This section describes methods and programs that have been devised and developed to enhance the transportation planner's ability to assess the transportation impacts of micro-scale design (MSD) factors. We can categorize these methods and programs in a number of ways: aggregate vs. disaggregate; empirical vs. conceptual; in process vs. post process; probabilistic vs. deterministic; and behavioral vs. mechanistic. Most of these methods have been devised to refine the estimates of travel most important for evaluating policy actions such as those required by the Clean Air Act Amendments, the Intermodal Surface Transportation Efficiency Act (ISTEA) or the Transportation Equity Act for the 21st Century (TEA21). Many of the current models are insensitive, for the most part, to changes in policies of that nature.
Incorporating MSD elements into a metropolitan planning organization's (MPO) existing regional models may prove to be a unique process in each case. The work required to implement a change in a methodology or program to accommodate MSD elements will vary according to the current levels of detail in the networks and zone systems, the level of detail regarding travel behavior in the calibration data sets (and per force the background surveys used for calibrating the models) and the structure of the models themselves. Another major concern is the capability of the MPO to measure and describe the MSD features of a particular site in a way that will stimulate the model's sensitivities.
If design factor rating schemes are used in developing travel demand models, care should be taken to ensure that the ratings are applied fairly in the forecast year. But even if no forecast is made of the design factors, it may be better to include these schemes in the model structure and then carry the base year ratings to the future at least the analysis will "acknowledge" present-day designs. These variables, some of which are quite creative, describe both the shape and texture of the neighborhood or transit station area infrastructure. Those that are subjective measures would benefit from consistency of definition. A first step would be to evaluate the design concepts recorded by Ewing, Calthorpe and other architects to arrive at a professional consensus on terms. While design factor rating schemes probably can be used to improve travel demand models, the cost of developing this data is substantial and therefore the use of this analysis should probably be given a low priority in the development of travel demand models, depending on the urban area.
Implementation of any of the methods or programs described in this report requires:
These requirements often spell disaster for the modeler who finds that the data are not available at a usable level of detail or that the connection that seems so clear turns out to be multi-colinear with several variables. The response to certain stimuli could appear to be based on non-travel factors (i. e., they might be highly correlated but not causally related).
Increasingly, as we expand the scope of personal and family data in our travel demand databases, we encounter reluctance on the part of those being surveyed to reveal private family decisions. Even though household data and travel behavior information require thoughtful and expensive surveys, it is in the area of MSD and land use data where we have made the least progress.
Thanks to the U. S. Decennial Census, the Bureau of Economic Activity forecasts, the Census Transportation Planning Package (CTPP), the Nationwide Personal Transportation Survey (NPTS), and the Public Use Micro-Sample Data (PUMS), we have a number of reasonable sources for personal information. The typical sources of travel behavior in larger urban areas is the household interview survey or the on-board survey supplemented by surveys on truck movements and other special generator effects. Until the 1990s, land use data were captured by surrogate measures such as at-place employment or at-place retail employment. Most of the attributes of the trip and the traveler were ascribed to the household or home end of travel.
The studies reviewed for this task link travel behavior with land use and MSD variables such as age of development, distance from the regional core (central business district (CBD)), block size, and neighborhood characteristics that correlate well with travel patterns. The causal relationship is less clear. In a number of recent cases, the development follows many TOD design rules but experience little transit use. Examples include Kentlands in Montgomery County, Maryland; Carnegie Center in Princeton, New Jersey; Metropark in Elizabeth, New Jersey; and Meisner Park in Boca Raton, Florida. In these developments, isolation from the rest of the region, limited and unattractive transit service, and unsupportive mixed use appear to offset the positive aspects of higher density and pedestrian-friendly urban design elements. A package of attributes is needed to describe a TOD adequately.
