Overview of the National Transit Database (NTD)

Pierce Transit 2
Close up of the Tacoma Link streetcar. Christopher MacKechnie

Federal Transit Reporting Requirements: National Transit Database (NTD)

As a condition for receiving funding from the federal government, transit systems need to report on several aspects of their operation. One such aspect is how well the system meets the requirements of the Americans With Disabilities Act . Another is how well the system meets the goals of Title VI, which mandate that transit services be distributed in an equitable and non-racist manner.

Perhaps the most important reports they make are of their safety, security, finances, and ridership, which form the basis of their entry in the National Transit Database .

Transit Safety and Security

Under the safety and security rubric, agencies need to report to the Federal Transit Administration (FTA) how many accidents and crimes occur on their vehicles every year. These statistics are then made available to the public on the site; for example, in 2010 there were ninety-six fatalities on America's heavy rail systems of which forty-one were suicides. Overall, from 2002 - 2010 there have been an average of 184 deaths; 21,468 injuries; and 12,726 security (crime) incidents reported annually to NTD.

Calculation of Transit Ridership

In addition to statistics about transit safety and security, the NTD measures annual unlinked (each boarding is counted separately even if a passenger is transferring) passenger trips (UPT) and passenger miles for all systems.

Average weekday, average Saturday, and average Sunday passenger trips and miles must be reported in addition to the annual total, and passenger trips must also be reported by time of day (A.M. peak, mid-day, etc.). Because the number of trips and miles is so high in comparison with the number of safety and security incidents, it has always proven more difficult to measure trips and miles.

Fortunately, the widespread deployment of Automated Passenger Counting (APC) systems has made it easier for agencies to calculate ridership in recent years. Procedures differ based on whether an agency has the ability to precisely measure 100% of their annual UPT's or not.

Agency Does Not Have the Ability To Measure 100% of UPTs

If an agency does not have the ability to measure 100% of their UPTs, then a method must be devised to randomly sample their scheduled trips. Scheduling software like Hastus can easily randomly select trips to sample, or the transit planner can use random number generators.

For agencies that operate different modes (like bus or light rail), different trips must be randomly selected for each mode. In addition, agencies that operate types of bus service that are very different from each other (i.e. express routes versus local routes) should separate them and select random trips for each type.

The FTA offers several different sampling methods that agencies can use to estimate their passenger trips and miles. Each of these sampling methods requires the agency to sample a different number of trips in a given time period. For example, a transit agency might sample three one-way trips every other day for an annual total of 549.

NTD sampling methods result in an annual range of sampled trips from 208 - 915. On each of these trips a surveyor must record the number of people that board and disembark at each stop, the number of people who are on the bus as it leaves each stop, and the distance between each stop.  By 2015 a new sampling method that focused on weekly or monthly sampled trips rather than daily became mandatory; this sampling method has resulted in a decrease in the number of trips that need to be sampled for most transit agencies while simultaneously improving the "randomness" of the sample.

As a result of this sampling the agency calculates the average number of passengers per trip and the average passenger distance. The number of annual passengers is therefore the number of passengers per trip multiplied by the number of trips per year, and the annual passenger miles total is calculated by multiplying the annual passenger total by the average passenger distance.

Limitations of Sampling

Each of the sampling methods NTD approves result in a 95% probability that the sampled annual passenger miles are within plus or minus 10% of the actual passenger miles. The first limitation is that the data can only be looked at on an annual basis. Although data is reported to NTD on a monthly basis the data is estimated only, and certainly subject to random fluctuations. While the breakdown by day-type (weekday, Saturday, Sunday) may be technically accurate, it troubles me to base average Saturday ridership on results generated from an annual total of seventy-eight trips. Since most transit agencies have a natural desire for as much ridership information as possible, the NTD process will either disappoint them or provide them with potentially inaccurate data.

The second limitation is the statistical sampling itself. If we were to run 100 separate NTD sampling methods over a given year for a particular transit agency, in 95 of them the results would be within 10% of the actual total. Although a 95% confidence interval is perfectly acceptable in science and other applications (although 99% is better), in many cases when we are experimenting we are less concerned with the actual numerical result (which will vary in each experiment) then its general nature. Since millions of dollars ride on passenger counts, basing the count on a system that will one time out of twenty not even produce a total within 10% of the true total seems problematic. Of course, this does not even take into account that for large agencies a +/- 10% range could represent a difference of 20 million annual passengers!

Agency Does Have the Ability To Measure 100% of UPTs

Fortunately, the advent of APCs solves the above two problems. Assuming you have APCs on 100% of your buses, you have a complete ridership count generated every day that, because APCs count alightings as well as boardings, is also a complete passenger miles count. Upper management receives endless ridership reports, and the inherent limitations of sampling are a thing of the past.

The FTA allows transit agencies with well-functioning APC systems to submit data in that manner rather than going through the random sample process described above.

Outlook on Passenger Counting Systems

The development of APCs has certainly given us more accurate ridership totals, and it is likely to upset conventional wisdom that fare hikes and service cuts cause declines in ridership. My report on service cut strategies reported that ridership changes were minimal when the Chicago Transit Authority (CTA) and Los Angeles Metro cut service, but were severe when Community Transit in Washington and Sacramento did so.

Although when I wrote it my point was that the type of service cut strategies employed affected the result, I wonder if the way ridership was measured also had an effect. Both Metro and the CTA use APCs, while at least at the time of the ridership losses neither Community Transit nor Sacramento employed them.

How might the NTD sampling process overstate ridership declines? Recall that NTD ridership is generated from multiplying the average passengers per NTD surveyed trip by the total number of trips operated in a year. If you operate fewer trips per year, then your NTD ridership will decline unless your average passenger statistic increase enough to offset the loss from multiplying by a lower number. But if I cut a trip that has zero passengers, then I am not losing any ridership. By treating every trip equally, the sampling method cannot take into effect the relative productivity of each trip.

A promising future line of research would be to examine how service changes affect ridership in systems that use APCs versus systems that do not, in order to see if any differences exist.

Regardless of how transit systems collect their data, they will be able to utilize software such as ArcView by ESRI to analyze it.