Click for next page ( 22

The National Academies of Sciences, Engineering, and Medicine
500 Fifth St. N.W. | Washington, D.C. 20001

Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement

Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 21
21 CHAPTER 2 MEASURING AND FORECASTING THE DEMAND FOR BICYCLING INTRODUCTION able datasets. A model easy to understand, use, and explain has more value even if it is necessarily limited in its detail This chapter describes strategies to estimate the demand and precision. This method has been incorporated into the for different types of cycling facilities. Such estimates form guidelines. the basis for user travel time and cost savings as well as esti- The literature review is described as follows: first is the lit- mates of reduced traffic congestion, energy consumption, and erature on measuring demand, including some original con- air pollution. Several relatively comprehensive reviews exist tributions; second is the literature on modeling demand, that that estimate the demand for non-motorized travel. These re- is, relating demand to bicycle facilities (the team makes orig- ports range from adapting traditional transportation modeling inal contributions by developing a demand model for the applications to devising specific applications and tools. Rather Twin Cities area); and third is the demand literature discus- than simply review these existing reports, this chapter focuses sion, which evaluates many of the problems with the tradi- on supplementing the knowledge gained from these reports tional approach of predicting demand by relating demand to with new perspective and original research. underlying explanatory factors. The team presents a way to understand the actual amount of cycling based on different settings (as opposed to how it is modeled or predicted). While several surveys and datasets LITERATURE REVIEW describe the amount of bicycling in the United States and various smaller areas within it, no single effort has previously Simple and reliable tools to estimate and predict the amount reconciled the results of these different surveys and data of bicycling in a given area, and how this amount depends sources to develop a general overview of the amount of bicy- on the availability of bicycle facilities and other conditions, cling. Supplementing its own data analysis with these previ- would be useful for a variety of investment and policy deci- ous efforts, the team reconciles several seemingly conflicting sions. However, while the desirability of such tools is gener- survey results and sets bounds on the amount of bicycling ally recognized, and there have been a number of efforts to that occurs in various geographic areas. The team uses this model demand either specifically or generally, no modeling as a basis for a simple sketch planning model for bicycle technique or set of parameter values or even rules of thumb planners to estimate demand in local areas. have emerged as definitive. Measuring the amount of bicycling The team also presents a detailed model to predict the occurring is an inexact science. amount of cycling relative to cycling facilities in the cities A good first step in thinking about how to model bicycling of Minneapolis and St. Paul, Minnesota. This analysis helps demand is to understand the types of questions that the model advance the state of the art beyond simply describing the might be used to answer. Porter, Suhrbier, and Schwartz (41) techniques for demand modeling to evaluating how these list three major questions, paraphrased as follows: techniques can be reasonably applied by a planner seeking reasonable results with limited resources. A principal utility How many people will use a new facility? of this exercise was to present many of the difficulties asso- How much will total demand increase given an improved ciated with practices of predicting demand. Such difficulties facility or network? illustrate the limitations of applying traditional demand mod- How does bicycling affect public objectives such as eling applications. reduced congestion and better air quality? The detailed bicycling demand analysis led to conclu- sions regarding the most practical ways of measuring and This question could be added: What are the total benefits predicting demand from the standpoint of a planner work- that bicycling creates, including the benefits to cyclists them- ing with limited resources and data. Based on these conclu- selves, such as improved health and recreational opportuni- sions, the team developed a draft sketch planning method for ties? The answer to this and the previous questions could measuring and predicting bicycling demand.i The method be useful in justifying public spending on bicycle-related pro- develops ranges of estimates from limited and easily avail- jects. The answers to the first two questions are likely to be

OCR for page 21