MPO travel forecasters have found it difficult to model TOD, since it is not a single measure. Some interesting avenues of approach, however, would be to examine:
The following categories of programs and methods represent the full suite of travel forecasting techniques, from off-line, relatively ad hoc post-processors, to concepts that include MSD elements throughout the full travel forecasting stream of programs. Since each MPO may use a slightly different travel demand analysis process with varying levels of detail and data availability, this report includes examples from each category - some of which seem quite mundane and others perhaps too complex for application in most areas.
Implementation of MSD-responsive techniques have taken place in four different ways:
Each approach has special needs and challenges. The last implementation approach would, in all likelihood, require scrapping a four-step or n-step sequential modeling process and may be thought of as more in line with the TRANSIM program, which is based on a different approach.
For purposes of this study, post-processor techniques refer to models that modify or pivot off the results of the four-step model process. In most cases, it occurs either after the traffic assignment step or, in fewer cases, after the mode choice step. This approach includes the techniques most widely used to estimate vehicle emissions, vehicle miles traveled (VMT) and traveler response to travel demand management (TDM) actions and, perhaps, urban design options during the past decade. In this approach, vehicle trips and VMT estimates from traditional travel models are manipulated in order to estimate diversion from single occupant vehicles (SOVs), or automobiles in general to other modes or to estimate a reduction in vehicle use. The development of this approach was motivated primarily by the Clean Air Act Amendments and the search for ways to reflect TDM and other policy efforts, including growth management. The Accessibility and Transportation Impact Study (JHK, 1996) represents one of the more comprehensive post-processors that estimates reduction in travel as a function of a large number of MSD factors. The impact factors that result calculate the percent reduction in trips by trip purpose by zone to reduce the number of trips. The change in VMT is derived from the number of trips and an estimated average trip length.
The JHK Accessibility Study is summarized in Appendix C to provide the reader with qualified assessments of parameters that describe MSD elements. The scoring system converts subjective values into numerical scores and then transforms the numerical scores back into qualitative indices; this makes it difficult to develop impact factors.
The advantages of the post-processor technique includes:
The post-processor approach has some serious deficiencies that limit its usefulness. The impacts calculated as a result of the implementation of MSD features are generally not reflected in the intermediate steps of the sequential modeling process. For example, there generally is no feedback that would reflect the change in trip generation rates, distribution patterns and, consequently, traffic assignment levels reflecting policies about growth and travel behavior.
The term "pre-processor" in this case indicates an operation within the standard sequential modeling process, but before the mode choice stage. The pre-processor can be applied to the trip generation calculations or the trip distribution process as well as just before the mode choice process. The pre-processor also comes into play as part of the TDM process where the impact of particular TDM measures falls only on some small sub-population and where the sensitivity to TDM measures is not found in the four-step process in use in a particular region. An example of such a module is one that discards non-motorized trips, either after the trip generation or trip distribution step, and therefore precludes the need for inclusion of these trips in mode choice. This approach seems to be well-suited for instances where data are not available to recalibrate the mode choice model or the trip distribution model, and where the trips estimated from the re-calibrated trip generation model can be modified with the results "pruned" before proceeding into the downstream models. However, this technique, while allowing for analysis of motorized trips without having to determine the characteristics of the non-motorized travelers, also eliminates the possibility of analyzing walking and bicycle trips further in the process.
The most appealing short-run techniques to measure the impact of MSD features involve the introduction of parameters that quantify MSD features into the existing models. The most common applications of this type involve auto ownership models, trip generation models, and mode choice models.
While the attractiveness of this approach is obvious, there are also some significant challenges. Modifying the trip generation model to estimate trips that are influenced by MSD features may require an additional re-calibration and re-estimation of the trip distribution and mode choice models. Previous research has reported problems with multi-colinearity when trying to estimate models with a number of parameters that involve urban design and urban form features, all part of MSD. Modeling issues are somewhat different if transit access enhancements are separate and distinct from trip generation and distribution issues. Examples of transit access improvement modeling is provided in Appendix F, which describes the development and application of transit-friendly factors.
