Household travel surveys: Where are we going?

0 downloads 234 Views 183KB Size Report
emerging capabilities in data fusion. Using some ... transport planning, large data sets were collected, using face-to-f
Transportation Research Part A 41 (2007) 367–381 www.elsevier.com/locate/tra

Household travel surveys: Where are we going? Peter R. Stopher *, Stephen P. Greaves The Institute of Transport and Logistics Studies, The University of Sydney, NSW 2006, Australia

Abstract In this paper, we commence by reviewing the recent history of household travel surveys. We note some of the problems that contemporary surveys are encountering throughout the world. We also review the data demands of current and emerging travel demand models, concluding that there are many new demands being placed on data, both in terms of the extent of the data required and the accuracy and completeness of the data. Noting that the standard method for conducting most household travel surveys is, and has been for some years, a diary, we briefly explore the evolution of the diary survey from the late 1970s to the present. In the next section of the paper, we explore a number of facets of potential future data collection. We include in this the use of GPS devices to measure travel, the potential of panel designs and some of the alternatives within panel designs, the development of continuous household travel surveys, especially in Australia, and the emerging capabilities in data fusion. Using some of these emerging methods for data collection and data simulation, we then propose a new paradigm for data collection that places the emphasis on a paid, national panel that is designed as a rotating, split panel, with the cross-sectional component conducted as a continuing survey. The basis of the panel data collection is proposed as GPS with demographic data, and the continuing national sample would also use GPS at its core. The potential to add in such specialised surveys as stated choice and process surveys is also noted as an advantage of the panel approach. We also explore briefly the notion that a special access panel or panels could be included as part of the design.  2006 Elsevier Ltd. All rights reserved. Keywords: Data collection; Travel demand modelling; Panel surveys; Data fusion; GPS; Travel diaries

1. Introduction It is axiomatic that a model can never be better than the data from which it is estimated. In the early days of transport planning, large data sets were collected, using face-to-face home interviews, with sample sizes that were often as big as 1–3% of the population. Over the past 20 years, sample sizes have dropped considerably, and are more often now in the range of 2500–10,000 households, often representing much less than 1% of households in the region (Stopher and Metcalf, 1996; Cambridge Systematics, 1996), and surveys in North *

Corresponding author. Tel.: +61 2 9351 0010; fax: +61 2 9351 0088. E-mail address: [email protected] (P.R. Stopher).

0965-8564/$ - see front matter  2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tra.2006.09.005

368

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

America, at least, are now customarily conducted by telephone, face-to-face interviewing having become both too expensive and too dangerous to accomplish in most urban areas of the continent (Stopher and Metcalf, 1996). Outside North America, face-to-face interviews are still done, although costs are becoming increasingly problematic, and threats to the safety of interviewers are also becoming more common. In Sydney, Australia, for example, current costs of the face-to-face interviews being conducted as part of the Sydney Household Travel Survey are around $350 per completed household. Another trend in household travel surveys is increasing non-response rates (Wilson, 2004). A typical (good) computer-assisted telephone (CATI) survey in North America will usually achieve about a 60% recruitment rate, followed by a 60% completion rate for recruited households, representing an overall response rate of about 36% (Stopher and Metcalf, 1996; Zimowski et al., 1997). Even when recruitment rates are increased to 70%, and completion by recruited households to 70%, the overall response rate is still only 49%. However, if the recruitment and response rates drop to 50% each, then the overall response plummets to 25%, which is a response rate that is more commonly associated with postal surveys in North America. Furthermore, it is known that many of the households that are non-respondents travel more than the average, or are larger households (DRCOG, 2000; NCHRP, 2006). Such households end up as non-respondents, partly because they are very hard to contact (those that travel more than the average), and partly because they see the survey task as being significantly more burdensome, because of both the number of members of the household, and the amount of travel that would need to be reported. Recent work by Wolf et al. (2003), Pearson (2004), and others has shown that diaries completed through telephone retrieval omit a significant number of trips. Wolf has reported a shortfall in trips of as much as 60%, but with most surveys showing a shortfall of 20–30%. A majority of these missing trips are fairly short trips, but they call into question the ability of diary surveys to obtain accurate travel data. Even face-to-face interviews are not immune from this problem, with current work in Sydney, Australia showing a shortfall of about 7–12% in reporting of trips in a face-to-face interview (Stopher et al., 2005a). Of particular concern here is the shift towards tour-based models, or activity models, where the omitted short trips may either produce an incorrect definition of a tour, or may omit important activities, and result in an underestimate of the number of certain types of activities that are undertaken. In addition to failing to report the correct number of trips, it is well known respondents vary markedly in their ability to provide accurate details of the other key components of their travel. For instance, the tendency to round travel times to the nearest 5, 10 or even 15 min, the inability to provide location details to the degree of precision required, and the failure to provide even basic route information (Stopher et al., 2005a) have all plagued the development of both conventional and emerging models. For surveys that rely on the telephone, whether as a means to recruit households, to identify the sample, or to collect significant amounts of data, there are other trends that are somewhat ominous. First, in the US, the establishment of the Do Not Call registry, while so far exhibiting a relatively benign effect on household travel surveys in North America (BTS, 2002; Stopher et al., 2005a), opens up the possibility that other countries may adopt similar procedures1 that may be more harmful to public-service orientated surveys such as household travel surveys. Second, there is increasing incidence of households that are moving to having mobile phones only, with no land lines. Given that most survey ethics standards currently do not permit calling to mobile phones for a survey, this trend is going to remove an increasing number of households from a telephone-based survey. Third, there are increasingly sophisticated call-screening devices available and call-screening options are being used more extensively, especially in North America. Notwithstanding the Do Not Call registry, there is at least anecdotal evidence to suggest that people use these call screening mechanisms to screen out not only marketing calls, but any calls from numbers that they do not recognise (Tuckel and O’Neill, 2001). Thus, it seems clear that the trend for surveys that rely upon the telephone as either a contact or data collection mechanism is that it is becoming more difficult to collect quality data by such a means (Zmud, 2003). Households are becoming harder to contact, the sampling frame is likely to miss increasing numbers of households, and the quality of the data has been called into question by the validation through GPS.

1

Currently, the Commonwealth government of Australia is considering introducing such legislation.

