CAR DRIVERS’ VALUATIONS OF TRAVEL TIME VARIABILITY, UNEXPECTED DELAYS AND QUEUE DRIVING Jonas Eliasson Transek AB, Sweden DRAFT 2004-08-16 Abstract In this project, we have studied car drivers valuations of travel time variability, unexpected long delays and driving in queues. We treat these three phenomena as distinct, even if they are all caused (or worsened by) congestion. We are also able to show that there is a significant difference between the valuation of ”anticipated” travel time variability and the valuation of unexpected long delays. The latter is valued higher, as expected. Further, driving in queues has a disutility which is significantly higher than can be explained by the fact that queues will increase travel time variability. We conducted a large, stated preference survey in Stockholm, covering three segments: morning travelers, afternoon travelers and business travelers. No previous such studies in Sweden have been made, but results are in mostly in line with earlier studies in the UK and the US. For example, the standard deviation in travel time are valued to around 0.7-1.1 times the value of travel time; unexpected long delays are valued to around 3-5 times the value of travel time; driving in queues are valued to around 1.5 times the value of travel time. Most valuations depend on socioeconomic factors and trip purposes, in mostly the expected way, but differences are in many cases rather small. Some results are interesting, such as the fact that men have higher values of time than women (in general), but women have higher relative valuations of travel time variability. Further, households with children have also higher valuations of delays and variability. Finally, the valuation of queue driving (relative to the value of time) is almost exactly equal for all socioeconomic groups and trip purposes, while for example the valuation of delays and travel time variability vary greatly across e.g. trip purposes. Along with the stated preference survey, contingent valuation questions are used. These confirm the results from the stated preference study, but the statistical significance is lower. We also discuss methods to incorporate measures of delays and travel time variability on standard large scale travel models and cost-benefit calculations. © Association for European Transport 2004 1 INTRODUCTION As congestion problems are growing more severe in urban regions, problems with delays, congestion and travel time variability are receiving more attention. In many cases, these problems are seen as worse problems than the increased travel time in itself. Investments in infrastructure, or other transport policies, are often motivated by the need to reduce unexpected delays and unreliable travel times. Since travel time unreliability, risk for delays and the tendency to avoid driving in queues affect people’s utility and their travel behavior, it seems natural to try to include these phenomena in cost-benefit analyses. the work presented here is a first step towards this. Neglecting to do this will introduce a bias in the CBA calculations, which will most likely underestimate the utility of investments aiming to reduce urban congestion, thus in general favoring e.g. rural or inter-urban investments. The next step consists of developing methods to evaluate how different proposed measures or investments will affect delay risks etc. The study focuses exclusively on car drivers in urban areas. However, it is most likely that the phenomena discussed here are relevant also in other settings – long-distance trips and public transit trips. It is also likely that similar valuations apply to those kinds of trips as well. Measuring those valuations, however, is not part of the present study. 1.1 Terminology We will distinguish between three related phenomena, which are all congestion-related but affect travel behavior differently: travel time variability, unexpected delays and queue driving. With these terms, we will mean the following: Travel time variability is the random, day-to-day variation of the travel time that arises in congested situations even if no special events (such as accidents) occur. If congestion is severe, this variation may be significant. The unreliability means that many travelers must use safety margins in order not to be late. In some cases, the margin will turn out to be insufficient, and the traveler will be late nevertheless. In the former case, an additional disutility (beyond that measured by pure mean travel time) arises since effective travel time can be said to be greater than actual mean travel time. In the latter case, the lateness will cause some sort of additional disutility. The social loss causes by the travel time variability is the sum of those two disutilities. We will use the term [unexpected] delays in a little more narrow sense than usual. With an “unexpected delay”, we will mean that something happens that will cause the travel time to be longer than one could reasonably have expected. Thus, this is something over and above the “usual” travel time unreliability. The distinction is important, although there is a fuzzy border, between the two, since it can be assumed that the traveler can take travel time variability into account by using safety margins whenever it is important not to be late, while © Association for European Transport 2004 purely unexpected delays are so long and so infrequent that it would be unreasonable to just use extra time margins. It turns out that the estimated valuations seem to support this distinction: the valuation of long, infrequent delays are significantly higher (per minute) than the valuation of travel time variability (small, frequent delays). The study also considers the disutility of driving in queues. There are reasons to believe that there is an additional disutility from queue-driving, in addition to the increased travel time (and also travel time variability) caused by the queues. 