The most ambitious, expensive and time-consuming approach is to develop a new set of travel forecasting models that are more responsive to growth management and MSD issues. Such an approach would involve: (1) the establishment of a different planning environment that would include ways to express MSD elements, and (2) surveys to develop behavioral relations between these parameters and travel choices. In most cases, development of the model would involve new travel surveys with information or with questions involving choices that could be influenced by MSD parameters. Some pioneering work has been going on with the use of stated preference surveys and panel surveys such as the Seattle panel. While this approach is exciting, it appears to go beyond the scope of this report, except to mention the possibilities and to urge travel forecasters to plan ahead when setting up a travel survey work plan.
All the information used in travel demand analysis is stored either in networks or in polygons. The networks are the repository of information describing the impedance to moving between points in space, and the polygon is the cluster of attributes for a small or large geographical unit. If the ideal travel model reflects individual choices, then the site data need to be stored on a very small polygon, very much smaller than the typical TAZ. It also requires disaggregating demographic information. Ideally then, we could evaluate the available choices and travel decisions made by each individual in going through their daily activities.
In a like manner, through time we have tried to disaggregate the various components that describe the overall impedance to making a journey. In addition to travel times and direct travel costs, we have tried to include safety, security, comfort, convenience, and other taste variables, all of which play a part in our travel decisions. The models in current use include only a small portion of the information described above, although we are making progress in some areas. Implementing a model change, hopefully an improvement, requires evaluation of a number of issues:
For each of the MSD variables and model improvements suggested in this report, we present several alternative approaches with increasing degrees of difficulty in data collection, manipulation, and model calibration. In this way, we can envision incremental improvements to the travel forecasting process, and the possibility of implementing changes to the existing model structure that would be within the grasp of the MPO or other agency performing the travel demand analysis.
Each model used in travel forecasting presents its own specific challenges when considering the incorporation of MSD variables.
Because a great many MPOs use a sequential modeling process with four or more distinct steps, sometimes with feedback and sometimes not, this section is organized in the sequential manner of a typical travel forecasting process:
The implementation of MSD elements in the modeling steps of the travel forecasting process usually depends on the ease of use and the effectiveness in contributing to the decision-making process. Table 3-1 represents the various MSD elements discussed in Chapter 2 and compares them in terms of how well they address the more common transportation issues and how effective they appear to be in enhancing the precision and accuracy of the typical travel demand models. Each MSD element is graded from good to poor, or high to low, or from yes to no, in terms of ease of understanding by lay persons and in how complicated data collection tends to be. The ease of forecasting the various variables and estimating their impact on VMT reduction represent two of the driving forces in our desire to enhance the travel forecasting process. The remainder of the table presents a summary of how effectively each element contributes to each of the eight models presented in the table. In some cases, such as the time of day model, the preponderance of poor ratings indicates that we uncovered no current uses of the MSD elements in the time of day model or factor used by MPOs. Additional discussion of these relationships occur in following sections of Chapter 3.
Computerized highway and transit networks provide the fundamental spatial relationship information used in travel forecasting. The earliest networks contained only information about highway travel time and, until very recently, networks contained information about motorized transit and roadway transportation. Recognizing that time and distance play major roles in our travel decisions, planners recognized the need for network data describing walking and bicycle trips as well as transit and highway trips.
Travel demand models have evolved in response to analytical and policy needs. Highway networks may contain bicycle infrastructure data, although infrequently. Description of the network development for Oahu is presented in Appendix H; area types calculated from population and employment density provide guidance on free-flow speed and capacity on each highway link as a function of the adjoining land uses implied by each area type. Transit network architecture has been refined to provide the information defining the difference between walk access to transit service and drive access to transit service.
One of the most widely recognized weaknesses in the early networks used for travel forecasting was the absence of a means to describe the variation of transit accessibility for the various points in a TAZ. It generally has become a matter of acceptable practice to have a means for describing walk access and auto access to transit services.