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

369

2. Data needs of current and emerging models The state of practice in travel demand modelling has been evolving relatively slowly, and much slower than the state of the art of modelling. In most regions of the United States, the standard models are still aggregate models of trip generation, trip distribution, and assignment, with a discrete choice model of mode choice that is applied as an aggregate model, after being estimated as a disaggregate model. The methods for trip generation (cross-classification and regression) and trip distribution (singly- or doubly-constrained gravity models) have changed little at all over the past 30 years. The discrete choice model may now be a more sophisticated nested logit model, although most still tend to be simple multinomial logit models. Assignment has probably undergone most change, with the old all-or-nothing and capacity-restrained assignments largely being replaced by equilibrium assignments. The data demands of these models have changed relatively little, although there is a growing trend towards reducing the size of traffic analysis zones, so that the demands on geographic precision of the data are increasing. Emerging models are another matter altogether. The cutting edge of the state of the practice is represented by probably three different types of models: simulation models (e.g., TRANSIMS, Rilett, 2002), activity models (e.g., AMOS, Kitamura et al., 1996), and tour-based models (e.g., New York MTC). More details of the state of the practice are provided elsewhere in this issue, and the reader can find an excellent summary of activity models in Veldhuisen et al. (2000). Each of these classes of models has increasing demands on the data. Of course, all of the models, whether cutting edge or not, require accurate and complete data, and will suffer from biases that are likely to be present in data that have low response rates, and where specific segments of the travelling public are missing. However, the newer models also have greater demands on the nature of the data collected. There is an increasing desire to move to point-geocoding, i.e., geocoding to the latitude and longitude, so that disaggregation below current TAZ levels is possible. Second, tour-based and activity models are each much more sensitive to missing trips and missing activity locations. There is increasing interest in looking at multiple activities taking place at the same time, in obtaining data on the duration of activities, and in looking at a wider range of activity definitions. In 1996, the NCTCOG survey used 26 activity codes (Goldenberg et al., 1995). A statewide survey in Michigan, currently being undertaken, uses 17 activity codes for trip ends. More information is desired on parking, as well as on trip segments (walk to the bus, etc.). An emerging area in travel demand modelling is that of process models, rather than outcome models. Process models are models that are based on the processes by which people make choices, rather than being focused on observed choices, which may have come about through the operation of a range of opportunities and constraints, as well as underlying behavioural processes. Some of the earliest investigations of decision processes in travel demand modelling are those of Jones (1979), using the Household Activity and Travel Simulator (HATS), and then followed much later by Doherty (2003), who devised the CHASE process to examine how planned activities change over time as one approaches the time to undertake them (Doherty and Miller, 2000). Data collection for travel demand models has largely focused on outcomes. However, the need for process data has recently received increasing attention (Bradley, 2004). The design of surveys to elicit behavioural processes has largely not been attempted, beyond the work of Jones and Doherty, but seems likely to require a significant increase in respondent burden. Bradley (2004) provides some lists of possible types of questions that would be needed, but also acknowledges that the respondent burden, as well as the interviewer burden, is likely to be very high in such surveys. Thus, it appears that this direction of potential modelling is likely to generate a much higher demand on respondents for information, and also to require much more in-depth survey processes and thinking on the part of the respondent, who is now challenged to why he or she made the choices, which choices were rejected, and how the decisions were made. Another direction that influences the relationship between models and data is the increasing trend to private–public partnerships in the provision of transportation infrastructure and services, and the complete privatisation of some facilities, such as toll roads. When private investment is involved in the provision of public facilities and infrastructure, there is a different level of scrutiny that occurs with respect to the likely use, and the generation of user fees (Baez, 2004). Essentially, the forecasts must be more accurate, and such measures as value of time achieve a much greater importance than has been customary in government investments. This

370

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

has implications for the travel demand models, which in turn imply that the data for estimating models must be more rigorously collected and must meet greater accuracy requirements. It is inevitable that these increases in data demands, coupled with already falling response rates, must lead to greater difficulty in obtaining the quality of data that is desired. Consequently, it seems appropriate to ask the question of where we are going in the area of data collection for the purposes of estimating travel demand models. 3. The diary survey Perhaps the first conclusion one might draw from the previous sections of this paper is that the diary survey is rapidly likely to become a thing of the past. The diary survey first appeared in the late 1970s, with a booklet design developed by Socialdata in Germany for the Kontiv survey (Bro¨g et al., 1983). This design was subsequently modified and introduced in the US in 1982 (Bro¨g et al., 1983), and used in surveys in the Detroit region and in Honolulu in the early 80s. Essentially, this was a trip diary, that asked people to report each origin-to-destination movement that they undertook for a day. At the same time, the idea of a prospective diary was introduced. Prior to this, all household travel surveys had tended to use face-to-face interviews, and to ask people to recall their travel for the previous day. With the introduction of the travel diary, this shifted to setting a date in the future as the ‘‘diary day’’. The idea behind this was that people tend to forget some trips that they make, so that the recall or retrospective method was assumed to lead to underreporting of travel. By setting the date in the future, it was argued that people could actually take their diaries with them on the diary day, and fill out their travel as they did it. Indeed, most of the trip diaries included a ‘‘Memory Jogger’’ for that purpose. This was an abbreviated version of the diary that allowed people to write down the start and end time of each trip and the purpose of the trip, as the trip was done, encouraging them to fill in other details later in the day, whenever it was convenient to do so. The trip diary became the most popular design for household travel surveys in the 80s, and continues in use to the present (Harvey, 2003). A variant on the trip diary was introduced in the 1990s, and is usually referred to as a place-based travel diary. It is basically a hybrid between an activity diary (discussed below) and a trip diary. It proceeds by asking respondents where they went to next, and then asks how they got there. This variant is quite popular in current and recent surveys. It is the basis of the Sydney Continuous Household Travel Survey (Battellino and Peachman, 2003), and is also being used at this time in a statewide household travel survey in Michigan. In 1990, the travel diary was modified by Stopher (1992) to an activity diary, in which the basic change was to re-order the questions. In the activity diary, instead of asking first the question of what trip was made, and later to ask about the purpose for the trip, the activity diary started out by asking the respondent what was the next thing that he or she did, and followed this by asking about how they travelled there. It differs from the place-based travel diary, in that the focus is on what the respondent did, rather than on the places where he or she did those things. Activity diaries can also be used more readily to find out about multiple activities done at one place (not easily done in the place-based survey) and also can be used to find out about in-home activities (Harvey, 2003). Activity diaries appeared in a number of surveys subsequently, including the Wasatch Front (1993), Southeast Michigan (1994), Oregon and Southwest Washington (1994), Oahu (1995), and the New York Metropolitan Area (1996). Activity diaries continue to be used (Harvey, 2003). Anecdotally, the trip rates derived from these diaries appear to be higher than those found from trip diaries, generally being as much as 20% higher. Most agencies, however, have been wary of using activity diaries to obtain information about in-home activities, although the Oregon and South West Washington survey probably achieved the maximum detail on in-home activities of any household travel survey to date. In 1995, the North Central Texas Council of Governments pioneered a time-use diary (Goldenberg et al., 1995). The primary difference between the activity diary and the time-use diary is that travel is treated as another activity, rather than as a means to reach an activity. The time-use diary has been used in several research efforts subsequently, including research conducted at Louisiana State University (Stopher and Wilmot, 2001). However, it has not proved popular with agencies conducting household travel surveys. On the other hand, there is a substantial body of research dealing with time use (Harvey, 2003) that has a significant overlap into the household travel survey arena.

