THE INCUMBENT AND THE CHALLENGER
On August 30th incumbent Florida Congresswoman and former Democratic National Committee chairwoman Debbie Wasserman Schultz faced a tough primary challenge. Her opponent, Nova Southeastern University law professor Tim Canova had managed to focus the national outrage against her following her forced resignation as chairwoman of the Democratic National Committee, and made the election a referendum on her politics, policies and ethics.
Tim Canova
Her DNC resignation followed revelations that Wasserman Schultz and other Democratic party officials were “conspiring to sabotage the campaign of Senator Bernie Sanders of Vermont." After Sanders lost the Democratic presidential primary, some considered this congressional district a proxy race, with Sanders endorsing Canova against his rival Hillary Clinton's long-time ally Wasserman Schultz. Canova translated that endorsement into a fundraising bonanza, sparking national attention by raising over 3 million dollars. Open Secrets, a non-profit that tracks money in U.S. politics, says that 100% of those donations were from individual contributions, although according to the Sun Sentinel ninety percent of the money was from outside Florida, and may not have translated into local votes.
It was a campaign of firsts: Canova’s first run for office; Wasserman Schultz’ first ever primary opponent after 6 terms in Congress, and the first time a new district map was in place. The 23rd Congressional District, which lies primarily in Broward, but also has some precincts in Miami-Dade County, was redistricted in 2015 as part of a court order that found that the districts were being drawn to favor Republicans and incumbents.
The stakes were high for both candidates and the outcome was hard to predict.
Debbie Wasserman Schultz with President Obama photo by Susan Walsh | AP
Wasserman Schultz rallied her supporters starting at the very top. President Obama endorsed her; Democratic Presidential nominee Hillary Clinton personally stopped by her office; Gabrielle Giffords appeared with her at an anti-gun violence event; civil rights legend Congressman John Lewis walked her across a bridge; and Vice President Joe Biden shared an ice cream moment with her saying, “She’s my favorite person.” But Canova adopted many of Sanders’ signature issues, and Sanders’ supporters were hungry for a new candidate.
“Gibson Lopez, 18, of Davie, attended Canova’s primary night party ... Lopez, who wore a ‘Berniecrats’ T-shirt, told Canova that he was the second person he ever voted for, after Sanders.” (Quote from the Sun Sentinel.)
There was one independent poll published during the race. The South Florida Sun Sentinel/Florida Atlantic University poll released nine days before the election had Wasserman Schultz ahead by 10 points. Not surprisingly, the candidates had starkly different perceptions of their odds. Canova's team put out an internal poll a month before the election that aligned with the Sun Sentinel's numbers. It showed Wasserman Schultz ahead by 8 points, but vulnerable with what the pollsters described as, "...a staggering decline from her popularity in past campaigns." Canova’s deputy campaign manager said that based on door to door canvassing they had done the last four weeks before the election, he thought it was going to be neck and neck. “We were identifying our supporters to her supporters five to one,” he said in a phone interview." In contrast, an internal Wasserman Schultz poll of 400 likely primary voters commissioned by the Patriot Majority PAC, showed Wasserman Schultz ahead by 33 points a month before the election. The Patriot Majority numbers were not representative of the Sun Sentinel poll. Florida Atlantic University professor Kevin Wagner, who helped conduct the Sun Sentinel poll, commented that the Wasserman Schultz survey, “...didn’t reflect what we came out with.”
On election night, Wasserman Schultz was announced the winner by a commanding lead of 13.5%. However, we have examined statistical analysis of the race from four separate analysts and after detailed demographic research have concluded that there are red flags that deserve further investigation.
STATISTICAL ANALYSIS OF THE RACE: “IMPLAUSIBLE”
In our analysis we use a method called the "Cumulative Precinct Vote Tally Chart," (CVT graph for short.) For detailed information on this type of graph please refer to our paper, An Electoral System in Crisis. Here is a brief description of the method:
The CVT graph shows the precincts added together cumulatively from the smallest precinct (the one with the least number of votes) to the largest precinct (the one with the most votes) along the X-axis (the horizontal line on the graph). On the Y-axis (the vertical line on the graph) it shows the average of each candidates’ percentage so far. Figure 1 shows a CVT graph with an expected statistical pattern.