1.2 The stated preference method The stated preference (SP) method is a general and widely applied method for measuring valuations and preferences, and for predicting choices between alternatives or goods. A typical SP survey confronts the respondent with the choice between two choice alternatives, e.g. two travel modes for a trip the respondent will make, two alternative houses if the respondent is a prospective house buyer or two cars. The alternatives differ in one or (mostly) several respects, one of which is usually the price. The respondent then indicates which of the alternative he or she prefers. Each respondent is confronted with a series (usually 5-20 depending on the situation) of such pairwise choices. By constructing the characteristics of the alternatives in a judicious manner (a process called survey design), the relative value of the variables can be measured through statistical analysis (usually logit analysis). The values can for example the willingness to pay for a car or a house with certain characteristics, or the willingness to pay for a time saving conditional on the travel mode. It is often of special interest to investigate how these valuations vary between market segments, for example young versus old, men versus women and private versus business travelers. In the next step, the data material can be used to construct computer models for market forecasts, predicting how demand will change if the characteristics of a “good” (in a general sense) changes. This is for example applicable to changed travel times on a train line, changes in accessories of a car or changed standard of house. 2 SURVEY DESIGN AND MODEL ESTIMATION 2.1 Respondent selection Respondents were recruited from car drivers in and around central Stockholm where traffic congestion is fairly severe even by international standards all day long, but in particular during morning and afternoon rush hours. Three groups were recruited: morning work trips, business trip and afternoon trips (with mixed trip purposes). After collecting car registration numbers, respondents were called and asked to answer a survey. After a few questions (used for drop-out analysis), respondents were asked o fill out a mail survey. Most (92%) were willing to do this, and most (63%) of those accepting the survey also completed the survey. The drop-out analysis showed no significant differences between those completing the survey and the no-shows. At the © Association for European Transport 2004 end, each group consisted of a little more than 200 completely answered surveys. To be included in the survey, the respondents had to live in Stockholm and travel by car at least a few times a week. Furthermore, the length of the recruitment trip (which the SP questions would deal with) had to be between 15 and 60 minutes. 2.2 The survey The mail survey consisted of a number of questions about attitudes towards delays etc. and opinions about it, and to what extent the respondent had adjusted his or her departure time in order to avoid queues. Of course, a number of additional variables were collected, such as sex, income, car availability, who paid the car costs (private or company car) etc. The main part of the survey, however was a stated preference survey. The stated preference (SP) part of the survey consisted of 12 questions regarding which of two roads the respondent would prefer. The two roads were characterized by different travel times and travel costs and one additional variable; they were assumed to be equal in all other respects. In the first four questions the travel time varied within an interval, and the width of this interval (the travel time variability) was different for the two roads. An example is shown below. The travel time variability varied between ± 5% and ± 40% (or between ± 1 and ± 16 minutes). Road 1 Road 2 Travel time 28-52 minutes 33-37 minutes Travel cost 25 kr 32 kr I choose ROAD 1 ROAD 2 Neither, abstain from the trip or use public transit instead In the analysis, the interval was interpreted as a 95% confidence interval for the travel time. in the next four questions, the two roads were subjects to occasional delays. About twice every month, bridge openings or something similar caused long, unpredictable delays. The length of the delays were different on the two roads, however. An example of the question is shown below. The delays varied between 5 and 45 minutes. Road 1 Road 2 Travel time 40 minutes 35 minutes Travel cost 25 kr 32 kr © Association for European Transport 2004 About twice a moth there is a delay for about… 25 minutes 10 minutes I choose Road 1 Road 2 Neither, abstain from the trip or use public transit instead In the last four questions, the roads two roads differed regarding how much of the travel time was spent “driving in a slow queue” , besides their travel time and cost. It was explicitly pointed out that their was no difference between the two roads as to their travel time variability or frequency of delays. One of the questions