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

371

While the diary has undergone significant evolutionary change over the past 20 years, mainly in an attempt to improve its reporting capability, and to reduce respondent burden, non-response rates have continued to rise. Most recently, the accuracy of diary results, as noted in the introduction, has been called into question by the parallel use of GPS devices to check the results of the diary surveys. Considering the reliance of the diary survey in North America on the telephone, the increasing expense of face-to-face surveys, and the problems of achieving high response rates by postal surveys in many countries, the future of the diary survey looks rather bleak. One must question whether or not the diary is capable of providing the increasing accuracy and detail required to support contemporary and emerging models, whether adequate response rates and representativeness can be achieved from diary surveys, and whether increasingly pressured lifestyles will result in even more bias in the results of diary surveys as only those who have the time to do them will complete them. It is always somewhat dangerous to predict the demise of a particular methodology, in this case, that of the diary survey. In 1973, Lee (1973) wrote of the demise of large aggregate four-step models of travel demand. However, those models have shown themselves to be particularly resistant to the morbidity that was suggested for them. Almost 30 years after their requiem was suggested, they remain as the mainstay of most metropolitan modelling efforts throughout most of the world. While new models and paradigms have been proposed and tried, the aggregate models have still retained much of their place in the planning process. Using this lesson, we are reluctant to forecast the demise of diary surveys. However, it seems clear that there are problems with the diary survey and that there is a need to look increasingly at possible alternatives to it. 4. Future directions for data collection While the diary survey will certainly not disappear in the next few years, there are a number of promising potentials for replacing our current reliance on the diary survey. Some of these surveys may be best in combination, as we explore further in this section. Some of the alternatives that are considered here are the use of GPS as a supplement and a replacement for diaries, panel surveys, continuous measurement surveys, and data fusion. 4.1. GPS surveys Clearly, one of the promising avenues for survey research is that of the GPS survey (Wilson, 2004, p. 73). This began with the proof-of-concept experiment run by the US Department of Transportation in Lexington, Kentucky in 1996 (Wagner, 1997). Since then, it has evolved significantly (Wolf, 2004a; Stopher, 2004) and there have now been a substantial number of surveys that have been conducted in the US, the Netherlands, and Australia, at least, all using GPS devices to record travel of individuals. An excellent summary of these approaches can be found in Wolf (2004b). In evolving work in both the US and Australia, there are probably three exciting developments occurring in the potentials of GPS surveys. First, wearable devices are becoming smaller and more easily carried. In current developments, a recording, passive GPS has been developed that is the size and weight of the average mobile telephone (Stopher et al., 2005b). This has also substantial potential storage capacity of up to 1 Gb of data, although current versions use much smaller memory than this. Battery power is supplied by a battery of the same type as used in mobile telephones, and is easily recharged, in just the same way that people are accustomed to recharging their mobile telephones. One version of this device is available that is also a functioning mobile telephone. Second, there is increasing evidence that the GPS data can be used to provide more than just the time, speed, and position of the user. In her early research into the use of GPS, Wolf showed the possibility of deriving trip purpose from a detailed GIS of land use (Wolf, 2000; Wolf et al., 2001). Subsequently, work has been done to collect information on the workplace addresses, school addresses, and frequently-visited shops that permits trip purposes of the majority of trips to be identified from the location visited, with most of the remaining trip ends being identifiable from land use data. Home, work, shop, and school generally comprise more than 55% of all trip ends, so that less than 45% of trip ends are left to identify from land use data. Vehicle occupancy by family members can be derived when all members of the household carry wearable devices. Also, using wearable devices, work at the Institute of Transport and Logistics Studies has shown, so far, that mode of travel can be identified very accurately, using speed and route information (Stopher et al., 2005c).

372

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

Thus, with bus routes included in the GIS, together with rail stations, and other terminal points for public transport, bus, train, and ferry trips can be identified readily. Waiting for a public transport vehicle, following a walk to the stop location is also readily identified. Therefore, equipping all members of a household with wearable devices provides rich information on mode, purpose, time, duration, route, origin and destination locations, and private vehicle occupancy by family members, without recourse to interactive PDAs, or subsequent data collection activities. The Institute of Transport and Logistics Studies at the University of Sydney is pursuing research on a fully passive GPS protocol that will require only a short recruitment interview, and otherwise is able to provide the majority of data normally collected in a diary. Third, one of the major problems with GPS devices is signal loss or serious degradation of the signal in various circumstances, including tunnels, urban canyons (which usually occur in the CBD where accurate data may be particularly important), heavy tree canopies, and in certain types of vehicles, and the loss of information due to position acquisition delays at the commencement of a trip. Here, research is currently underway to combine the GPS with other positioning methods such as dead-reckoning devices and mobile telephones. For example, it is now well-established that position can be determined to within ±40 m from mobile telephone networks. Using this capability to fill in when the GPS signal is corrupted or absent, while not providing the precision of the GPS positioning, does have the potential to provide significant improvements under signal loss situations. As this research proceeds, it appears likely that GPS records will be more complete than has been the case in the past. At the same time, however, the latest antenna/receiver available is capable of recording inside buildings, acquires position in 5 seconds or less, and also gives more complete records in public transport vehicles and through urban canyons than has previously seemed possible (Stopher et al., 2005b). A significant advantage of GPS surveys is that they make it readily possible to collect data over many days. For the most part, diary surveys have been limited to one day. This, at least, has been the standard in US applications. In a few cases, two-day diaries have been used, although it is always noted that there is a drop off in reporting on the second day. Therefore, if the first day is already under-reporting travel, the second day will provide an even greater level of under-reporting. There are a few instances of diaries being collected for a week, and one instance in Europe (Axhausen et al., 2000) of a diary survey that required respondents to complete diaries for six consecutive weeks. Such instances are very limited, and there do not appear to be recent instances of more than a two-day diary anywhere outside Europe. In contrast, the GPS survey is not subject to respondent fatigue, or reporting drop off in second and subsequent days. Indeed, on the contrary, as people become used to carrying the device with them, it may actually be easier to collect data for two or three weeks than for one or two days. The other advantages of GPS surveys are obvious. The devices provide very precise geography of the beginning and ending points of travel, and also provide detailed data on the route used (data that have hitherto not been feasible to collect). The ability to pinpoint the actual origin and destination of travel may make it possible to abandon the use of the traffic analysis zone as the basis for analysing urban travel, and move, instead, to continuous representation of space in our travel demand models. The devices also provide extremely accurate information on the time when the travel took place, and the duration of the travel. With geographic location and time recorded so precisely, it is also possible to obtain very accurate speed data (Bullock et al., 2003). Information can also be obtained on acceleration and deceleration, thus potentially providing the ability to estimate much more precisely the emissions characteristics of a vehicle movement. One of the detractions from GPS surveys is their expense. At present, GPS devices are fairly expensive, with passive devices, capable of storing many days’ worth of data, costing on the order of US$750 each. However, one of the unknowns here is the effect of collecting multiple days’ worth of data on the sample sizes required for modelling and other applications. A week of data will generally require the device to be deployed for 10 days (one day for delivery, seven days for data collection, one day for return, and one day for data downloading and preparation for re-use). Assuming an average household size of 2.6 persons per household, and that each person carries a personal GPS device, one household requires 26 device days for measurement. Typical household travel surveys currently are conducted over two time periods of the year, usually from about March to June and September to November, amounting to a total of about seven months of data collection. This is usually sufficient to collect data on 3500–5000 households or more. Assuming that this entails 210 days of surveying, each device can be used 21 times for a one-week GPS survey. This would mean that approximately 12