To visualize how the graph is created, imagine the precinct with the least votes, then picture adding in a precinct with slightly more votes, then add in one with slightly more. Continue doing that until all of the vote have been added together. Each time a new precinct is added in - the total of the votes so far is charted on the X-axis and each candidates' percentage - of the votes so far - is graphed on the Y-axis. In the end, there is a graph that shows whether or not the candidates' percentages are changing from small to large precincts, and if so - by how much.
Figure 1 - The Republican Congressional primary in the 23rd district follows the expected statistical pattern.
The precincts are being added together cumulatively, so as you move further right on the graph, it becomes harder and harder for any individual precinct to overcome the average percentage of all the votes that have been added up so far. Because of that, after enough votes have been added in, we expect the data to begin to graph as a flat line. Figure 1 shows a graph of the Republican counterpart to the Wasserman Schultz/Canova race. This race follows the expected statistical pattern. Once enough votes are counted, the average level of support for each candidate is clear, and the graph settles into a relatively flat line. The more votes counted, the flatter the line becomes. This behavior is predicted by the law of large numbers, a mathematical principle that is the basis for all polling. Investopedia provides an easy to understand explanation for it, “A principle of probability and statistics which states that as a sample size grows, its mean [average] will get closer and closer to the average of the whole population.”
In other words - as the number of votes sampled increases - the expectation is that each candidate's percentage of support will get closer and closer to the average of their support in the whole population; exactly what happens in the Republican primary of the 23rd district (Figure 1).
In the race between Wasserman Schultz and Canova, that does not happen (Figure 2). As more and more votes are added in, instead of the pattern approaching the average level of support for each candidate (a flat line) Wasserman Schultz' percentage climbs continuously in a mathematically precise pattern, and Canova's percentage decreases steadily. We will discuss later in the post exactly how precise the pattern is. This is not the expected statistical pattern. This graph is by Anselmo Sampietro, a mathematician who holds a Master of Statistics degree from the University of Bologna, Italy.
Figure 2 - The Democratic Congressional primary in the 23rd district shows an unexpected statistical pattern that favors Wasserman Schultz.
To realize how unusual this is, it is important to visualize what is happening during the course of the graph. Along the Y-Axis, each point is an average of the candidate's support up to that point (using the language of the law of large numbers - it is an average of the "sample" so far.) Along the X-Axis, the votes are being added together cumulatively, so by the time the final precinct is added in, the total is no longer a "sample" of the votes - it is at that point, all the votes in the race (again, using the law of large numbers language, it is the "whole population.") The far right side of the graph is then an "average of the whole population." Why does the average of the candidate's support (average of the sample as it grows larger) not grow closer and closer to the average of all the votes in the race (average of the whole population)? The average of the whole population is actually getting further and further from the average of the sample, contradicting the law of large numbers.
Four separate analysts examined the results independently and found the same trend, so there is no question that the trend exists, only a question of what it indicates. Figure 2A, by data analyst and engineer Phil Evans, shows the same graph with an estimate of the difference between the expected statistical pattern and the reported results. He estimates the difference at 10%.
Figure 2A - There is a 10% difference between the expected statistical pattern and the results in this race.
When looking at the data, Dr. Fritz Scheuren, a member of the statistics faculty at George Washington University, said he found the results, “Potentially implausible." He continued, "What you show are results that could have been manipulated to get to this graph.” Dr.Scheuren suggested that the best way to resolve whether there is a problem with the election is to do a manual count of the ballots. “We have to find a way to find out if they were manipulated, and that requires a recount, of at least a sample of locations.” He added that it would be best to start with precincts “where the gaps are very wide,” meaning the precincts with the largest distance in percentage between the candidates. If there are differences between the manual count in those precincts and the machine count, he recommended expanding the recount to other precincts.