22 more useful for technical analyses to prioritize projects the Twin Cities area. The team does not describe existing given limited resources. literature in depth. It has already been well covered in recent Another way of approaching the problem is to note that reports. It does address applying these models to practical there are three different demand prediction objectives: demand estimation, especially in situations where a planner is constrained by limited data, time, or technical expertise. Predicting the total amount of bicycling in an area or on These issues are explored in depth in section 2.23, which a facility, develops an argument that the common demand modeling Predicting the marginal amount that total demand will objective to develop relationships between facilities and usage change given a change in facilities or policy, and by comparing different geographic areas is unlikely to pro- Identifying areas where inadequate facilities appear to vide useful results for a variety of reasons. These reasons are be holding the level of bicycling below its potential, as derived from the analysis of bicycling rates and from some in the "Latent Demand" approach (42). findings from the team's attempt to develop a demand model for the Twin Cities area. In principle, a model that explains the total amount of bicycling as a function of "basic" factors including demo- graphic, policy, and facility variables would answer all of Measuring Bicycling Demand these questions at the same time. Most past work has taken this kind of approach. The FHWA (43) and Texas Transportation This section describes the results of several surveys and Institute (44) completed major surveys of non-motorized mod- other measurements of general bicycling demand that have eling techniques in the late 1990s; the majority of the efforts been conducted during the last decade. The first objective here they describe focused on predicting either commute shares or is to combine the results of many different measurements to total bicycle travel by referring to characteristics of the pop- show that they are in fact all roughly consistent and to place ulation and land use of the area being considered and to some general bounds on the numbers that are likely to be observed. measure of the bicycling environment. The second objective is to demonstrate that the various mea- A second, less common type of demand prediction method sures of bicycling demand can be reconciled by a conceptual uses census commute-to-work shares, often combined with framework in which there is a distribution of bicycle riding other data, to provide an area-specific baseline of bicycle frequencies over the population (see Appendix A). usage. This can substitute for some of the unmeasured fac- Most of the information about the amount of bicycling tors, attitudes toward bicycling in community and accessi- addresses the number of people who ride bikes, as opposed to bility of neighborhoods to employment centers, that typically the number of trips or miles of riding. Because of the amount are found to have a large and often unpredictable impact on of information that is available about riding frequency, the demand in the more "traditional" models. team uses this as the measure of bicycling demand. At the end At the extreme, these represent two completely different of this section, the team addresses how this can be converted ways of approaching the problem. Use of the traditional into trips or distance calculations. approach relies on an (often unstated) assumption that rela- The surveys and other sources that address the frequency tionships between demand and underlying explanatory factors of bicycling produce a wide variety of results. Each source will be stable over time and transferable from one place to asks about a different time frame; the number of people who another. The second approach relies more directly on what is ride a bike in a week will be larger than the number who ride already known to be true about demand in a given area. This in a day. A key distinction that has to be tracked is that adults method in principle is more limited because it does not are considerably less likely to ride a bike than are children, directly relate demand to changes in the underlying environ- regardless of the time frame being considered. These two ment. However, within the short to medium time frames that groups must be studied separately to avoid confusion or ambi- most bicycling forecasts are concerned with, it is more accu- guity. This is generally not an issue with most bicycling sur- rate to base predictions on known facts rather than on theoret- veys, which tend to focus on adults. It is, however, a factor in ical and possibly unproven relationships. deriving numbers from general travel data collection surveys. Section 2.21 describes the results of several surveys and In the ensuing discussion and tables, the numbers refer other measurements of general bicycling demand that have to adults 18 years old and older. In each case, the survey sam- been done