is shown below. The share of the trip spent in queue varied between 0 and 90% (or between 0 and 27 minutes). Travel time Travel cost I choose Road 1 Road 2 22 minutes of which 2 30 minutes of which 27 min are min are spent in queue spent in queue 30 kr 17 kr Road 1 Road 2 Neither, abstain from the trip or use public transit instead 2.3 SP survey design Each of the three SP games – travel time variability, unexpected delays and driving in queues – consisted of 12 questions distributed on 4 survey types. Difference designs were used, which means that four linearly independent columns were used for each game: one for the difference between travel times, one for the difference between travel costs, and two for the right-hand side and left-hand side of the remaining variable. Travel times varied between 16 and 40 minutes, travel costs between 17 and 35 kr (~ 1.8 – 3.8 €). Design were first evaluated using two methods: the distribution of trade-off points and the “theoretical efficiency” – what standard deviation the estimates would have, theoretically (assuming a “correct model”). A pilot survey was also carried out, with satisfying results. 3 GENERAL RESULTS A number of questions regarding travel behavior and opinions about congestions and related phenomena were asked. We present a few of the most interesting results here. According to the respondents, travel times during the morning rush are about twice as long as free-flow travel times, on average. Over 80% stated that their travel time was unreliable. The mean travel time variation (according to the respondents) was about 30 minutes. © Association for European Transport 2004 Despite the quite severe congestion and the fact that nearly 90 % had flexible work hours, only about one third of the car drivers during the morning rush stated that they had changed their departure time to avoid congestion. A little more than half traveled earlier than they would have, had there been no congestion. Of the remaining two thirds, half (i.e. one third of all car drivers) said that this was because they could not choose another departure time (due to e.g. leaving children at school or that they had to be at work a given time). The remaining car drivers stated no particular reason for not adjusting their departure time. The most common answer on the question “why have you chosen to travel by car to work, rather than e.g. public transit?” was that the respondent needed the car in his or her job; around half of the respondents . The second most common answer was that the travel time as shorter than with transit, and the third most common was that the respondents had other errands along the way. About three quarters had free parking at work. This is interesting, since a sometimes proposed way to reduce congestion is to force car drivers to pay full parking costs, or at least to pay tax for the “hidden” benefit of free parking. 4 ESTIMATED VALUATIONS 4.1 Travel time variability To estimate valuation of travel time variability, we used the utility function u = αt + βc + H∆ where t is travel time, c travel cost and ∆ is half the width of the given travel time interval (i.e., travel time is presented as t ± ∆). α, β and H are parameters to be estimated. In the table below we present results in the form of valuation of travel time and travel time variability (complete estimation results are presented in appendix). Values within parenthesis are not significant. Value of value of travel time (Value of standard travel time variability ± X deviation of travel (kr/h) hours (kr/h) time)/(value of travel time) α/β H/β H/(1.96α) Morning 68 33 0.95 Business 120 (20) (0.32) Afternoon 58 14 0.59 Table 1. Valuation of travel time variability The first column contains the values of travel times. Morning and afternoon valuations are mot significantly different. The business valuation is a bit low compared to what is usually found. © Association for European Transport 2004 The second column presents the valuation of a travel time interval of ± X minutes (in kr/h; 10 kr is a little more than 1 €). The morning valuation is almost twice as high than the afternoon valuation, which seems reasonable since most many travelers want to be in time for work. A little surprising is that the business valuation is so low – it is no even significantly different from zero. The reason is probably that most of these trips had purposes that were not time-critical (not meetings, for example). The third column contains the ratio between the valuation of travel time and the valuation of the standard deviation of the travel time, often called the “reliability ratio” (Black and Towriss, 1993). To translate the presented time interval to a standard deviation, we assume that the interval was interpreted as a 95% confidence interval. In most cases, it is this ratio that can be compared across studies. Earlier studies have obtained similar valuations for private travelers (see the table below). Afternoon travelers are a little lower than most earlier studies, which may be because most earlier studies have dealt with work trips exclusively. Study Reliability ratio Remark Abdel-Aty et al. (1995) 0.35 According to Small et al., 1995 Black and (1993a) Towriss 0.55 According to Cohen Southworth, 1999 Black and 1 (1993b) Towriss 0.79 According to Bates et al. (2001) Noland et al. (1998) 1.27 According to Noland et al. (2001) Lam and Small (2001) 1.3 Small et al. (2001) 1.3 Bates et al. (2001) 1.1 – 2.2 Rietveld Black and (1993a) Towriss and Claims (without reference) that this is “typical values from the literature” 2.4 Public transit headways with 0.70 All modes and trip purposes long Table 2. Reliability ratios from the literature. 