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

373

GPS devices could be used to collect one week of data from 100 households over this time period. Thus, 120 devices would permit a sample of 1000 households to be collected. As yet, we do not know how much daily variability of travel is reduced when looking at a week of data, and we also do not know whether the information from a week may improve the quality of models of travel behaviour. Clearly, to achieve the sample sizes that are customary from one-day diaries, but collecting GPS data for one week, would require over 400 GPS devices. At a cost of US$750 each, this is now an expensive survey. On the other hand, the major costs involved in executing the survey would be a recruitment activity, courier delivery and pick up of the devices, and downloading and analysis of the data. Thus, while initial capital costs are high, the actual costs of the balance of the survey are probably quite low. In Australia, we have estimated that these costs would be on the order of US$100 per completed household, including the cost of processing and interpreting the data. This cost is lower than most CATI surveys, and substantially lower than a personal interview survey. We currently use a daily rental rate for our devices of $3 per day. This would add about $70 to the cost of a completed household, leading to an overall cost of about US$170 per completed household, that is similar to, but probably a little lower than, current CATI surveys. However, the accuracy, completeness, and richness of the data would be far in excess of current diary surveys. One other disadvantage that requires further research is the potential bias in who responds to the GPS survey. However, in a study undertaken in Sydney, Australia, (Hawkins and Stopher, 2004), it was found that the only significant differences between the population and those who took GPS devices were that they were less likely to be: • • • • •

People from non-English speaking countries; Couple households with older children; Secondary school students; Low income earners; Large households.

However, these are very similar to biases that afflict diary surveys. Additional research is warranted, as the devices are used more extensively, to determine what biases exist in the samples obtained, and to develop methods to correct these biases. At the same time, current evidence suggests that the biases are no more serious than those already existing in most conventional household travel surveys. 4.2. Panel surveys Panel surveys are widely used in fields such as medicine, health sciences, economics, and entertainment. For some reason, they have not been embraced by the transport profession. A few panels exist, or have been conducted in the past, most notably the German Mobility Panel (Armoogum et al., 2004) and the Puget Sound Panel (Murakami and Watterson, 1990). In the past, the Dutch conducted a panel for several years, and a panel was also conducted some years ago in Reading, UK. The potential advantages of panels have been discussed in the transport literature (Golob et al., 1997; Stopher et al., 2004; inter alia). Nevertheless, use of panels has really not penetrated the transportation profession. Perhaps the most important reason behind this is that no one is quite sure how to use panel data in modelling. As a result, panels have been reduced somewhat to a matter of curiosity and provision of anecdotal data about change, rather than being looked upon as a serious source of model estimation data. Probably the first reason for this is that travel demand models have never been formulated to use lagged variables (i.e., variables describing a characteristic or situation at a prior time period). A Panel Survey, of course, does not represent a survey methodology, but rather is a sampling methodology. As such, it may not be immediately obvious why panel surveys are being discussed in this context. However, there are several reasons for introducing this notion as a response to the declining productivity of the diary survey. Before discussing those reasons, there are three aspects of panel surveys that are of interest to the discussion – rotating panels, split panels, and special access panels. In addition, we should note that regular panels also come in various designs. The two principal designs are the subsample panel, in which each successive wave represents a subsample of the previous wave, as a result of attrition, with no attempt being made to make

374

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

up for the lost panel members. The second design is a partial replacement panel, in which the respondents lost to attrition are made up by additional samples at each wave, so as to maintain the sample size at the same level. This is also referred to as a refreshed panel, because the sample is refreshed at each wave. Rotating panels are a special case of a refreshed panel and are defined as panels where respondents are recruited to remain on the panel for only a limited time. After that time, respondents are rotated off the panel and replaced. As an example, a 3-year rotating panel would recruit respondents to remain on the panel for 3 years. To start the panel, the initial cohort may be made up of one third of respondents who will remain only for 1 year, one third who will remain for 2 years, and one third who will remain for 3 years. At the end of the first year, those respondents who were recruited for only 1 year will be rotated off, and replaced by respondents who are recruited for 3 years. Similarly, at the end of the second year, those who were recruited for 2 years are rotated off, and replaced again with respondents who are recruited for the full 3 years. Thereafter, each year, one third of the panel is rotated off and replaced by respondents who will remain for a period of 3 years. Practically speaking, the rotation is usually made up partly of households that take themselves off the panel (through attrition) and partly by rotating off any additional panel members up to, in this case, one third of the panel. The value of the rotating panel is that it places a limit on participation, and thereby probably suffers from less attrition than would be the case in a regular panel and, second, that the panel can be kept continually updated to the population it is intended to represent. The only large-scale rotating transportation panel in the field is the German Mobility Panel, which has recently completed 10 waves (Armoogum et al., 2004). Table 1 indicates how the Panel has evolved over time and in particular how attrition has been reduced through the use of this method – the 2000 figures are explained by expansion of the survey to the former East Germany and specific problems in the first year this was done. In a split panel, the split is between a cross-sectional survey and a panel (Kish, 1985; Raimond and Hensher, 1997). At each wave of measurement of the panel, which may be either a subsample panel or a refreshed panel, a separate, non-overlapping cross-sectional sample is also drawn and surveyed. This provides greater Table 1 Sample sizes and sample development for the german mobility panel (adapted from Armoogum et al., 2004) Recruitment Cohort 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Annual sample size Total rotated outa % Remaining for 3 years Total attrition % Attrition a

Survey Wave Panel Attrition Panel Attrition Panel Attrition Panel Attrition Panel Attrition Panel Attrition Panel Attrition Panel Attrition Panel Attrition Panel

1994 517

517 – – – –

1995 297 220 447

744 0 – 220 43%

1996 161 136 203 244 1123

1487 161 31% 380 51%

Note, these are rotated out after completion of the wave.

