Better Origin Destination Surveys

Sampling

One is never able to observe all trips in a study area at every hour of the year. Therefore, one must compromise, take a good representative sample and then expand it to get a good representation of the full travelling population.

Intercept surveys involve asking a sample of drivers and passengers of vehicles (e.g. cars, public transport, goods vehicles) crossing a roadside station, a limited set of questions. The conduct of these interviews requires a good deal of organisation and planning to avoid unnecessary delays, ensure safety and deliver quality results. The identification of suitable sites, co-ordination with the police and arrangements for lighting and supervision are important elements in the success of these surveys. Sample size and accuracy are some of these concerns.

It is important to undertake surveys on representative days, avoiding atypical periods when the schools are on vacation, close to holidays or when the weather is exceptional (unless these are precisely the days of interest). The sample rate for a particular site depends, of course, on the number of vehicles or persons passing through it. Ort?zar and Willumsen (see Ort?zar, J. de D. and Willumsen, L. 2011; ?Modelling Transport? Fourth Edition. John Wiley & Sons) discuss in some detail how to determine the sample rate at a particular location. They conclude recommending the following table as a function of the flow per time period (hour, peak period or day).

The fieldwork procedure requires stopping at random the corresponding number of vehicles, interviewing the driver (or passengers) and asking origin, destination and trip purpose. Random sampling each vehicle category may be tricky as some drivers are less likely to stop or accept being interviewed. So a positive effort must be made to avoid surveying only easy targets (nearby lane traffic, those volunteering to stop).

In the case of public-transport trips, given the practical difficulties associated with stopping vehicles for the time required to interview all passengers, the survey may be conducted with the vehicles in motion. For this, it is necessary to define road sections rather than stations and the number of interviewers to be used depends on the observed vehicle-occupancy factors at the section. However, even this approach may be unworkable if the vehicles are overloaded.

One is never able to observe all trips in a study area at every hour of the year. Therefore, one must compromise, take a good representative sample and then expand it to get a good representation of the full travelling population.

It is very important that each survey site has at least enough interviewers to achieve the sample rate suggested, plus some back-up personnel to allow for comfort rest during the survey. In addition, there should be a well trained supervisor and an enumerator counting traffic and recording effective sample rates. Total vehicle flows will be critical to expand the sample during the analysis stage.

Expansion and correction

The sample rate (number of persons or vehicles interviewed over the total number of persons of vehicles crossing the site) should then be used to expand the observations to the universe of travellers; in other words, each observed Origin Destination flow must be expanded to represent the full number of trips. In doing so, one must take care with double and triple counting; if one had located a third survey site between junctions U and V some trips from, say, A to D will be interviewed twice. In that case, one should only expand with the average expansion rate of both sites; otherwise, one would be exaggerating the number of trips from A to D. Avoiding multiple expanding requires, therefore, identifying those OD pairs that were potentially intercepted in more than one survey site and applying a corrected expansion rate to those. If a full transport model is available, it is possible to undertake an even better correction more or less directly using matrix estimation techniques. The sum of the expanded OD pairs constitute the Origin Destination Trip Matrix to be used to plan the new facility.

At this stage, there are two options. One can model the representative day (or time period) that corresponds to the period of the survey or one can correct the flows and matrices to represent an average for the year. The first approach is the most common. One will

represent the proposed new tolled road correctly in a transport model (capacity, speed-flow relationship and toll structure) and then assign the trip matrix to minimum generalised cost routes. The resulting revenue will then be expanded as a function of the relationship between flows on the survey days and Annual Average Daily Traffic (AADT or similar acronym) on existing roads in the vicinity. This requires a Public Works authority to have undertaken surveys throughout the year to establish this AADT. The relationship between the modelled period and the year is called annualisation factor; if permanent or at least regular traffic counts have not been undertaken in advance the estimation of this factor is difficult and subject to significant potential errors.

Quality assurance

It is very easy to undertake OD surveys cheaply and badly; taking non-random samples and completing survey forms in the comfort of a coffee shop rather than the field are known problems even in the most advanced countries. The errors and omissions will only be discovered later on and will be impossible or very expensive to correct; even worse, they may not be discovered at all. It is important, therefore, to implement a good quality assurance procedure as the cost of errors is very large. This should include good supervision and random checks by senior professionals.

It is very important that each survey site has at least enough interviewers to achieve the sample rate suggested, plus some back-up personnel to allow for comfort rest during the survey. In addition, there should be a well trained supervisor and an enumerator counting traffic and recording effective sample rates.

The provision of well designed and tested survey forms together with good interview sites with enough protection, safety and security for surveyors is of paramount importance. Roadside interviews are dangerous for both vehicles and interviewers and the cost of designing and implementing a good survey site is well compensated by the quality and reliability of the results. The revenue projections based on a poorly estimated trip matrix will be erroneous and cost dearly in terms of the financial feasibility of a toll road.

Dr Luis Willumsen
Luis Willumsen Consultancy, London

The auther is an internationally recognised authority in transport demand modelling and planning. He has a distinguished academic career and extensive experience as a consultant for over 20 years, working in some 30 different countries. He has made major contributions to modelling techniques and the estimation of future traffic and revenue projections for Toll Roads, Mass Transit and Rail concessions. He is co-author of Modelling Transport, a text published by Wiley and now in its fourth edition.