Waiting with the TSA

So I was traveling for a conference this week.  I was consequently interested in an article in Monday’s New York Times about growing passenger frustration with the Transportation Security Administration’s passenger screening procedures (Flier Patience Wears Thin at Checkpoints, Nov 8). The article notes that complaints about the TSA’s procedures have to do both with their increasing intrusiveness and their speed. And, if you run British Airways, you may also complain that the TSA puts a greater burden on foreign carriers than domestic ones. Interestingly, those in the business seem more concerned that the hassle of getting through the airport will discourage travel than that another terrorist event may shut them down.

Giovanni Bisignani, chief executive of the International Air Transport Association, said in a speech at an aviation security conference in Frankfurt last week that the airlines would like to see an overhaul of the checkpoint screening process — with a greater focus on finding bad people, rather than bad objects. “Discouraging travelers with queues into the parking lot is not a solution,” Mr. Bisignani said in his speech. “And it is not acceptable to treat passengers as terrorists until they prove themselves innocent.”

That raises the question of just how bad waits are and whether they are getting worse. There is some evidence on the latter point. The wider use of body scanners has supposedly slowed down the process as those who are chosen to experience the latest in privacy invasion have to completely empty their pockets etc. Further, baggage fees mean more large carry ons and hence slower screening.

But just how long is the wait? It turns out that over the years the TSA has changed how they publicly report waiting times.

Although the T.S.A. used to track security line wait times and post that data on its Web site so travelers knew what to expect, the agency stopped publishing that information in 2008. It is now searching for a way to automate the process of collecting wait-time data, said Lauren Gaches, an agency spokeswoman, but does not know when it will resume sharing that information with the public.

Historical data posted on tsa.gov indicates that average peak wait times were about 12 minutes in 2006 and crept up to 15 minutes in early 2008. Since then, the T.S.A. has shifted to a system that tracks the percentage of passengers who wait 20 minutes or less to go through security, and says that 99 percent of travelers have waited less than 20 minutes in security lines in 2010.

So there are a couple interesting points to this. First, how hard is it to measure the wait? Conceptually, it’s easy but it could be very time-consuming to get a lot of data if the system isn’t automated. As a I wrote last October, I once got to participate in the TSA’s data collection procedure. I was handed a card with a time written on it which was collected when I reached the magnetometer. That gets the TSA a sample of one passenger’s time in the system but it would be very hard to collect lots of data that way. What one really needs is enough data for every hour of the day and every day of the week (and maybe for every month of the year) so that you can accurately estimate the waiting times for each period. So you need lots of data for lots of time periods. I fly a decent amount but not only have I been asked to help with this measurement only once, it is the only time that I have seen anyone get asked to help sample the queue. Also, an accurate estimate will require collecting data when the line is really long. I would not want to  be the TSA officer who has to approach the 200th person in line for help in measuring their wait.

Of course, even if the TSA did manage to collect enough data, the wait they would measure is dependent on their staffing policies and procedures. Add staff or drop some procedures (Oh, goody! I get to keep my shoes on while a stranger uses an x-ray to peer through my clothes!), and that data largely becomes useless.

An alternative take on measuring the queue is to focus on the current state instead of long run averages. If they choose to measure how long it takes to get through the line when the line is a certain length given the current staffing level, they could post a reasonably accurate estimate of the wait given the current line regardless of whether it is Monday morning or late Wednesday night. The caveat to that is that the mix of travelers doesn’t change too much over the day or that people move through the process at a given rate regardless of whether they are families with small kids or road warrior executives. If those assumptions are true, the estimates should be fairly stable. I don’t really expect that those assumptions to hold, but I would still venture that this is an easier way to estimate the wait.

There is next the question of how to report the wait. Arguably, reporting a percentile of the waiting time distribution is more informative than reporting the mean. If I go through the security line multiple times over a short horizon, I might get a sense of the overall average time. But that’s insane. Customers are not going to see the average time and it is clearly more informative to know that you will get through the process under some time with a high degree of confidence. Of course, that also invites gaming. Suppose I am certain to get through security in 20 minutes. I could then reason that I should be able to make my flight (assuming I am not checking a bag and have my boarding pass) if I get to the airport, say, 40 minutes before my flight. Suppose enough people reason that way. Then suppose that airlines cluster flight departure times (which will be likely a hub airport). A lot of people will miss their flights because too many people will hit the TSA lines at once.

Finally, there is a question of just how the TSA can measure wait times. One possibility is an automated system that takes video images and tracks how long customers spend in the queue. I have had a chance to see some of these systems.  They are pretty cool but I think there are issues with how TSA lines work that will cause problems with them. For example, they naively track heads moving through the system.  Or at least heads that spend sufficient time in a box drawn on the video screen. Thus they cannot distinguish easily between passengers moving through the line or TSA officers rotating positions. Further, they are limited by camera positions and thus by ceiling heights etc. Plus they will only count where they are set up for. If the line really does back up to the parking lot, the cameras will miss that.

There may be an easier way. Passengers already have to show their boarding pass twice. Once to prove they belong in the line and once to verify their identity. Why not just scan every boarding pass at both points? That would give a measure for every passenger going through. Admittedly, it will only measure part of the process since you still have to undress and remove your toothpaste.  But it would capture a large part of the wait for essentially all passengers rather easily.

By the way, the logos on this post come from a contest to come up with a more appropriate emblem for the TSA. Go here to see the rest of the entries.