The general methods for achieving this delineation, listed in order of increasing elegance, include:
The availability of more high-powered computing equipment now allows for subdividing the region into many more and smaller TAZs than in the past. For example, in Honolulu, the new models now being developed will include zones that are small enough to be either totally within walking distance of transit or not. MPOs developing more disaggregated models for pedestrian and bicycle analysis in the Midwest are using quarter sections as used in Chicago or even quarter-quarter sections as their small area zone system, which they contend supports a network of sufficient detail to be used for pedestrian and bicycle trips.
For those regions with sufficiently detailed and public parcel files, it will be possible in the future (and now on a limited basis) to provide even finer detail about the environment around each point in the region. There is some inertia among model builders because of the expense of assembling and maintaining files at this level of detail. But the ability to evaluate the transit accessibility for each plot in the region has obvious interest. This level of detail would also allow the models to go directly from the venue at the parcel level to the transportation network, bypassing the zone aggregation and averaging assumptions about median or mean travel times between the zone centroid and the network.
The most disaggregate methodology of all is based on a geographic information system (GIS) where attributes of place can be developed at the raster cell level, using each of the geographic layers at whatever level of aggregation is required for the data being presented by layer. The calculation of distance and accessibility to transit service can then be estimated for each traveler. This last method for evaluating TOD or transit orientation of the geographic elements of a region may fall beyond the scope of assimilation into current modeling practice, but it has exciting potential.
The cost of the methods listed above relies on geographic and geometric assessments. Uniform distribution throughout a zone assumes that the distribution of population, employment, and activities are uniform with respect to the location of transit and that the walking infrastructure (or bicycle access infrastructure) is uniform throughout the zone. If zones are small enough, this may not be too serious a problem; but larger zones commonly found in most suburban areas of medium and large regions do not have uniform development patterns. In some cases there may be large areas of low-density development with just a few high-density developments within one section of the zone. In others the clusters of commercial and residential developments may be located in areas that have either very good or very poor transit service.
Recent examples of this approach to network coding with attempts to include more detail about factors fostering walking or bicycle use include Delaware DOT; Oahu, Hawaii; and Portland, Oregon. The available data until recently have been vehicle travel times, sometimes with congestion effects, and transit travel times, sometimes reflecting congestion. Transit travel times also have included walk access and auto access travel times, but these have frequently been estimated for the average resident of a zone that is too large to be traversed on foot. Only in rare cases do the network files contain information relevant to MSD attributes. Thus, walk access is assumed to be of the same quality at each transit stop or station area. Very few forecasting models have dealt with walk access explicitly, other than to estimate the percentage of the zone that lies within walking distance. Models that used a network expression for walk access include Dallas and San Diego, which used a block system.
The MSD factors that affect highway and transit networks are found in two categories: those factors that are essential parts of the network configuration and those factors that represent wayside or roadside development. The elements that are generally used to describe attributes of network links include:
These elements allow the model to estimate the difference in speed, capacity, and relative safety of the roadway and pedestrian networks. The pedestrian network, of course, has implications for transit access.
An approach providing more sensitivity to pedestrian and bicycle traveler's needs includes calculating the density of development, either population, employment, or households and perhaps accessibility and connectivity measures to provide information about the suitability of the network for pedestrian and bicycle travel. It would also be possible to prescribe smart growth policies and other development constraints and incentives to capture the relative attractiveness of different locations as affected by their proximity to desirable activities and transit facilities.
Responses to the challenge of capturing the variation in transit access impedance based on variation in MSD/ urban form factors have usually included either creating micro-scale TAZs or stratifying the TAZ demographic data according to distance from a transit facility. Several applications of the methods used to split the TAZs whether physically or logically - have been compiled into a TMIP publication, "Guidelines for Network Representation of Transit Access, State-of-the-Practice Summary," 1994.
TAZ systems are usually based on census geography, either groups of blocks or census tracts. In most cases the zones and the associated networks are at too coarse a scale to reflect in any meaningful way most of the design features and policies that fall under the rubric of MSD. Given this state of affairs, the MPO planner is faced with using either aggregate measures (usually in a post-processor format) or MSD measures requiring substantial modifications to the databases used for travel demand forecasting.