1997

173 30 837 286 513

1523 173 39% 316 21%

1998

632 205 364 149 504

1500 632 56% 354 23%

1999

296 68 402 102 1188

1886 296 58% 170 11%

2000

294 108 760 428 553

1607 294 58% 536 28%

2001

623 137 461 92 970

2054 623 52% 229 14%

2002

309 152 764 206 696

1769 309 56% 358 17%

2003

617 147 524 172 856 1997 617 64% 319 18%

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

375

accuracy about the changes occurring in the population, but is also a very expensive design, in that the crosssectional sample must still be fairly large, and the panel size cannot be reduced significantly. A variant on the split panel is where the cross-sectional survey is conducted much less frequently than the panel waves. However, this loses much of the benefit of the split panel, and is useful only as an occasional check on the make up of the panel. A special access panel is a panel set up with the purpose of conducting (potentially) several different types of surveys on the same group of respondents. It is widely-used in consumer behaviour and marketing studies and typically involves recruitment through the use of incentives. Such an approach is currently being used to try to gain greater insights into the different facets of scheduling behaviour for 271 households in Toronto, Canada (Roorda and Miller, 2004). In this study, participants are surveyed three times using different survey techniques and instruments, each of which is trying to uncover a particular component of scheduling and travel behaviour. Special access panels may also be drawn to provide information on particular subgroups of the population, such as bicyclists, public transport users, carpoolers, etc. The value of the special access panel is to provide detailed measurement of special subgroups of the population, and especially to monitor changes occurring in those subgroups over time. They are particularly useful when the primary concern for measurement is what is happening in specific small subgroups of the population, and where changes in the majority of the population are either of little interest, or are well known from other measurement procedures. The point of this elaboration on panels is to suggest that a combination of a split and rotating panel for national measurement of travel behaviour has enormous appeal, and could potentially provide a mechanism to overcome some of the growing problems of multiple independent cross-sectional samples in metropolitan areas, including a move away from reliance on the diary survey. While we have observed that there is a problem of declining response rates to standard household travel surveys, it is also apparent that, provided with the right incentive, many people will still respond to a survey. Therefore, one of the directions that may be well worth pursuing is the use of national paid panels. Such panels are common in a wide variety of application areas, but not currently in transport planning. How much it would be necessary to pay people to participate in the panel is unknown at this time, and would require research. However, in Australia, currently, we are using some short-term (3-year and less) GPS panels, where no incentive is offered to participants. Preliminary evidence suggests that the attrition from such panels is relatively minor. Reliance on a panel alone is probably not useful partly because of insufficient sample size for modelling activities, and partly because of the dangers of the panel becoming unrepresentative of the entire population. Therefore, the adoption of a split panel is appealing, because it contains a cross-sectional element that will continually check the representativeness of the panel, and will allow appropriate adjustments to be made in the panel rotation. We elaborate on this notion in a later section of this paper. It should also be noted that the recent review of the Bureau of Transportation Statistics’ Surveys recommended research on panels (Wilson, 2004, p. 75), as one of the directions for improving transport surveys. Probably the most important component of data that should be collected from a panel is travel/activity data, because it is changes in travel patterns that are of most concern and interest to travel demand modellers. Of course, it is relatively easy to continue to collect updates on sociodemographic data of the households in the panel. Given the earlier discussion, one obvious thought that springs to mind here is to set up a panel that will provide GPS data on an annual basis, along with updates to sociodemographic data at each wave. Changes in vehicle kilometres of travel (VKT), mode use, and the amount of walking and bicycling are probably among the most important indicators that a panel might be used to collect. 4.3. Continuous measurement Most household travel surveys are conceived as one-shot surveys that are undertaken at intervals of 5, 10, or even more years. Kish (1965, p. 477 et seq.) argues for continuing surveys on both statistical and economic grounds. Similar arguments have been put forth recently in studies to determine if the US Decennial Census should move to a continuing survey (Edmonston and Schultze, 1995). So far as the authors of this paper are aware, only Australia has embraced the continuous survey as a mechanism for household travel surveys. The Victoria Activity and Travel Survey (Transport Research Centre, 2003) and the Sydney Household Travel

376

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

Survey (Battellino and Peachman, 2003) have been conducted as continuous surveys for some years. Western Australia has also considered adopting the continuous household travel survey, as have other locations in Australia. The principle of the continuing survey is that sampling is undertaken on a continuing basis with the data providing rolling averages over some pre-defined period of years. For example, the Sydney Continuous Household Travel Survey samples approximately 5000 households each year. Data from these households are pooled for a period of 3 years, and after completion of each year’s survey, averages and totals from the sample are updated by dropping out the oldest sample and adding in the most recent one. Thus, at any time, the survey provides data on about 15,000 households, measured over a 3-year period. In addition, unlike many other household travel surveys, the sampling for the Sydney HTS is conducted throughout the year. Therefore, approximately one-twelfth of the sample comes from each month of the year. The sampling is spread geographically, so that all geographic regions are sampled each month, thus preventing any biasing of a region to a particular time of the year. As noted by Kish (1965), continuous surveys generate many organisational and economic advantages, including: ‘‘. . . adequate survey staffs in the office and in the field, accustomed to working as teams and experienced in their special problems . . . savings from merely distributing over many surveys the costs of expensive sampling materials, such as maps, instructions, forms, manuals. These advantages and savings are substantial; they may be properly allocated with cost accounting, similar to investments in large machine tools for mass production.’’ (p. 477) There is also the added benefit for government agencies that the continuous survey becomes a more or less fixed expenditure year after year, and is therefore easier to budget as a constant expenditure. It also tends to reduce the scrutiny sometimes given to periodic surveys that require a massive budget appropriation on an infrequent basis. It is often the case that there is little political memory of the last survey when the next one must be budgeted, thereby making it more onerous to argue again for the expenditures required to undertake the survey. At the same time, however, continuing data must meet the requirements of modelling to be useful. During September 2005, there was a lively debate conducted on the CTPP list serve about the 2004 American Community Survey (ACS). The general conclusion of this debate was that the ACS data are not useful to transport planners and are seriously flawed. To quote one of the participants in this discussion: In summary, there are several serious problems in the ACS program, including: 1. The errors in the annual ACS data for 2000–2004 are very large and the data cannot be used to make rational conclusions in transportation planning. 2. Like 2004, the current 2005 ACS (Full Nationwide Implementation) for areas with 65,000 plus population and areas with population between 20,000 and 65,000 will be useless because of the reasons mentioned above. It will not produce reasonable data as promised by the Census Bureau. 3. The Census Bureau is promising to produce zonal (TAZ) data for transportation planning after accumulating zonal data for 5 years after 2005. Such data will not be comparable to the decennial census because of many reasons including the ACS sample is smaller than the decennial census, does not include group quarters population, and is weighted to an estimated population rather than census counts (Zakariah, 2005). The general consensus of the list serve discussion was that the ACS would be a significantly inferior product to the old Census long form data, produced every 10 years, and that such data would not assist transportation planners in most of their endeavours, especially relating to modelling. 5. Data fusion The costs and associated difficulties of conducting surveys have led to the intriguing question of whether it is possible to take advantage of data from disparate sources to create artificial (or synthetic) travel data for a region of interest. The potential payoff in terms of substantial resource savings for data collection for all regions regardless of size makes this a particularly appealing avenue of development. The conceptual under-

