I. Introduction
I’ve had a lot of conversations with friends about traveling over the holidays. There have also been articles written about the risk of traveling. Of course, at the end of the day, people need to make their own decisions, but how are they supposed to do that? How are regular people supposed to assess risk in these situations?
My goal in writing this is to share the way that I’ve tried to estimate risk. I won’t pretend to be an authority here, because there are no authorities. But my hope is that, in clearly explaining how I’ve estimated risk, it will give other people a place to start as they try to make decisions about when and how it is safe to travel.
II. Characterizing Exposure Risk While Traveling
My new favorite piece of COVID-related public health advice comes from the mayor of Los Angeles: “don’t share your air.” It’s succinct, it rhymes, and it communicates the principle underlying the CDC recommendations in a way that empowers people to make intelligent judgment calls in situations where following the letter of CDC guidance (wear a mask indoors, stand 6-feet away, etc…) doesn’t make sense. There are some situations where following the guidance isn’t necessary, other times when it isn’t sufficient. “Don’t share air” is a good heuristic when facing these situations.
So, how is this advice be applied to air travel?
Basically, how much air you share with another person depends on 1) the distance between you, 2) the amount of time, 3) whether you are both wearing masks, and 4) how well ventilated the area is.
In what situations are you likely to share air with people?
In lines (e.g., at security, boarding, baggage check)
Lines are risky because you’re standing near your neighbors in line for multiple minutes. Airports are (hopefully) going to require that people stand at least 6 feet apart and keep their masks on, but even if these rules are observed, you will still share some air with your neighbors if the space is poorly ventilated and the line is moving slowly.
At the food court
Here, this risk is high. You have a bunch of people hanging out in the same area for multiple minutes (eating, standing in line, and waiting to get their food), and the people who are eating aren’t wearing a mask. My view is that you should just stay the hell away from airport food courts. Pack food at home, and find a secluded corner to eat it in.
On the airplane
Airplanes are a mixed bag. The good news is that airplanes are apparently quite well ventilated, which means that you mostly only share air with the people sitting next to you. The bad news is that, because you’re sitting still for multiple hours, you share A LOT of air with the people sitting next to you.
This means that the amount of shared air depends predominantly on how packed the seating is on an airplane (i.e. how many people are sitting near you), and increases if passengers and flight staff move around the cabin a lot.
On airport transit
Some of the bigger airports have multiple terminals that are connected by trams. These trams should be avoided, if possible. You’re standing in a closed area with other people for multiple minutes. Masks help here, but I wouldn’t count on them.
Walking through the airport
I actually don’t think you’re likely to get COVID here. Yes, you’re indoors with a lot of people, but everybody is moving quickly, everyone is (hopefully) wearing a mask, and it should be relatively easy to keep your distance from other people.
Waiting at the gate/boarding
This issue was brought to my attention by my partner, who flew recently. She points out that airport terminals were not designed with social distancing in mind, and if there are multiple crowded gates in close proximity, it’s hard to stay more than 6-feet away from other people
III. Estimating Exposure Risk Part 1, Number of Contacts
Let’s try to quantify the risks. In the table below, I’ve estimated the number of people you’re likely to share air with each time you do one of the activities above.
| Activity | Number of people you share air with (per instance) | Notes |
| Standing in Line & Waiting to Board | 0-4 per line/wait | This assumes that you could potentially be exposed to your neighbors in line. Varies with how spaced-out the line is and how quickly the line moves. |
| Onboard the airplane | 0-9 per flight | This assumes that you’re mainly sharing air with the people sitting in your row, or immediately in front/behind you, and maybe the odd person who walks past. Varies with how crowded the plane is and how much the people move around the cabin. |
| On Airport Transit | 0-10 per car | This assumes that you get exposed to everybody in the airport transit. Varies with how crowded the transit car is. |
The big takeaway from this table is that the number of people you’re exposed to can vary wildly.
One factor is your travel schedule. Each flight you take adds another line (boarding) and another plane flight (duh), which practically doubles the number of people you share air with. The other factor is how crowded the flights and airports are.
Here’s a table with estimated “contact” numbers in different travel scenarios.
| Direct Flight | 1 Connection | |
| Low Crowds | 2 | 5 |
| Moderate Crowds | 6 | 10 |
| Heavy Crowds | 14 | 25 |
The good news is that flying can be extremely safe, probably comparable to grocery shopping, if you take a direct flight on a day when the crowds are small.
The bad news is that, on crowded days, the exposure compounds across each leg of the trip, because the lines are likely to be more congested AND the seating on airplanes is likely to be more crowded. On a busy travel day, a trip with multiple connections could have you sharing air with up to 25 people.
In short, the travel schedule matters a lot when you’re estimating risk, and avoiding busy days can reduce risk dramatically.
IV. Estimating Exposure Risk Part 2: COVID Case Count
The other half of this risk estimate is how likely people you run into are to have COVID. How many circulating cases are there?