Current practice results in TAZs that increase in size with increased distance from the regional core. Through time the number of zones in many regions has increased due both to extension of the regional cordon line outward and to reduction in zone size in the most densely developed areas. Even so, the majority of regional models rely on zones of 1 square mile or greater.
To incorporate MSD features, the analysis zone should be as small as possible. There is consensus that small zones are better than large zones for modeling purposes, but there are weaknesses and strengths in both cases. One major weakness is that land use models, in particular, and land use planning, in general, deal with large zones, and so small zones are difficult for land use forecasting. Another problem, but one of diminishing importance, is the increased cost of models as the number of zones increases. This is becoming less of a problem because of high-speed small computers. One alternative is to use a windowing system, although not generally satisfying for transit analysis.
An example of an MPO using small zones is the OMPO zone system for Oahu, Hawaii, which takes advantage of the small size of the developed area (about 400 square miles) and advances in computing power to support a system of 767 TAZs, many of which are a single block in size. Advantages include specific transit access designation; e. g., walk access to transit is designated for each zone without the need for estimation and calculation of the proportion of the zone within walk distance of transit service.
Census blocks relate primarily to physical blocks in urban areas and tracts or polygons frequently delineated by natural features outside the urban area. The use of census blocks provide a richer variation in demographic and land use data than the use of zones or even districts, which are generally combinations of zones. The use of census blocks leads to questions of confidentiality, particularly when using U. S. census data since the Census Bureau is reluctant to present data that can be identified as coming from a single household. Even when using census blocks, we are forced to use averages of demographic and land use attributes.
An even more detailed analysis can be performed if data are stored by individual parcel. Parcel data indicate the exact land use since, in almost all cases, a single parcel will contain a definable, if not singular, use. Parcel data tend to be more expensive to collect since it has to be performed at the local level. What is perhaps more of a challenge is that forecasting parcel level data tends to be difficult and may not be politically correct. A further problem arises from the fact that private sector activities frequently are kept secret and don't necessarily follow the time frame envisioned by planners. An even smaller unit of geography is that defined by the raster cell, a very small unit of geography located at a particular point and, in some cases, defined by the size of a pixel. The assumed benefit from using these very small geographic units is the ability to minimize the equivalence problem caused by the different geographic bases used to assemble data. In this way, if we know the location of a single traveler, we can collect all the geographic, demographic and system attributes. It also obviates the need for dealing with averages, but it does not solve the problem of confidentiality.
Should the selected spatial system be relatively rigid, it is often possible to modify the access coding to reflect walking access conditions. Accessibility measures can be modal or multimodal, depending on the zone-to-zone impedances. When accessibility is associated with transit time, the analyst should be very careful in defining the measure. Many studies separate zones into walk-to-transit zones and drive-to-transit zones, while other studies disaggregate individual zones into walk-to-transit areas and drive-to-transit areas. When the accessibility measure is being calculated for these studies, care must be taken to separate out the information into the walk-to-transit time, the drive-to-transit time and the cannot-use-transit. This step is usually performed using the following steps for each zone-to-zone interchange:
Travel forecasting models use a variable reflecting household auto ownership. Numerous trip generation models also use auto ownership as one of the variables. In this way, one can observe that the presence of an automobile supports trip-making while absence of a motorized vehicle constrains the choices for trips to those that can be reached on foot or by transit.
Auto ownership is an important factor in travel choices. Appendix D describes the research and development of an auto ownership model by Ryan and Han for Hawaii, which uses MSD elements to define parameters in the model. The cost and convenience of auto storage at the home location and the "importance" of owning a vehicle are based on (1) development density and (2) the accessibility of attractions by transit or non-motorized travel.