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

377

pinning for such an approach is that travel behaviour could be transferable across space and time, if we can sufficiently encapsulate the characteristics known to account for variability in travel in the data generation phase (e.g., demographics, city contextual characteristics, the transportation system, etc.). If we can, there is seemingly no reason why, for instance, we could not take travel data from say a national or regional travel survey, and combine this with known local socio-demographic and spatial characteristics to create artificial/ synthetic travel data. Such a concept has been explored through an ongoing research effort, which uses Monte Carlo Simulation techniques to generate travel data for individual households, based on demographic and locational characteristics of each household, and distributions of pertinent travel characteristics derived from a national or regional survey (Greaves and Stopher, 2000). Using the US 1995 Nationwide Personal Transportation Survey (NPTS) as the source of the distributions and Census micro-data as the source of the local demographics, the technique is able to generate synthetic travel data sets, which compared reasonably well with observed data sets in three urban locations of differing characteristics in the United States (Baton Rouge, Salt Lake City, and Dallas-Fort Worth), and Adelaide and Sydney in Australia (Stopher et al., 2004). A particularly important outcome of this research has been that marked improvements in the resulting data have been observed by using travel data from a small sample of households in the region of interest to ‘‘update’’ the distributions from the national survey using Bayesian or other appropriate techniques. For instance, recent work in Sydney has shown an update sample of as few as 500 households in combination with the 1995 NPTS data (which, it must be stressed is a non-Australian national survey), results in travel data, which are remarkably similar to that obtained from the Sydney household survey (Pointer and Stopher, 2004). To provide one example of when this might be particularly useful consider the issue of generating data for a corridor or sub-regional survey. Typically, a large regional survey may provide only a handful of observations in that area, restricting the modelling effort. However, if a sample of say 300–500 households could be surveyed in that sub-region, this could then be used to update the distributions from the larger survey. These updated distributions could then form the basis for generating the travel patterns for households in the subregion. Indeed with techniques now available to generate synthetic households (e.g., Beckman et al., 1996; Ton and Hensher, 2001), it is conceivable data could effectively be simulated for all households within the region, enabling modelling to be done with entire populations as opposed to (often unrepresentative) samples. In the work to date, the approach has focused on data needs of conventional models, namely trip rates/purpose, mode shares, departure times, and travel times. This univariate ‘‘trip-based’’ approach can clearly be extended in various ways to incorporate greater behavioural reality in the data, which is critical for emerging modelling needs. The basic need is to incorporate the interdependencies between travel choices, something which is achievable by moving away from a trip-based to a ‘‘tour-based’’ approach (Stopher et al., 2004). Several challenges exist within such an approach, such as whether tours should be sequentially or holistically generated, how intra-household constraints should be incorporated, how to simulate the geography of travel, and ultimately how we validate and assess the value of such an approach. While the focus in simulating data described previously may have originally been on the data needs of estimating existing models per se, the question must be asked as to why, if we can conceivably generate data for entire populations, can we not use the (national) data itself to estimate travel directly for a local region? Such a concept lies behind data-driven micro-simulation modelling approaches, such as the RAMBLAS modelling system in the Netherlands (Veldhuisen et al., 2000). In this system, no specific travel diary data are required – rather available population, land use, and road system data are used with a national time-use survey as the source of the activity-travel patterns for households in the region of interest. The authors themselves note that while the modeling system may not be as intrinsically appealing as other activity-based models, that aim to use economic and behavioural principles to model travel, a clear appeal of the approach is that it does not require the large amounts of detailed data these models typically require. To date these types of approaches are largely descriptive as opposed to predictive. The challenge in using such techniques directly as a modelling tool is how to build in sufficient explanatory power in the population segmentation to reflect reactions to policy changes, such as a change in public transport pricing. To date, the simulations are dependent only on demographics of the population, and attempts to introduce transport system characteristics and city size and density characteristics have been unsuccessful. However, we are confident that further research will change this situation.

378

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

6. The way ahead? Currently, a number of countries undertake nationwide surveys. In the face of declining response rates, increasing costs of surveys, and other problems, the nationwide survey may be the best means to maintain up-to-date information on travel behaviour. However, a number of redesign issues are probably necessary for all nationwide surveys currently being undertaken to be of greater value for current and future regional and sub-regional travel-demand modelling applications. Our ideas take advantage of all of the aspects that have been discussed in the previous section of this paper. The design may be described as follows: First, a paid, national, rotating split panel would be recruited. The rotation might be for 3 years, and the cross-sectional survey might be conducted every 2 years or every 3 years, rather than with every wave of the panel, or may be conducted as a modest sized continuous survey. The panel, however, would be surveyed every year. Part of the panel would also be constructed as a special access panel, the make up of which could change from time to time, as new issues arise, and new minority groups of the population become the major focus of policy or analysis. The primary method of surveying the panel would be a demographic survey accompanied by a one-week or more GPS survey. A critical issue is the form-factor of the GPS, because fundamentally it must be a device people are willing to carry around with them and in this day-and-age of concerns on personal security, it must be is inconspicuous as possible. A logical option is integration with mobile telephones, which is already an available option in countries such as Japan. The additional appeal of this suggestion is that it is likely to appeal to those segments of the population who have traditionally been the most difficult to include in surveys, such as teenagers and busy professionals. Of course, we must be cautious of infringements on civil liberties and the danger the devices could be stolen. For panel households that have been measured in a previous wave, the demographic survey would be an abbreviated one, designed to determine what changes had occurred in the household since the last wave. Because the panel will be a national panel, households that move would not be removed from the panel, unless they emigrate (i.e., move out of the nation). It is worth noting here that there could be advantages to undertaking the one-week GPS survey twice each year, because the more frequent contact with the household would provide a reinforcement of panel importance and involvement. Another alternative, that may be of some interest, would be to combine an annual GPS survey with a quarterly odometer survey of all vehicles available to the household. This would provide several advantages – more frequent contact with the household thereby helping to maintain a current contact address and keep the household in the panel, information on seasonal variations in car travel, and a method to assess national and regional growth in vehicle kilometres of travel (VKT). The GPS survey would be conducted by couriering GPS devices to and from households, and the demographic survey could be done by mixed modes, including mail, telephone, and internet or e-mail. For the cross-sectional element of the survey, we prefer the idea of a continuous survey, which would cover all months of the year, and would use a sample that is independent of the panel. The continuous survey would be of the same nature as the panel, using a combination of a demographic survey and a GPS survey, although the GPS component could be shortened to a few days, instead of a week or more. Because it is run as a continuous survey, in which data could be aggregated over a period of 2–4 years, the sample size could be relatively small, say, on the order of 5000 households per year or less. This would probably be sufficient to provide useful data for modelling purposes, especially because the panel data, which might consist of data on no more than 1500–2000 households, could be used to augment and to factor the continuous survey data. The panel would serve to illuminate changes in travel behaviour, especially those that arise in response to economic and policy changes, and could also provide time series data for more sophisticated models, such as process models. Indeed, there is every reason to suppose that the data collection from the panel could be augmented from time to time with such methods as process-based surveys, SP surveys, and other measurement procedures that would provide a rich source of data for special purposes. For any region that desired to undertake modelling requiring a much larger sample size for its locality, the potentials of simulated data become interesting. Theoretically, a sample of almost any desired size could be simulated for any region or locality, using the national panel and continuous sample as a source of the distributions for simulation, and a small sample (say 500 households) collected in the local area for Bayesian updating (Pointer and Stopher, 2004). Alternatively, a present and a number of future scenarios could be simulated,