The way to estimate this to do this would be to download COVID data from covidtracking.com, look at the new cases diagnosed in the past ~9 days, and then calculate the probability that none of the individuals in a randomly selected group of a particular size have COVID.
Or you can use this online tool, created by the good people at Georgia Tech, which does all of this for you.
I plugged the numbers from the previous section into the tool, and calculated the likelihood that you’d share air with at least one person who has COVID in each travel scenario.
| Direct Flight | 1 Connection | |
| Low Crowds | 5% (1% to 10%) | 12% (2% to 22%) |
| Moderate Crowds | 14% (3% to 26%) | 22% (5% to 40% |
| Heavy Crowds | 30% (7% to 50%) | 46% (11% to 72%) |
The bold number is the Best Guess. The range in parentheses is what I call the “range of sanity” (which I’ve discussed in greater detail here). The uncertainty here comes from the assumptions we make when we estimate the true prevalence of COVID-19 among airport travelers.
Unfortunately, we don’t actually know what % of airport travelers have COVID. The best reference point we have is the number of confirmed COVID cases that were identified in the past ~10 days in the U.S. That number is about 1.6 million, about 0.5% of the U.S. population.
Unfortunately, this number cannot be taken at face value, because not everybody who contracts COVID is tested, and these people don’t show up in the official data. To get the “best guess”, I assumed that, for every confirmed case of COVID, there are 4 cases that go undiagnosed, meaning that the true number of active COVID in the United States is 5x the number of confirmed cases (8.1 million, or ~2.5% of the population). This is the value that was recommended by the people who created the risk assessment tool, and it’s roughly consistent with the estimate gotten using the methodology outlined at COVID19-projections.com. It also happens to be a default setting on the Georgia Tech tool, which is handy.
Of course, even if we’ve estimated the number of COVID infections in the general population, this might be a poor estimate of the % of airport travelers with COVID. Airport travelers are not a random cross-section of Americans. On one hand, airport travelers are less likely to be sick than the general population, because people who feel sick tend to stay home. On the other hand, airport travelers are more likely to have recently risked COVID exposure, because people who are cautious about COVID tend to avoid airports. It’s not clear which of these forces is stronger.
In other words, we have some uncertainty about the initial estimate, and we’re not sure about the direction of the estimate. This is why I’ve put in the range of sanity.
For the lower bound of the “sanity range”, we’ve assumed that the true COVID incidence is equal to the number of confirmed cases. It’s what you would expect if people are being really good about getting themselves tested before they travel and are generally willing to cancel their travel plans if they feel sick.
For the upper bound of the “sanity range”, I’ve assumed that the true incidence is 10x the number of confirmed cases, instead of 5x (conveniently, also a default setting in the Georgia Tech tool). This is represents a world where there are a huge number of undiagnosed cases and/or the airport population is full of the the riskiest of risk takers. You could also get these number if, for example, the tests that most people are taking before they travel have a really low false-negative rate.
If you want to re-create my analysis (which everyone should definitely do, because my estimate of the prevalence will be out of date, and local data might be better than the full US data), here is how to do it:
- Go to the Georgia Tech Risk Assessment Tool
- Click the “real time U.S. and State-Level Estimates” tab.
- In the text box on the left-side of the screen, select the state whose data you think best reflects the population of your airport (or, if you want to use the U.S. total, uncheck the box that says “limit prediction to state level”)
- Enter the number of people you think you’re likely to share air with during your travels, based on your travel plans
- The Best Guess will be listed as the “Chance someone is COVID19 positive at 5x CI “
V. Some Philosophical Points
First, risk of being exposed to COVID is different from the risk of actually contracting COVID. This risk depends on exactly how much air you’re sharing with the people you’re exposed to, how much virus the infected person is shedding, and other things that we just don’t know. My initial guess put the number between 10% and 80%– a range which speaks to my uncertainty on the on the subject. Zvi Mowshowitz (who has been putting out excellent weekly roundups of COVID data) thinks that 10% is at the high-end of plausible in situations where you’re near another person for 10s of minutes, assuming that everyone is keeping their masks on, not talking at each other, and generally not being stupid. I don’t understand his reasoning well enough to reproduce it, but I’m inclined to trust it.
Second, there is an open question about how we should think about risk tolerance. I’m of the view that one should be fairly risk adverse right now. We have a vaccine that will be widely distributed in the next ~3 months, which will make a substantial fraction of the population immune to COVID. This means that any risky activity that you want to do in the next three months could be done risk-free in May.
(Yes, this means we should consider rescheduling Christmas celebrations for May. Jesus was born while the shepherds were in the fields in any case.)
Finally, a note about this exercise. I have tried to be transparent about the uncertainties in this estimate. This is not the gospel truth. But I think it’s helpful to lay the foundation for constructive conversations about risk assessment in the final stretch. Having thought about risk today, even with uncertainty, makes you a lot better at processing and contextualizing information that you receive tomorrow.
For now, that’s all.
Narrative Leaps 