1 Apparently, the Black and Towriss report (which was only published as a working paper from the UK Department of Transport) was published in two versions, here denoted a and b. © Association for European Transport 2004 4.2 Unexpected delays To estimate valuation of long unexpected delays, we used the utility function u = αt + βc + γF/n where t is travel time, c travel cost, F is the length of the delay and 1/n its frequency. α, β and γ are parameters to be estimated. In the table below, valuations of unexpected delays are presented. The ratio commonly compared across studies is the ratio between value of time and value of delay time. Usually, this ratio lies between 3 and 5, a result our study confirms. Valuation of travel times are reprinted for ease of comparison. Value of travel Value of delay Ratio delay time (kr/h) time (kr/h) time/travel time α/β γ/β γ/α Morning 68 360 5.3 Business 120 425 3.5 Afternoon 58 228 3.9 Table 3. Valuation of delay time. Analogously with travel time variability, morning delay valuations are higher than afternoon variation, and business valuations are remarkably low. Just as before, this is probably due to morning travelers being more eager to be in time, and most business travelers being on their way to “visit another if the firm’s offices” or “collect something” (two of the available trip purposes for business trips), rather than to a meeting. 4.3 Driving in queues To estimate valuation of queue driving, we used the utility function u = αt + βc + δq where t is travel time, c travel cost and q time spent in queue. α, β and δ are parameters to be estimated. The table below presents the valuation for queue driving. Valuation of travel times are reprinted for ease of comparison. The last column shows how much the presence of a queue increases the value of travel time. This is the value that can easiest be compared across studies. Value of travel time Value of queue Increase in (kr/h) time (kr/h) value of time α/β (δ+α)/β δ/α Morning 68 96 42% Business 120 174 44% Afternoon 58 89 53% Table 4. Valuation of queue driving © Association for European Transport 2004 The valuations are not significantly different across the three segments. This seems reasonable, since queue driving is essentially a “discomfort” and could reasonably be expected to be unconnected to the trip purpose. It may seem counterintuitive that car drivers are prepared to accept around 50% longer travel times just to avoid queues. Perhaps one would expect that pure travel time (and travel cost) would decide which road one would choose, Our results are, however, highly significant and stable across all segments (apart from the three segments presented here, we have also studied various other segmentations). Several other studies have found similar results. Wardman (2002) presents a meta-analysis for several time valuations, and concludes that queue driving seems to be valued 48% higher than free-flow travel time. 4.4 Delays, unreliability and queues – just “three sides of a coin”? The distinction between “[infrequent ] delays” and “travel time variability” is obviously a bit blurred. The distinction we are after is between the normal, daily variability that is an inevitable consequence of road congestion, and the delays that are so long and infrequent that travelers normally cannot compensate with taking a “reasonable margin” when choosing their departure time. But is this distinction really meaningful? A natural question is if there is a way to investigate this question quantitatively. One way to do this is to rewrite the utility function we used to estimate delay valuations in terms of the standard deviation these delays imply. It can be shown that this implied standard deviation can be written as σ(F,n) = ηF, where η is a constant depending on n. With n = 10 (one delay every 10 days on average), we get η ≈ 0.32. It can be shown that the relative valuation of standard deviation to travel time (the “reliability ratio”) implied by the delay SP game is γ −1 θ α = α nη where θ is the valuation of standard deviation and θ/α is thus the reliability ratio. We can now compare the reliability ratios from the “variability” game” with those from the “delay” game: delay variability Morning 1.34 0.95 Business 0.79 0.32 Afternoon 0.92 0.49 Table 5. Reliability ratios from the “variability” SP game and the “delay” SP game Apparently, the delay valuations are considerably higher. This is fairly convincing evidence that it makes sense to make a distinction between (long and infrequent) “delays” and (daily and anticipated) “variability”, both for methodological and conceptual reasons. © Association for European Transport 2004 There are at least two possible reasons for this. One is delay valuations may be non-linear: long delays may be worse per minute than short delays. This may be contradicted by the fact that the magnitude of the “delays” and the “variability” were in fact of comparable sizes. Another, perhaps more plausible reason is that it may be easier to compensate for daily variability than really infrequent delays: In other words: the valuation of variability may reflect the extra cost incurred by using extra margins when punctuality is important (rather than a direct measure of the “lateness