Figure 3 is a graph of the precincts in the district that are in Miami-Dade County. The graph is by Dr. Beth Clarkson, a quality control engineer with a doctorate in statistics. It shows Canova’s percentage of the vote decreasing as precincts with larger numbers of votes are added in. It demonstrates that the trend exists in the Miami-Dade precincts as well as in the district as a whole.
Figure 3 - The pattern where Canova's percentage decreases as the precincts get larger is evident in both Miami-Dade and Broward County. Here the pattern is shown in the Miami-Dade precincts.
WHAT IS CAUSING THE PATTERN?
There are three general explanations for the pattern.
- It could be caused by an error in the computation of the results.
- It could be due to demographic differences between the precincts. If the larger precincts had a much larger representation of a group that supports one of the candidates - it is possible that it could explain the increase in one candidate's percentage. The larger demographic group would need to be much larger than the actual increase, since no group experiences 100% turnout and it is unlikely that any candidate receives 100% support from a community.
- It could be caused by manipulation of the results.
POSSIBILITY OF ERROR
It is possible that there are errors in the results. However the pattern we are examining, seems too precise to be an error. Figure 4 shows how precise the pattern is. In this graph we have divided all of Wasserman Schultz' results by 2.5 and multiplied Canova's results by 2.5. The result is two straight lines. We are not suggesting that these are the actual results. We are simply illustrating that when the reported results are divided by a single multiplier, they create straight lines. This demonstrates how mathematically precise the pattern is, arguing against it being caused by an error.
Dr. Clarkson, who is the chief statistician at an aviation research institute went so far as to say, "It's definitely not an error." She pointed out that the pattern has been documented consistently in too many other races to be an error.
Figure 4 - If Wasserman Schultz' results are divided by 2.5 and Canova's are multiplied by 2.5, the result is two straight lines. This type of precision argues against the pattern being caused by errors.
DEMOGRAPHIC EXPLANATIONS
We did extensive analysis of all the demographic data that is available from the Broward County Supervisor of Elections for this race. Most of the precincts for the district are in Broward County, so this represents the majority of precincts in this race. Figure 5 shows a graphic representation of all the demographic data that was supplied for registered voters.
Figure 5 - Cumulative analysis of all demographic groups supplied by the Supervisor of Elections of Broward County.
The only demographics that show much variation are the Black, Hispanic, White, and Other Race voting blocks. So we focused our research on those groups.
There are demographic differences between the small precincts and the large precincts, but they are not enough to account for the pattern in the election results. Figure 6 compares Wasserman Schultz' increase with variations in the Black and Hispanic voting blocks. Black registered voters increase as the precinct size increases, but not by enough to account for the increase in Wasserman Schultz' support. The increase in Black registered voters is less than 4%, and only a small percentage of those voters turned out. (Turnout for Broward County overall is listed as 16.57% for this race.) Furthermore, the Hispanic vote decreases in the same precincts, almost cancelling out the increase in the Black vote.
Figure 6 - Black registered voters increase in the larger precincts, but not enough to account for Wasserman Schultz' increased percentage.
SIGNS OF MANIPULATION
The Broward County Supervisor of Elections actually provided us with even more detailed data, listing the number of eligible voters in each demographic, and the turnout percentage of each group of eligible voters. With that data, we were able to figure out how many people from each demographic group actually cast a vote. We then constructed a formula and a graph that demonstrates what percentage of the vote each candidate would end up with, depending on what percentage of the vote they received from the key demographic populations. Sounds confusing, but it's easier to understand when you look at the graphs.
We graph the projected votes from our formula alongside the reported vote totals - to see if we can re-create the election. We were told by the candidates that Wasserman Schultz was doing well in the Black community, so we start by giving her a hefty percentage of the Black vote: 85%, as well as even or leading percentages in the other demographic categories (Figure 7). This projection does not match the reported totals.
Figure 7 - When Wasserman Schultz receives 85% of the Black vote and 50% of the White vote, the projected race does not match the reported totals.
In order to more closely approximate the reported totals, it is necessary to give Wasserman Schultz 100% of the Black vote, and lower her support in other demographics. This projection is closer, but still does not match the reported totals (Figure 8).