Density measures have been used to estimate vehicle ownership, household attributes, parking cost, assignment 2, and mode choice. In vehicle ownership and household attribute models, density measures are usually used to imply urban congestion. In one vehicle ownership model, the dependent variable was the share of the households by automobile ownership and the model structure was a logit model. 3 In this case, the population density was a continuous variable in the utility equation. The use of population or household density would appear to be an obvious variable in the estimation of household size (persons per household) and income.
Accessibility measures can be very useful in developing automobile ownership models and other socioeconomic 4 models. The automobile ownership model developed by Ryan and Han used a combination of the accessibility measures to develop a measure they called the "auto importance measure." This measure was the highway time accessibility measure divided by the sum of the highway time accessibility plus the transit time accessibility plus the walk time accessibility.
Auto Importance Measure = Highway Time Accessibility / Highway time accessibility + transit time accessibility + walk time accessibility
The transit time accessibility measure was calculated using only the walk-to-transit portion of the transit accessibility calculations because the drive-to-transit accessibility required an automobile and, therefore, should not be included for the automobile ownership model. This auto importance measure could range from 1.0 (where no transit or walk accessibility is available) to 0. 33 (where the three times – highway, transit, and walk – are equal). In extreme cases it might drop below 0.33 (where transit is considerably better than highway). The auto important measure was found to be an important variable in estimating car ownership; as auto importance increased so did automobile ownership.
Micro-level mixed uses should provide increases in the walk accessibility5 measure. This, in turn, will lead to a reduction in the auto importance measure and therefore to reduced automobile ownership. At the macro level, closer home and work connections may increase transit accessibility faster than highway accessibility and therefore reduce the auto importance measure.
Design factor ratings may be a reasonable variable in the estimation of vehicle ownership. It is anticipated that these ratings schemes could be used in the Ryan/ Han vehicle ownership model as a "dummy" variable. That is, a variable that is either a 1 or a 0. Depending on the urban area, the model might have a couple of these variables – one to signify if the area had a high rating (perhaps above 8 on a scale of 1 to 10) and one to signify if the area had a low rating (perhaps below 3 on a scale of 1 to 10).
Trip generation models in current use tend to relate travel to household characteristics. These models are sometimes stratified, such as CBD versus non-CBD models. This stratification accounts for MSD features of the CBD, but in an aggregate manner, so that all parts of the CBD are deemed to be identical. This is obviously not the case. In many model structures, trip generation models predict fewer trip ends for CBD or dense urban sectors than in the less densely developed areas. This is because the primary focus is on vehicle travel, either private or public, totally ignoring pedestrian and bicycle modes.
Incorporating non-motorized travel into the traditional sequential model introduces additional requirements: the trips that have been introduced into the trip generation phase have to be distributed and, if they are not going to be incorporated into the mode choice model, they need to be dealt with at some point prior to that operation. If non-motorized trips are going to be dealt with in the trip distribution step of the sequential model, it is necessary to have zones and networks that can estimate the impedance for non-motorized trips in a way that makes them compatible with the estimates of travel by vehicle. CATS has developed non-motorized modeling capability using a walk network and small zones.
The role of density in trip generation is difficult to quantify. Some studies have used density both as a continuous variable and as a cross-classification variable in both production and attraction models, with the density being inversely proportional to the trip rate (trips per household or trips per employee). But normally these models were used to estimate "motorized" trips 6 rather than total trips and the density measures were surrogates for removing the walking trips. That is, as density increased, so did walk trips with a subsequent decrease in motorized trips. But state-of-the-practice travel demand models should include both motorized and non-motorized trips (walking and bicycling). In which case, density should probably not be extremely important in trip generation.
A possible exception to this is the non-home based trips made while at work. A state-of-the-practice model should, at least, identify non-home based trips (if present) as trips made while at work, to or from work, and non-work related. For the non-home based trips made while at work, the employment density measure might be useful. An area of high density might promote more trips since there are many more potential opportunities nearby, although many of these trips might be walk trips. Thus, a person might leave the office to go to lunch or walk over to another person's office (instead of calling).
