Methodology

The ETA uses the term “down-ballot difference” (also known as “drop-off”) to mean the difference between the number of votes cast for a presidential candidate when compared to the number of votes cast for the next down-ballot candidate of the same party.

For example, if Kamala Harris won 100 votes and the next-down ballot candidate for Senate won 95 votes, that would be a down-ballot difference of 5 votes. When calculated as a percentage, that’s a down-ballot difference of 5%. 

 Across all parties, it is more common for Presidential candidates to get slightly more votes than other candidates down-ballot. In the U.S., where significant emphasis and media attention is given to the Presidency, it is ‘normal’ for the candidate for President to receive on average between 2-3% more votes than the next down-ballot race. (Source: The Center for Politics)  For this reason, the ETA calculates down-ballot difference in relation to Presidential votes. 

A negative drop-off difference is shown when a down-ballot candidate receives more votes than the Presidential candidate.  

The ETA uses the term “Vote Share by Vote Count” to mean examining the percentage of votes that candidates received across different precincts (or voting centers) relative to how many votes were cast at those locations. 

The usual way that we show this information on a chart is through scatterplots. Scatterplots can be helpful in that they allow us to look at all of the data points at once. For example, each point on a scatterplot may reflect data associated with one candidate's results in one precinct. 

The example below shows a series of data points (the little colored circles), each representing election results for one of two candidates competing for the same office in a given county. Each pair of 1 yellow + 1 green colored circles = data from one precinct.

The location of each dot vertically shows what percentage of the vote the candidate won at that precinct. The location of each dot horizontally shows the number of votes received by the candidate at a given location (e.g. a precinct). The darker trend lines indicate the overall trends between the data points.  

An overall trend emerges across precincts of varying sizes. Some degree of variation is normal, but a common pattern is present. People in this county voted for Candidate A about 75% of the time and for Candidate B about 25% of the time. When examining at the county level, the ETA would expect for there to be a relatively consistent relationship between candidate popularity and the number of votes per precinct – visually represented by a horizontal trendline. 

The trends displayed in the chart above would not be of immediate concern were they to appear in real-world election results. 

In this case, the chart above shows a ‘binary’ election – meaning that all (or the vast majority) of votes have been cast for two candidates. This means that when Candidate A gets about 60% of the vote, Candidate B gets about 40% of the vote.

This is why it is normal for election results for each candidate to ‘mirror’ each other in binary elections. 

 What Is ‘Expected’ About These Election Results?

• Studies have shown that there is a spatial (geographic) relationship in voting patterns, meaning the place you live and the people you live close to influences the way you vote. Local shared experiences tend to inform a common mindset that nudges people’s decisions – including the decision of who to vote for. 
• We can therefore anticipate some similarities in voting results among people who live in proximity to each other (for example, within the same county).
• We would also expect to see normal human variation across the different places where people cast their vote. 
• No one candidate benefited, unexpectedly or disproportionately, in precincts where more or fewer votes were cast.

What Would Be Unexpected?

A major skew in how well one candidate did relative to how many votes were cast at a certain location could be cause for additional scrutiny. 

1. An election official in Russia or Georgia stuffs handfuls of falsified paper ballots for one candidate into a ballot box. When the votes are counted, that voting location shows very high voter turnout.
   
2. An election official picks up the phone to call and report the election results to a central authority. The election official lies and claims there were 500 extra votes for their preferred candidate at that location. The original ballots are taken away and stored where they cannot be easily accessed, and 500 extra falsified voters are added to the poll books. When the votes are reported to the public, that voting location shows a very high voter turnout.
   

In both examples, the end result is that high voter turnout is reported as a result of additional falsified votes being added to the tally – and that high turnout only benefits only one candidate or party. 

The exact same principle applies to electronic forms of election fraud. 

• Instead of an election official stuffing a ballot box or lying about the vote count, imagine a vote-counting tabulator being compromised by bad actors and reporting inaccurate results. 
• Instead of a paper pollbook having falsified records of voters added by hand, imagine a hacked “ePoll book” that claims more eligible voters cast a ballot than actually turned out.

There are several ways to show turnout analysis visually, including:

• Line charts, a type of chart that visually represents data by plotting points and connecting them with lines;
• Histograms, a type of chart that groups together data into bars of varying heights to show the frequency of data points in relation to each other; and,
• Scatterplots, a type of chart that shows a series of data points.

Which chart is best depends on type and how much data is available. When possible, the ETA tries to use histograms to show turnout analysis.

Each data point shown reflects all the votes received of whatever vote type is being measured by that candidate at that location. There is nothing temporal (‘time-based’) about what is being measured on the x-axis.

Our English-speaking brains tend to see ‘numbers getting bigger, left to right’ and instinctively assign some kind of time-based meaning to the visual. This is not the case in our Turnout Analysis charts. 

In the chart above, each bar represents the number of votes cast for a candidate at precincts within a certain turnout range. For example, the highlighted bar in yellow is showing that in precincts where about 40-42% of registered voters turned out to vote, there were around 175 votes for that candidate. 

In the example above, the distribution of election results in relation to voter turnout looks as anticipated. The election results form a bell-shape (or a “bell curve”). 

When data forms a bell curve, most ‘values’ (the things that we’re measuring) fall close to the average (the ‘middle’ of the bell curve). The further away from the average a value is, the less prevalent it is. When representing data visually, whether or not the data you are representing ‘fits’ a bell curve can be an indicator of whether something unusual or unexpected happened.

Evidence of manipulated election data in other countries has been identified through this approach.

One known indicator of unfair elections is an anomalous deviation from normal distribution, wherein one candidate receives a disproportionate number of votes in areas where turnout is high. Referred to as a ‘Russian tail’, such a spike may indicate election result falsification, particularly if only one candidate benefits from the unusually high turnout. However, there may also be honest reasons for deviation from normal distribution -- such as local dynamics, demographics, and/or geography.