Figure 8 - The reported totals exhibit a steeper increase from the smallest to the largest precincts than any combination of the actual votes.
No matter what percentages of the actual votes we distributed to the candidates, we were unable to replicate the reported results. If the division of the votes is close to what we started with in Figure 7 and then a manufactured 3.5% increase in Wasserman Schultz' results, and a manufactured 3.5% decrease in Canova's results are input into the graph, then the reported totals can be replicated almost exactly (Figure 9). This would seem to imply some kind of manipulation was necessary to obtain these results. It is possible that the reported results are based on demographic trends, but that those trends are being exaggerated in some way.
Figure 9 - If a 3.5% manufactured increase for Wasserman Schultz and a 3.5% manufactured decline for Canova are added to the actual votes, then the reported results can be created almost exactly.
Another possible permutation is to give Canova a higher percentage of the Black (30%) and Hispanic (55%) vote, although still giving Wasserman Schultz the lead in the Black community that both sides agree she had (Figure 10). Under that scenario, Canova would win the race, and the statistical pattern of those results looks closer to the expected statistical pattern from Figure 2A, although with a reversal of who wins the race. In this projection, each candidate finds and maintains an average percentage of support, more closely following the law of large numbers. This interpretation of the data shows Wasserman Schultz with a bump up in the precincts that are known to have a higher registration of Black voters, something that corresponds to both the the narrative from the campaign and the demographic data.
Figure 10 - With this division of the vote, the race maintains both the expected statistical pattern and the known demographic influences. In this scenario, Canova would have won the election.
We would like to mention that the Supervisor of Elections' office of both Broward and Miami-Dade Counties were helpful and provided data quickly, in a polite and friendly manner. We really appreciate their professionalism and help.
We do not know what the actual vote totals from the race might be. We are concerned, based on the data available, that the votes that were cast may not be fully represented in the reported totals, and that those reported totals may not be accurate.
SOLUTIONS
There is so much concern in the electorate that the system is rigged; that the elections will be, or are being hacked; that our candidates do not reflect our best possible leaders. All of these are indications that our election process is broken. The first step in fixing it would be to move to a secure, transparent election system. Our team supports:
- Paper ballots
- Marked by the voter
- With secure chain of custody
- Hand-Counted (at the precinct immediately following the vote if possible)
- In a transparent process that is open to scrutiny by the public and the media
- The results of the count, along with any concerns, must be posted publicly in each precinct immediately following the count
Hand-counted paper ballots are used in most of the democracies around the world. It is an affordable, efficient solution that some U.S. counties have already implemented. We would like to see it adopted nation-wide.
This is a clip from my documentary Holler Back - [not] Voting in an American Town that shows the hand-counted process in action.
MORE INFORMATION
ABOUT THIS RESEARCH
Is available at ElectoralSystemInCrisis.org
ABOUT ME
The blog is by lulu Fries'dat. I'm an Edward R. Murrow award-winning journalist and documentary filmmaker. My work focuses on systemic problems with our elections that discourage participation. Follow me on twitter @luluFriesdat
This is a longer bio:
Ms. Fries’dat received a Best Documentary award for her first feature-length documentary (producer/director) Holler Back — [not] Voting in an American Town, a film that explores why more people don't vote and participate in politics. Clips are available for viewing here. Her network news experience includes editing assignments for CBS Evening News, Nightline, Sunday Morning, The Today Show, and Good Morning America. She produced and edited profiles of Democratic candidates for MSNBC, and has done long-format documentary work with NBC News and CNBC. She was on the editing team of Gideon’s Army, an Emmy-nominated documentary that follows the personal stories of public defenders in the Deep South. Gideon's Army received the U.S. Documentary Editing Award at the Sundance Film Festival and the Ridenhour Prize, fostering the spirit of courage and truth. Other film credits include Are We Not Men - The DEVO documentary, Joe Papp in 5 Acts (PBS/American Masters) and Slamnation by five-time Emmy winner Paul Devlin.