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

379

avoiding the use of models, and working directly from large data samples to describe the expected behaviour of the population. 7. Conclusions There are growing concerns that the diary survey that has become a standard for household travel surveys over the past two to three decades may be less and less viable in the future. Among the threats to the diary survey are its tendency to rely on telephone contact for recruitment and retrieval of data, increasing nonresponse rates, increasing survey costs, and new concerns about the accuracy and completeness of the records obtained from the surveys. In the face of these concerns, this paper has explored some promising new avenues for data collection, especially the use of paid panels, supplemented with a continuous cross-sectional survey, the more extensive use of GPS devices as a means of measuring person travel much more accurately and completely, and the use of simulation of data to augment samples or to bypass modelling altogether. While the proposed new method for household travel surveys outlined in this paper is probably unlikely to be embraced warmly by many government agencies, we believe that it offers some substantial advantages over continued reliance on the telephone-based diary survey, especially in North America. There will clearly be a need to demonstrate that such a survey process can provide the data that are needed by local governments to execute the planning that they are mandated to undertake. However, such demonstration can only be achieved by taking the first step of adopting at least a scaled down version of the proposed survey process. As new travel demand modelling methods move into the forefront of our work, and as new demands are placed on data collection, alternatives to the current reliance on self-reported travel are probably mandatory. The models we develop will never be better than the data from which they are estimated, and therefore attention to the quality of the data, and the methods by which better quality data can be collected, is clearly required. References Armoogum, J., Chlond, B., Madre, J.-L., Zumkeller, D., 2004. Panel surveys, paper presented to the 7th International Conference on Travel Survey Methods, Costa Rica, August 2004. Axhausen, K.W., Zimmerman, A., Scho¨nfelder, S., Rindsfu¨ser, G., Haupt, T., 2000. Observing the Rhythms of Daily Life: A Six-Week Travel Diary, Arbeitsberichte Verkehrs- und Raumplanung, 25 Institut fu¨r Verkehrsplanung und Transporttechnik, ETH, Zu¨rich. Baez, G.A., 2004. Investment-quality surveys, paper presented to the 7th International Conference on Travel Survey Methods, Costa Rica, August 2004. Battellino, H., Peachman, J., 2003. The joys and tribulations of a continuous survey. In: Stopher, P., Jones, P. (Eds.), Transport Survey Quality and Innovation. Pergamon Press, pp. 49–68. Beckman, R.J., Baggerly, K.A., McKay, M.D., 1996. Creating synthetic baseline populations. Transportation Research A 30, 415–429. Bradley, M.A., 2004. Process data for understanding and modelling travel behaviour, paper presented to the 7th International Conference on Travel Survey Methods, Costa Rica, August 2004. Bro¨g, W., Fallast, K., Kateller, H., Sammer, G., Schwertner, B., 1983. Selected results of a standardised survey instrument for large-scale travel surveys in several European countries. In: Ampt, E.S., Richardson, A.J., Bro¨g, W. (Eds.), New Survey Methods in Transport. VNU Science Press, pp. 173–191. Bro¨g, W., Meyburg, A.H., Stopher, P.R., Wermuth, M.J., 1983. Collection of household travel and activity data: Development of a survey instrument. In: Ampt, E.S., Richardson, A.J., Bro¨g, W. (Eds.), New Survey Methods in Transport. VNU Science Press, pp. 151–172. BTS, 2002. Omnibus Household Survey Results, August 2002, July 2002, June 2002, April 2002, March 2002, Bureau of Transportation Statistics, US Department of Transportation. http://www.bts.gov/omnibus/household/2002/july/month_specific_information.pdf (accessed November 2005). Bullock, P., Stopher, P.R., Pointer, G., Jiang, Q., 2003. GPS measurement of travel times, driving cycles, and congestion, paper presented at the 6th International Symposium on Satellite Navigation Technology and Applications, Melbourne, July 2003. Cambridge Systematics, Inc. 1996. Travel Survey Manual, Prepared for US Department of Transportation and Us Environmental Protection Agency, Travel Model Improvement Program (TMIP), Washington, DC. Doherty, S.T., 2003. Should we abandon activity type analysis? In: 10th International Conference on Travel Behavior Research. Available from: . Doherty, S.T., Miller, E.J., 2000. A computerised household activity scheduling survey. Transportation 27 (1), 75–97. DRCOG, 2000. Denver Regional Travel Behavior Inventory: Describing and Reaching Nonresponding Populations – Analysis and Project Report, Denver Regional Council of Governments, Denver, CO. Edmonston, B., Schultze, C. (Eds.), 1995. Modernizing the US Census. National Academy Press, Washington, DC.