during the last decade. The central aim is to ple was randomly selected households, regardless of their describe the results of these many different measurements in propensity or use of cycling. A summary of the findings of all one place to show that they are all roughly consistent when these sources is shown in Table 6. differing methodologies are considered and to place general Davis (50) takes a different approach by actually counting bounds on the numbers that are likely to be observed. This the number of bikes on a fairly large sample of roads and bike information forms the basis for the demand prediction guide- facilities in the Twin Cities and calculating the total amount lines. Section 2.22 describes the existing demand prediction of biking in the region. In this approach, there is no informa- literature as well as efforts to develop a demand model for tion on who is biking or why, but only an estimate of the total

OCR for page 21
23 TABLE 6 Measures of adult bicycling frequency Source and Area Measure Average Range Travel Behavior Inventory, Twin Cities MSA %/day 1.4% - National Household Travel Survey, U.S. Total (45) 0.9% - NHTS, U.S. MSAs - 0.2%-2.4% NHTS, U.S. States - 0.0%-2.2% NHTS, U.S. Total %/week 6.7% - NHTS, U.S. MSAs - 4.5%-12.7% NHTS, U.S. States - 3.5%-12.4% Parkwood Research Associates (46) %/month - 16.6%-21.2% Bureau of Transportation Statistics (45, 47) %/summer 27% Parkwood Research Associates (46) %/year 37%-46% National Sporting Goods Association (48) %/6 times per 10.7% year - Minnesota DOT (49) %/ever ride 50% - amount being done (and where it is being done, to some not necessarily in a given year. Mathematically consistent here extent). Davis's approach provides a very powerful and objec- means that the fraction of each population frequency group tive alternative to the biases that are always inherent in who will ride during a given time span can be calculated using survey-based data (for example, the number of people who a simple probability formula, and the groups summed to arrive say they would consider commuting by bike exceeds by a at a population total. factor of 20 the number who ever actually do). It is also a Evidence from the TBI and NHTS, although not exactly method that could be exactly duplicated in other cities and consistent, can be interpreted to imply that the average person- towns, providing an objective baseline of how much cycling day of cycling for an adult generates about 40 minutes and actually takes place, and how it might vary by location, facil- 7 to 8 mi of riding, although there is a great deal of variation ity, and even weather conditions. Finally, this method has the around these averages. advantage over surveys of being a relatively inexpensive method for the amount of information that is generated. Some users cycle almost every day; others may only ride Modeling and Predicting Bicycling Demand once per year. The longer the time frame being considered, the more people will have ridden at least once. It is possible to The objectives in this literature review are to outline the divide the population into different frequencies of riding in a general types of models and methodologies that have been way that is consistent with these numbers derived from differ- used and to evaluate their potential usefulness to planners ent time frames. Table 7 shows an example of what such a seeking demand estimates with practical value. The team con- breakdown might look like, based on trial and error. siders three criteria that are likely to be important to planners: These riding probabilities and population frequencies are accuracy, data requirements, and ease of use. The focus is on mathematically consistent with about 1% of adults riding in the FHWA report (51) of 1999 because it is the most recent a given day, 5.3% in a week, 16% in a month, 29% in a sum- comprehensive survey. The earlier TTI report (52) provides mer, 40% in a year, and 50% "sometimes" riding, although more detail on specific models and methods but does not add to the breadth of the FHWA report. The major FHWA report documenting non-motorized TABLE 7 Cycling frequency travel estimation methods identifies five major methods. The Frequency of cycling % of adults first two of these, comparison studies and aggregate behavior models, are criticized by FHWA for their low accuracy. The 3 of every 4 days 0.1% low accuracy is derived from the difficulty of comparing one 1 of every 2 days 0.2% 1 of every 4 days 0.5% location with another (or transferring parameter values esti- 1 of every 10 days 1.2% mated in one location to another) because of the large impact 1 of every 20 days 3% of unobservable factors such as attitudes. While these meth- 1 of every 50 days 10% 1 of every 100 days 15% ods can be easy to use, and could require limited data, the dif- 1 of every 200 days 20% ficulty of not knowing the area(s) from which the demand Never 50% numbers were generated severely limits its applicability.

