Measuring the Quality of the US’s most Important 2020 Election Technology
Yes. The Internet. We train our global internet measurement platform on the US and discover the swing states most at risk.
In just over two weeks, the United States faces what could be the most consequential election in US history. The election comes at the end of a tumultuous and harrowing 2020 for America, which has tragically seen over 200,000 deaths at the hands of COVID-19, as well as country-wide protests over police brutality amassing more than 15 million Americans since May.
At the heart of any election, and 2020 moreso than any, access to information is absolutely crucial to ensure that citizens make properly informed choices. For a growing number of Americans, the internet is fast becoming their preferred means of news consumption. The endemic spread of misinformation on the internet has been a major theme of the election run-up, indeed, of Trump’s entire presidency, ever since the role of Cambridge Analytica and Russian troll farms in impacting the 2016 election. And as we approach the 2020 election, the integrity of information communication technologies is at a critical point, especially for vulnerable states.
A recent examination of state cybersecurity in the US revealed that as many as half of traditional swing states, including Iowa, Florida, Nevada and New Hampshire, faced increased vulnerability to cybersecurity breaches coming up to the election. This poses a real threat to the quality of information consumed, which then rests on users’ ability to quickly assess the veracity of information, see flagged messages on social media, and check the facts.
Accessing high quality information communication technologies can be a significant challenge for low income households, who also face a range of other barriers when it comes to voting.
Given the high stakes of the 2020 presidential election, and the crucial role of digital media and news in the formation of voter opinion, we have applied our unique global internet measurement infrastructure to explore the quality and implied robustness of internet connected end-point devices, by state.
To be clear, our approach focuses on the immediacy of user internet connections as a proxy for quality. Others have looked at the overlay of cybersecurity challenges that US battleground states face ahead of the election. Both dimensions will be important to understand as the US goes to the ballot.
Our Approach
Our approach begins by combining a commercially available geo-located IP database with our powerful scanning technology to measure the online or offline status of millions of Internet addresses globally every hour.
Our observational methodology uses the most basic Internet messaging protocol that is widely used billions of times a day to establish routes for your email, tweet, or share. After developing a carefully selected set of Internet addresses (IPs) to measure, we periodically send them one of these tiny messages, essentially asking, ‘Are on you online?’. These online/offline answers form the basis for our ‘connectivity’ indicators.
In addition, we also receive back from these measurements the immediacy, responsiveness, or latency, of the IP (measured in round-trip-time, or rtt). Latency is a reasonable proxy for the experienced speed of connection, especially for any user who is interacting with a major social platform where even basic chat activities to other users nearby must travel to a server well beyond state and even national borders (and back again).
Importantly, the Monash IP Observatory has no access to any content being shared, viewed, visited, or generated by a user at a given IP, and all Monash IP Observatory activity works in aggregates of thousands of randomly sampled measurements across geo-spatial sub-regions.
Measuring the United States’ Internet Experience
For this exercise, we focus on measurements across the 51 US states over 1–12 October 2020.
During this time, our infrastructure measured at least once over the period more than 21.6 million unique internet-connected end-points. These end-points, or unique Internet Protocol (IP) addresses, are typically attached to a home or business router, or a mobile cell-tower.
Given that internet connectivity follows population, it is not surprising that our measurements comprise over a million unique end-points in states like Texas, but just over 23 thousand in D.C. Nevertheless in all but five states (South Dakota, North Dakota, Alaska, Rhode Island and DC), our measures are built from more than 100 thousand end-points.
Specifically, our measures are breadth-first: we first collapse measurements at a single IP down to an average latency for that IP, and then aggregate across IPs in each state.
We focus on measures of latency, since, unlike bandwidth which is more about how long it takes to down-load or upload a very large file and relates to overall capacity, latency tells us about the immediacy of a connection, which is not only crucial for any synchronous online activity (chat, VoIP, tele-conferencing, tele-health, tele-education, tele-law, etc.) but also relates to the reliability of the connection to receive in general.
If the delay (latency) of the connection is slow, then the sending and receiving device can get muddled up about what has arrived, what is still ‘in flight’, and what is still to send. The usual remedy is to resend everything after a certain point. As you can imagine, that’s not great, and a lot of the time, content will just not load (or more like it, the user will just walk away).
Finally, since we find that latency measures across end-points in a state vary widely, and do not often conform to a simple, symmetric, normal distribution, we focus our attention on the median, and the 99th percentile.
The median is the middle experience — if we lined out all of the end-points, ranked by latency, then read the latency off the middle end point in the line, that would be the median. It’s a great central measure for distributions with long tails and is used extensively in areas like house prices.
The 99th percentile is the ‘worst 1%’ — going back to our line up, if we went to the very end of the line .. the end-points with the highest (slowest) latency, and took the reading 99% towards the end of the line, this is the 99th percentile. It is helpful as it avoids the ‘max’ which can be affected by strange one-off shocks.
The Standard Experience
So what did we find?
