prior). The ballots received by mail after election day often double the total number of ballots cast up to that point.
This peak of votes for Biden could have come from a large precinct that experienced high turnout, an organized ballot harvesting operation, or both. It is also possible that there was a large number of fraudulent mail-in ballots corresponding to a large precinct sent to the elections office. Again, research into the precincts represented in these batches and any “boots on the ground” GOTV efforts by political groups could clarify if this explanation is a valid alternative to an assumption of fraud.
An Argument With No Possibility of Resolution
We have two opposing sides to the election integrity argument. On one side, data scientists are providing credible evidence based on CVR data that points to possible algorithmic control of election outcomes. On the other hand, election officials ignore or belittle these findings and claim that the voting systems they certify count votes with 100% accuracy. The solution to this argument could be found in the source code of the software in the voting systems. Bear in mind that neither your state’s secretary of state nor your county clerk has the training or ability to do a line by line review of this source code. County election clerks are not data scientists. Just like the public, they have to trust the two testing laboratories that are responsible for reviewing the code.
A theory can and should always be questioned and tested. If source code were open sourced and available to the public then theories and assumptions of fraudulent interference in elections could be either proven or debunked. Unfortunately that is not possible, because the voting system source code is “proprietary” and kept from public view. It is shameful that both state and federal laws allow for voting system source code to be hidden from the public. Everything this software does should be subject to public scrutiny. Our entire election process, indeed our entire government, hinges upon the results produced by secretive code. This lack of transparency should bother voters from all political parties.
Until voting system source code is made public, both sides of this argument will go unresolved. It is only natural that people will continue to make assumptions and develop theories to explain their findings. A government that truly serves its voters would give them all of the information they need to verify any election outcome.
Recently, the California legislature and Secretary of State passed laws and regulations that restrict access to election records, proving their commitment to opposing transparency.
Why aren’t any discrepancies caught in the 1% audit?
Another argument against “machine” fraud is that the 1% Manual Tally would catch at least some discrepancies between the machine count and the original paper ballots. The 1% manual tally is a process performed by hand counting paper ballots to verify that the hand count matches the machine count. You can observe the 1% manual tally at your county’s elections office, and you will see for yourself that the hand count of the original paper ballot does match the machine count. If fraud were “widespread and systemic” as election integrity influencers believe, then we should expect to see the manual tally catch at least some discrepancies more regularly.
A larger percentage audit (like, 2% or even 10%) would go a long way toward resolving the reservation that the 1% sample is not large enough to catch discrepancies, but clerks and Secretaries of State disregard anyone asking for this kind of reform. Imagine if the IRS performed an audit of just 1% of a business’s financial records. Is that ever done? Yet an audit of 1% of the paper ballots is considered sufficient to verify our elections.
Move Forward, Cautiously
Hyperfocusing on voting systems as a source of possible fraud distracts from other major issues. Look to Trump’s EO on election integrity (hotly contested by Democrats) and the ACE Act to get an idea of where Trump and congress have identified problems that need fixing. While the prospect of a federal voter ID law seems to be making progress, another issue to consider is how many voters are not receiving their vote-by-mail ballots, and is this a source of voter suppression?
We should continue to insist on transparency in election records, accurate voter rolls, and election systems that use only open-sourced code. I am not advocating for the current paradigm. But until it changes, we should never forsake our constitutional right to vote, or worse, be manipulated into hopelessness. Let’s just say for the sake of argument that there is systemic fraud. Lack of legitimate votes would only make that fraud easier to pull off.