OCR for page 21
24 Comparison studies attempt to predict bike use in one area addition to, rather than instead of, auto trips; they would not or facility by measuring use in a similar area. However, it is be captured by a model that starts out by assuming that a difficult to know whether the two locations are similar. Areas given household will make a certain number of trips (as identical in demographics and land use can generate bicycling these models do). Ignoring recreational trips could be justi- rates that differ by a factor of 10 or more. Similarly, for aggre- fied if it were assumed that they had no value to society (or gate behavior models, the fact that a certain relationship exists at least a very small value compared with a "utilitarian" in one area between the amount of bicycling and certain trip) but the team is not willing to make this assumption, explanatory variables generally does not mean that the same and indeed the benefits analysis indicates that it is probably relationship will exist in other similar areas. Even if relation- not true. A good model of bicycling demand should capture ships were not consistent across places, these types of models all types of bike trips. could still be useful if the range of likely error is known and To these criticisms could be added that the accuracy of is relatively small. But this does not appear to be the case. these models is unproven in the context of bicycle forecast- The third method, sketch planning, is described as relying ing. FHWA rates these types of models as highly accurate, on data that already exist or can be collected easily, such as but it is not clear whether this rating is based on actual com- census data. This is the sort of model that is described later parisons with other models or on their complexity and high in this report. Sketch planning methods use readily available data needs. data such as commute to work shares from the census as a To supplement the research associated with this task, the tool for estimating behaviors of interest, rather than estimat- team conducted original research based in the Twin Cities ing these behaviors directly from underlying conditions. The metropolitan area. This application develops a disaggregate FHWA report rates these methods as not being very accurate, model relating bicycle facilities to the probability of an indi- although this assessment seems to be derived from the fact vidual riding a bicycle in a given day. The work is described that these methods are simple and rely on limited data. It is in detail in Appendix B. The following text provides an not clear that the relative accuracy of sketch planning meth- overview of the methods and results of this analysis. ods compared with others has ever been formally evaluated. The primary aim of the investigation was to understand the This method has been criticized for being difficult to apply effect of proximity to a bike facility on the odds of cycling. In accurately to other geographical areas. The team is not con- other words, does living closer to a bike facility increase the vinced that this is the case. While the relationship between likelihood of traveling by bike? The hypothesis is that subjects commute shares and other measures of bicycling may not be living in closer proximity to a bike facility will be more likely perfectly consistent from one place to another, it does seem to travel by bike compared with those who live more than 1 mi from the analysis to fall within a fairly limited and predictable from the nearest bike facility. range. The team believes that such methods could be quite The outcome of interest (any bicycle use in the preceding accurate, especially when supplemented with local knowl- 24 hours) was ascertained from standard travel data fur- edge and judgment. Perhaps even more important, the degree nished by the Twin Cities Travel Behavior Inventory. The of accuracy can be known with some precision. This can explanatory variable (or exposure) of interest is the proxim- make the forecasts more useful to planners hoping to do a risk ity of bicycle facilities in the form of on- and off-road bicy- analysis. And, these methods can have other advantages cle lanes and trails. Three continuous distance measures were of being very easy to use (and explain to policymakers) and calculated using global information system (GIS) layers fur- requiring limited and easily accessible data. nished by the Minnesota DOT, with separate map layers for The last two methods described in the FHWA report, dis- on-street and off-street trails. Using household locations crete choice models and regional travel models, are widely (x-y coordinates) and the GIS map layers, the distance in meters respected in the transportation profession because of their was calculated to the nearest on-street bicycle lane, the nearest longstanding application to the forecasting of auto and transit off-road trail, and the nearest bike facility of either type. travel. However, it is not clear that they are appropriate for Distance variables were used to classify subjects into one understanding bicycle travel, in part because of the significant of four categories. The four categories represent the dis- amount of data and technical expertise that is needed to exe- tance from home to the nearest bike facility as less than cute them and in part because "unobservable" factors play a 400 meters (one-quarter mi), 400 to 799 m, 800 to 1,599 m, greater role in determining the amount of bicycle travel than and 1,600 m or greater (greater than 1 mi). Given that dis- they do in either auto or transit.ii tance cut-points with relatively simple interpretation were Both these types of models are based on the assumption used, it provides a compelling way to grasp the reported that bicycle trips are made after considering a decision findings in terms of comparing individuals who live within among a number of alternate modes. However, the evidence 400 m of a bike trail and those who live more that 1,600 m strongly suggests that a majority of bicycle trips are recre- from a bike trail. Attributes of the built environment are ational in nature--a person going for a bike ride for fun theorized to influence the likelihood of cycling--namely, probably did not consider whether to go for a drive or a bus having destinations to which individuals can bicycle matters. ride instead. Recreational bike trips are probably made in Three spatial attributes were measured that are indicative of