The use of accessibility measures in trip generation is difficult. Some studies have used the cutoff accessibility measure in the shopping attraction equation. In this case, the shopping attractions per retail employee increase as the number of households within the given time range increases. Similar cutoff accessibility measures have been used for non-home based trips, especially those associated with the work place. For shopping attractions and non-home based trips, the use of accessibility measures appears reasonable. These are discretionary trips and as the number of potential "customers" increases, so should the total number of trips per employee or per square foot of establishment space.
Studies that estimate travel stratified by the income of the traveler have used accessibility measures to help "split" the attractions into income groups. In New Orleans, accessibility was the number of households within 35 minutes by transit; this measure was used in the work attraction model to stratify the attractions by income group. Accessibility in this case was calculated as the ratio of the income group value to the value for all income groups. Essentially this variable said that as workers available in income group X increased, the number of jobs for income group X increased – a fairly reasonable result.
Mixed-use policies in trip generation may be more difficult to ascertain. The micro-level policy should increase the potential for the walk mode and this would reduce trip generation for a model estimating motorized trips. The decrease in automobile ownership would also tend to decrease the total travel. On the macro level, a policy of matching laborers and employment opportunities might increase some accessibility measures and therefore affect shopping trips and non-home based trips, but this is not necessarily true.
The effect of design factors is not known for trip generation, aside from the density and accessibility affects. But if a rating scheme has been developed, the analyst should be able to review the effects of this rating on trip productions by including the ratings as one of the variables in a cross-classification model or by using an analysis of variance. Similar analyses could be performed on trip attractions, depending on the observed data available.
Trip distribution is the process of connecting trip productions (origins) and attractions (destinations), thus creating a web of travel throughout a region. The fundamental theory behind most trip distribution models in the past has been that people are attracted to large accumulations of trip ends and are reluctant to travel any farther than necessary. This theory is very similar to Newton's theory of gravitation, hence the name of the most widely used model, the gravity model. In the last decade, planners have recognized that other attributes of origins and destinations play an important role in the distribution pattern and they are still working on defining the nature and extent of these attractions.
Very little has been done in using density measures in trip distribution. In some cases, the terminal times (the travel times off the highway network, including parking the automobile) are modeled as a function of densities. Since terminal times usually range from 2 to 10 minutes (with the higher values being very rare) a cross-classification model using residential and employment density would be an adequate model. This type of model has been used in Atlanta to estimate terminal times. Density measures have an indirect impact on the distribution model, since they can be used to estimate capacity and free-flow times on the highway networks. These, in turn, would affect the travel times being used by the distribution model.
Accessibility measures are very important in trip distribution models. The accessibility measure (defined as total accessibility) is the denominator in the gravity model where the travel time is defined using a set of F factors rather than by a power. In a logit distribution model, the log sum term is essentially total accessibility – again with a modification of the time function.
In Chicago, the distribution model is an intervening opportunity model. The calibration parameter for this type of model is the L value. The initial calibration of the L value was based on geography, but a later calibration effort related the L values to a cutoff accessibility 7 measure – which was the number of attractions within a certain generalized cost. This is a very interesting case where the accessibility measures modify the travel time function. Neither the F factors in the gravity model nor the coefficients in logit distribution models are made a function of the accessibility measure.
Mixed-use policies have significant effects on trip distribution. The placement of the origins and destinations of trips has a major influence on the trip length distribution of travel. This, in turn, affects the person and VMT and, potentially, the market shares for the different modes. It is difficult to specify, even in general, the decrease in travel that mixed-use policy would generate, but it can be substantial. It is suggested that an interesting policy analysis would be to specify, for a land use forecast, that each additional worker (household) should be accompanied by an additional employment opportunity (job) no more than X miles from the resident – where X is a value between 1 and 5. Indeed, it should be possible to "program" this policy and build several future land use forecasts with different values of X. Using these land use forecasts, the travel demand models could then be applied and the change in VMT determined. This analysis would allow the analyst to provide policy makers with a measure of the efficiency of mixed-use policies for their specific urban region. Since the potential of mixed use (especially at the macro level) is so great, an analysis of the affects of mixed-use policy is highly recommended as one of the initial evaluation analyses for long range planning.