Figure 1 presents our findings for the median (recall, ‘middle’) internet latency experienced by end-points (houses, businesses, cell-towers) in each state.
It is not difficult to see the pattern. States towards the North-West are universally worse off than their Eastern counterparts.
By the median, the worst three states are Idaho, New Mexico, and Washington, on 181ms, 180ms, and 180ms respectively. Compared to the best states on this measure, New Jersey and Delaware on 154ms and 155ms respectively, the three trailing states are experiencing around 16% slower latency.
It is worth noting that the trip time across the US, by light, on fibre optic cables should take no more than around 10ms. And, our infrastructure measures from one on-shore, and three offshore source locations, with entry points on the East, West, and South of the US. To be sure, we dropped our only US based infrastructure source and repeated the analysis. It didn’t materially change the results.
So the variation cannot be explained simply by distances.
That said, at a 25ms difference, the median user of the internet in the US is not greatly impacted by their geography.
However, democratic elections imply that everyone gets the chance to vote. And that means, the key comparison is not really the median internet user, but actually the experience of the user with the worst internet experience in a given state.
In other words, democratic principles means we should consider the inclusive quality of supportive infrastructure.
Only by looking at the bottom 1% will we be able to assess if the internet is likely to be a barrier for voters engaging with the political process, and even getting out to vote.
And Inclusive Quality?
In Figure 2, we present the comparison for the 99th percentile — the latency of the slowest 1% of end-points in each state.
Here, the differences are very stark. Pay close attention to the scales.
The differences between the winners and losers is out in the hundreds of ms, or over 100%.
At the top (best) end, Delaware and Rhode Island again feature, with 99th percentiles of 207ms and 219ms respectively. Note, this tells us that the worst 1% of users in these states experience only a fraction slower immediacy of their connection than the median user in states like Idaho, New Mexico, and Washington.
At the lower end, the numbers are eye-watering.
Oklahoma stands out on 455ms, with Iowa at 407ms, and West Virginia on 383ms.
What this means is that between the internet quality of say Delaware and Oklahoma stands 248ms, or a relative quality drop of 119%.
But what does this analysis mean for battleground states?
The 2020 Election Battleground States
In Figure 3 we present the full set of states, ordered by the 99th percentile, and showing the gap between the median and 99th.
We colour these bars by accepted norms around the state being likely republican, democrat, or swing in the forthcoming poll.
As you can see, Republican states are more heavily represented on the right hand side, reflecting the more extreme differences between internet speeds for the average user and the worst 1%.
Indeed, of the bottom 15 states, 9 (60%) are Republican and just 2 (13%) are Democrat, with the remainder, 4 (27%) being swing states. Conversely, if we looked at the top 15 states, the picture would be almost perfectly reversed.
So which swing states have the worst performance on this metric of inclusve internet quality?
Iowa, Wisconsin, Pennsylvania and Michigan are all in the lower pool.
But of these, Iowa is the standout with a bottom 1% experience of slightly over 400ms. Contrast this with Delaware (D) where the worst 1% experience is half of this value, at just over 200ms.
To put this in context, the typical latency or immediacy of internet end-point devices in Sydney Australia, connecting to servers in the US is 200ms.
Suggesting somewhat depressingly that the bottom 1% users in Iowa are effectively receiving packets of internet content as if they were being re-routed to Australia and back again before popping up on screen.
400ms or above is around 0.4 of a second. This may not sound like much, but consider that human speech is typically 150 words per minute or above, which equates to around 0.4 or less of a second per word.
So practically, the user in Iowa is going to struggle with any synchronous activity online as they’ll be more than a word behind everyone else.
And, as we’ve discussed already, poor latency typically points to dropped packets, and so, a poor overall internet experience.
This is a problem.
Countdown to Nov 3
So as the election draws closer by the day, spare a thought for Iowans, or residents of Wisconsin, or Pennsylvania, trying to get updates on the Presidential debates, or crucial information on where to vote, or how to lodge mail-in ballots, or what their rights might be as they approach in-person polling stations.
Aside from everything else they’ve had to face this year, a good number of them may be starring at an interminable spinning wheel or the word “loading”, blandly blinking on their screens.
For the sake of democracy in the US, let’s hope they have the patience to stick it out online.
Though really, it shouldn’t be so hard.
Acknowledgement: Hayley Lock assisted with writing, background research and analysis in the preparation of this article.
The mission of the Monash University IP Observatory — ‘internet insights for social good’ — is to monitor the availability and quality of the Internet during critical events such as elections, natural disasters or conflicts. The IP-Observatory is fully compliant with the EU’s General Data Protection Regulation (EU-GDPR). The IP-Observatory does not collect, hold or process personal data. The IP Observatory was founded by Klaus Ackermann, lecturer in Econometrics and Business Statistics, and Simon Angus, and Paul Raschky, Associate Professors in Economics. The observatory is a project of SoDa Laboratories at the Monash Business School, and tweets @IP_Observatory.