OCR for page 21
25 one's home location--open space, regional accessibility, and This section discusses some issues with using this approach neighborhood retail. to model bicycling demand. The arguments are based in part For the sample of central city residents studied, 86 subjects on some of the facts about bicycling discussed in the previ- (4.8%) had at least one documented bike trip. This rate is ous section, and in part on some preliminary findings from higher than both the larger TBI sample and national averages, an original attempt to estimate a demand model for the Twin which tend to hover around 1 to 2% of the population (53). As Cities area. While this model is not described here, in part expected, the proportion of bikers varied with proximity to due to the lack of useful results, it is used to illustrate some bike facilities, with more bikers living closer to bike trails of bicycle demand modeling issues more generally. and fewer bikers living further from bike trails. Of interest, There are several reasons why a bicycling demand model these distributions differed depending on which measure of derived from basic information is not likely to be accurate bicycle facility proximity was used. or useful. These can be illustrated in part by the team's own A priori, it was assumed that the type of bicycle facility attempt at developing a demand model, in which was found a matters, that is, the type of bike facility may have different statistically significant result that off-road paths were associ- effects on the likelihood of bicycle use. Therefore, the team ated with lower rates of bicycling. This result is counter- used separate models to estimate the effect of proximity to intuitive. Davis (50) found that off-road facilities in the Twin off-road facilities on the odds of bike use. Examining the Cities are more intensively used than other options. The team's simple logistic regression model to the fully adjusted model result was not due to an obviously underspecified model; a wide for off-road bicycle facilities, the odds of bike use did not dif- variety of demographic and land use variables were included fer significantly by proximity to a trail. No effect of proxim- in the regressions. There are several reasons for this outcome. ity to off-road bike facilities on bicycle use was detected (see One is a possible shortcoming related to measurement; the Appendix B for details). manner in which facilities were defined did not correspond to Finally, the effect of proximity to on-road bike facilities on how people perceive them. For example, many of the sub- the odds of bike use was examined. Using a series of logis- urban "off-road" facilities run next to busy highways, with all tic regression models, it was found that subjects living within the associated crossing of driveways and roads. They are off- 400 m of an on-road bike facility had significantly increased road in the sense that there is a barrier separating them from odds of bike use compared with subjects living more than the road, but they are not off-road in the sense of eliminating 1,600 m from an on-road bike facility. As expected, those potential conflicts or of being appealing facilities on which that lived within 400 to 799 m of an on-road bike facility also to cycle. For example, it is conceivable that elaborate bicycle had significantly increased odds of bike use compared with modeling efforts would incorporate traffic volumes on major subjects living more than 1,600 m from an on-road bike facil- streets, travel times by bicycle (given traffic signals and other ity, although the odds of bike use were slightly lower than for sources of delays), crash locations, or number of street cross- those living closest to an on-road facility. ings by off-road paths. Such data are available in many met- After adjusting for individual, household, and neighbor- ropolitan planning organizations. Other issues and factors hood characteristics, the effects were somewhat attenuated. include lane width, pavement quality, and the presence of on- Subjects living near an on-road facility (less than 400 m) still street parking. These measures were not captured because they had statistically significantly increased odds of bike use com- are considerably more difficult to obtain. Proximity to high pared with subjects living more than 1,600 m from an on- traffic corridors along a route also has important implications. road bike facility. Subjects within 400 to 799 m still tended It would be useful to have information about impedance fac- toward increased odds of bike use, but this failed to reach the tors along a specific route, difficulties with the transit/cycling level of statistical significance. While not the focus of this interface, or other issues. These factors are important and the analysis, this part of the study reaffirmed that many of the fact that they were absent from the data might limit the broader socio-demographic and economic variables used in other applicability of the results. This problem is only compounded studies are important. Bicyclists are more likely to be male, to when trying to develop a model based on results in different be college educated, to come from households with children, locations because cities may have different ways of defining and to have higher income. and measuring their own facilities. The second issue with this sort of model is that there are Discussion of Prediction Methods very large and seemingly random differences from one place to another. In one area in Minneapolis, 16% of the adults Traditional approaches to modeling bicycle demand derive made bike trips on the day they were surveyed, while the rate at some level from the standard methods used for forecasting in many other areas was 0%. Even across entire metropolitan auto travel (i.e., they start from basic information about the areas the differences can be large; the metropolitan areas and people and the transportation environment in an area and use states with the most bicycling can have rates that are 10 times this in some way to predict an amount of bicycle travel, either that of the places with the least. While there are some well- directly, or as the solution to a mode choice problem in a documented population and land use characteristics that are larger travel model). associated with higher levels of bicycling, the impact of these