380

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

Goldenberg, L., Stecher, C., Cervenka, K., 1995. Choosing a household-based survey method: Results of the Dallas-Fort worth pretest, paper presented to Fifth National Conference on Transportation Planning Methods Applications, Seattle, Washington, April 17–21, 1995. Golob, T.F., Kitamura, R., Long, L. (Eds.), 1997. Panels for Transportation Planning. Kluwer Academic Publishers. Greaves, S.P., Stopher, P.R., 2000. Creating a Synthetic Household Travel/Activity Survey – Rationale and Feasibility Analysis. Transportation Research Record 1706, pp. 82–91. Harvey, A.S., 2003. Time-space diaries: merging traditions. In: Stopher, P., Jones, P. (Eds.), Transport Survey Quality and Innovation. Pergamon Press, pp. 151–180. Hawkins, R., Stopher, P.R., 2004. Collecting data with GPS: Those who reject and those who receive, paper presented to the 27th Australasian Transport Research Forum, Adelaide, September–October. Jones, P.M., 1979. HATS: A technique for investigating household decisions. Environment and Planning A 11 (1). Kish, L., 1965. Survey Sampling. John Wiley and Sons. Kish, L., 1985. Timing of surveys for public policy. Australian Journal of Statistics 28 (1), 1–12. Kitamura, R., Pas, E.I., Lula, C.V., Lawton, T.K., Benson, P.E., 1996. The sequenced activity mobility simulator (SAMS): An integrated approach to modelling transportation, land use, and air quality. Transportation 23 (3), 267–291. Lee Jr., D.B., 1973. Requiem for large-scale models. Journal of the American Association of Planners 39 (3), 163–178. Murakami, E., Watterson, W.T., 1990. Developing a household travel survey for the Puget Sound Region. Transportation Research Record 1285, 40–48. NCHRP, 2006. Standardization of Personal Travel Surveys. Report to the National Cooperative Highway Research Program on Project 08-37, Transportation Research Board, Washington, DC. Pearson, D., 2004. A comparison of trip determination methods in GPS-enhanced household travel surveys, paper to be presented to the 84th Annual Meeting of the Transportation Research Board, January 2005. Pointer, G., Stopher, P.R., 2004. Monte Carlo simulation of Sydney household travel survey data with bayesian updating using different local sample sizes, paper to be published in Transportation Research Record, 2004. Raimond, T., Hensher, D., 1997. A review of empirical studies and applications. In: Golob, T., Kitamura, R., Long, L. (Eds.), Panels for Transportation Planning. Kluwer Academic Publishers, pp. 15–72. Rilett, L.R., 2002. Transportation Planning and TRANSIMS Microsimulation Model, Transportation Research Record 1777, Transportation Research Board, Washington, DC, pp. 84–92. Roorda, M.J., Miller, E.J., 2004. Toronto Activity Panel Survey. A multi-instrument panel survey. Paper presented at the 7th International Conference on Travel Survey Methods. Costa Rica, August 1–6. Stopher, P.R., 1992. Use of an activity-based diary to collect household travel data. Transportation 19, 159–176. Stopher, P.R., 2004. GPS, location, and household travel. In: Hensher, D.A., Button, K.J., Haynes, K.E., Stopher, P.R. (Eds.), Handbook of Transport Geography and Spatial Systems. Elsevier, Oxford, pp. 433–449. Stopher, P.R., Metcalf, H.M.A., 1996. Methods for Household Travel Surveys, NCHRP Synthesis 236, Transportation Research Board, Washington, DC. Stopher, P.R., Wilmot, C.G., 2001. Development of a Prototype Time-Use Diary and Application in Baton Rouge, Louisiana, Transportation Research Record 1768, pp. 89–98. Stopher, P.R., Greaves, S.P., Xu, M., 2004a. Using nationwide household travel data for simulating metropolitan area household travel data, paper presented at the NHTS Conference: Understanding Our Nation’s Travel, Washington, DC, November 1–2, 2004. Stopher, P.R., Alsnih, P. Bullock, Ampt, E., 2004b. Evaluating voluntary travel behaviour interventions, paper presented at the 27th Australasian Transport Research Forum, Adelaide, September 2004. Stopher, P.R., Xu, M., FitzGerald, C., 2005a. Assessing the accuracy of the Sydney household travel survey with GPS, paper presented to the 28th Australasian Transport Research Forum, September, Sydney, Australia. Stopher, P.R., Greaves, S.P., FitzGerald, C., 2005b. Developing and deploying a new wearable GPS device for transport applications, paper presented to the 28th Australasian Transport Research Forum, September, Sydney, Australia. Ton, T., Hensher, D.A., 2001, Synthesising population data: The specification and generation of synthetic households in TRESIS. Paper presented at the 9th World Conference on Transport Research, Seoul. Transport Research Centre, RMIT 2003. User Manual Volumes 1 & 2: A Companion Document to the VATS94-VATS99 Databases. Tuckel, P., O’Neill, H., 2001. The vanishing respondent in telephone surveys, paper presented at the annual conference of the American Association for Public Opinion Research, May, Montreal, Canada. Veldhuisen, J., Timmermans, H., Kapoen, L., 2000. RAMBLAS: A regional planning model based on the microsimulation of daily activity travel patterns. Environment and Planning A 32, 427–443. Wagner, D.P., 1997. Report: Lexington Area Travel Data Collection Test: GPS for Personal Travel Surveys, Final Report for OHIM, OTA, and FHWA, Washington, DC. Wilson, J., 2004. Measuring Personal Travel and Goods Movement, Transportation Research Board Special Report 277, National Research Council, Washington, DC. Wolf, J., 2000. Using GPS Data Loggers to Replace Travel Diaries in the Collection of Travel Data, Dissertation, Georgia Institute of Technology, School of Civil and Environmental Engineering, Atlanta, Georgia. Wolf, J., 2004a. Defining GPS and its capabilities. In: Hensher, D.A., Button, K.J., Haynes, K.E., Stopher, P.R. (Eds.), Handbook of Transport Geography and Spatial Systems. Elsevier, Oxford, pp. 411–431. Wolf, J., 2004b. Applications of new technologies in travel surveys, paper presented to the 7th International Conference on Travel Survey Methods, Costa Rica, August, 2004.

P.R. Stopher, S.P. Greaves / Transportation Research Part A 41 (2007) 367–381

381

Wolf, J., Guensler, R., Bachman, W., 2001. Elimination of the Travel Diary: An Experiment to Derive Trip Purpose from GPS Travel Data, Transportation Research Record, Number 1768, pp. 125–134. Wolf, J., Loechl, M., Thompson, M., Arce, C., 2003. Trip rate analysis in GPS-enhanced personal travel surveys. In: Stopher, P., Jones, P. (Eds.), Transport Survey Quality and Innovation. Pergamon Press, pp. 483–498. Zakariah, T., 2005. 2004 American Community Survey (ACS) Data Release, Memorandum to the [email protected] list serve, September 9, 2 p. Zimowski, M., Tourangeau, R., Ghadialy, R., Pedlow, S., 1997. Nonresponse in Household Travel Surveys, Report DOT-T-98-4, Prepared for the Federal Highway Administration by the National Opinion Research Center, Chicago, IL. Zmud, J., 2003. Designing instruments to improve response. In: Stopher, P., Jones, P. (Eds.), Transport Survey Quality and Innovation. Pergamon Press, pp. 89–108.