penalty”) , while the valuation of delays reflects the “pure” cost of a delay - the “lateness penalty”. As to queue driving, this valuation of seems to be higher than can be explained by longer travel times etc. in itself. In our opinion it is now well supported (also by other studies) that driving in queues is a source of extra discomfort, over and above the prolonged travel time, unreliability etc. This would support introducing some measure of queue driving in cost-benefit analyses. On the other hand, it would be reasonable to argue that also other “comfort factors”, particularly in public transit, should also be reflected in the CBA. In particular, lack of seating, train/bus standards and similar transit comfort factors should also be reflected, to avoid biasing the CBA. 4.5 Segmentation Analyzing the valuations of different population segments shows that most relative valuations (i.e. delay valuation divided by travel time valuation etc.) are stable across segments, except for a few interesting cases presented below. For example, relative valuations were not affected by age, income or the extent that the respondent felt that their travel time usually varied. That the valuations are so stable across segments increases the credibility of the results. Valuations of delays and variability The valuations of delay time and travel time variability correlate; if one is high (relative to other segments), the other is also high in general. Valuations of delay time and travel time variability and delay time are fairly similar across segments, except in a few cases: - People with children have considerably higher delay/variability valuations for afternoon trips Women have higher delay/variability valuations for afternoon trips This conclusion can be formulated more dramatically: the fact that delays and unreliability are neglected in current cost-benefit analyses means that they are currently biased towards favoring men without children at the expense of women with children. © Association for European Transport 2004 Valuations of queue driving The value of queue driving is very stable across segments. The exception is that women tend to value queue driving a little more negatively than men for private trips. (Complete estimation results for segments are available in the considerably longer Swedish version of the paper.) 5 CONCLUSIONS Valuations of travel time variability, delays and queue driving have now been study in quite a few studies. With this paper, we want to further underpin the conclusion that delays and travel time variability should be included in costbenefit analysis. Valuations of queue driving, provided similar “comfort factors” for other modes are also included) Many earlier studies, however, do not separate the phenomena. Our studies have convinced us that the three phenomena of travel time variability, long and unexpected delays and queue driving per se are really three different things. Even if they have a common cause (road congestion), consequences and valuations are different, and probably also the measures necessary to reduce them. Only a few earlier studies have investigate differences in valuations between population segments and trip purposes. It turns out that relative valuations (the ratio between the valuation and the travel time valuation) are surprisingly stable in most cases. the most striking difference are that women and people with children value delays and unreliability considerably higher during afternoon trips. This is, in our opinion, an interesting result, especially since many studies focus on morning trips to work, and on men’s’ valuations (since they comprise a considerably higher share of the car drivers). The valuations are similar to most earlier studies. In our opinion, this increase the credibility of both our and previous findings. We feel that we are now reaching a point where we can be sufficiently sure on what the valuations are to include them in formal cost-benefit calculations. The main remaining obstacle is how to do this, i.e. how policy measures, investments and changes in traffic volumes affect delays etc. quantitatively. Currently, an on-going research project at Transek is investigating this by using traffic statistics and micro-simulation. We hope to present first versions of this work fairly soon. 6 APPENDIX: COMPLETE ESTIMATION RESULTS Morning travelers Observations Final log(L) D.O.F. Rho²(0) time © Association for European Transport 2004 2341 -1299.23 7 0.1993 -0.12368 -6.4 cost -0.10916 -6.7 variation -0.11794 -3.8 delay time -0.06553 -3.5 queue time -0.05159 -7 LSM1 0.944718 -4.8 LSM2 0.940303 -3.7 VOT (kr/h) stddev/travtime delay time/travtime queue time (in addition to travtime) 68 0.95 5.3 0.42 Business travelers Observations Final log(L) D.O.F. Rho²(0) 2504 -1193.32 7 0.3125 time -0.13249 -5.6 cost -0.06606 -4.1 variation -0.04295 -1.5 delay time -0.04679 -3.2 queue time -0.05874 -7.4 LSM1 1.087088 -4.5 LSM2 1.23877 -4.1 VOT (kr/h) 120 stddev/travtime 0.32 delay time/travtime queue time (in addition to travtime) 3.5 0.44 Afternoon travelers Observations Final log(L) D.O.F. Rho²(0) 2301 -1375.73 7 0.1374 time -0.1035 -6.1 cost -0.10747 -6.6 variation -0.05046 -2.4 © Association for European Transport 2004 delay time -0.04082 -4.3 queue time -0.05522 -7.6 LSM1 0.972807 -4.8 LSM2 1.207709 -4.3 VOT (kr/h) stddev/travtime delay time/travtime queue time (in addition to travtime) © Association for European Transport 2004 58 0.49 3.9 0.53