It is probably difficult to include design factors in the trip distribution model with the exception of the ability to estimate transit walk access more accurately. The analyst might attempt to develop a separate set of F factors for the gravity model based on zones grouped by the ratings (perhaps three groups); or, if the distribution model's structure is a logit model, the analyst might try using the rating schemes as a set of dummy variables in the utility expression.
For most of the last three decades, travel models did not include walk and bicycle trips in their trip generation equations. The reason was actually quite simple: the trip distribution, mode choice, and assignment modules did not deal with walking and bicycle travel as primary modes. This meant that estimating walk and bicycle trips in the trip generation phase of the modeling process required removing those trips before proceeding with the analysis. The realization was that walking and bicycle use were, in fact, modes that could be substituted for auto and transit use, given the right infrastructure. For most modeling activities, acknowledgment of bicycle and pedestrian travel was accomplished in reverse; trips generated in walkable neighborhoods were generally reduced or lower in number than trips for the same purpose by households with the same composition but located in neighborhoods without pedestrian amenities.
The time of day functions in use in most MPOs are relatively crude relationships between the proportion of trips and the time of day. The percentage in each time period is frequently determined from observations of ground counts. In general, the peak hour and/ or peak period factors are regional or subregional, but rarely reflect the site design characteristics of the origin and destination zones. Investigation and further research into the time of day distribution of trips related to MSD features appears to be a most interesting research topic.
In the typical n-step travel-forecasting model, mode choice has received the greatest emphasis during the last decade. In part, this is due to long-standing observations that "transit works" where densities are above some threshold (for example, more than 10 dwelling units per acre). The following sections describe a number of interesting techniques used to increase the sensitivity of the mode choice models to MSD features. The influence of MSD on mode choice models appears to be stronger than on any of the other models in the travel demand forecasting process.
There are two examples of mode choice models calibrated on transit accessibility measures that include not only distance but also the amenities and security requirements necessary for attractive transit access (1000 Friends, 1992; Douglas and Evans, 1997). The weakness in these two methods is their reliance on subjective evaluation of transit and pedestrian environmental factors (PEFs) or transit-friendly factors (TFFs) resulting in replication and transferability issues. On the other hand, the development of subjective criteria, if carefully described and applied by local/ regional planners with intimate knowledge of the region's infrastructure and pedestrian and bicycle infrastructure attributes, can be a useful tool to explain some of the variation in transit use that's not explained by service and demographic characteristics.
Density measures have been used in mode choice models as direct variables; that is in the mode choice equations, and as in-direct variables, used as variables to estimate other measures used in the mode choice model. Some mode choice models have used density measures as continuous variables in logit utility equations, both residential and employment density. Other models have used indicators of density (such as area type) and a CBD as dummy variables so that the coefficient on the variable was in essence a modal bias coefficient. One common method is to have a CBD dummy (a 0/ 1 variable) as one of the utile equation variables with the CBD defined as having an employment density over a certain value. The use of these dummy variables, though, should be used with caution, because they contain the "cliff" feature previously discussed.
As a macro-scale measurement, density is usually estimated as an average density for a TAZ. The simplest method of making this calculation is to divide the population (or households or employment) by the area of the zone. In some cases, this can produce anomalies; for example, a park in the middle of the CBD might decrease a zone's density to a low level, or a large building in a small zone might substantially increase the zone's density. A better measure of density for a TAZ would be land use information (population, employment, etc.) and area within a certain distance of the center of the zone. For example, one-half mile and three-quarters of a mile have been used in some urban areas. This type of calculation becomes fairly easy with GIS systems. One simple method of performing this calculation was in New York where the GIS system produced a data file of the percent of each zone within a specified distance of the centroid of the subject zone. These percentages were then multiplied by the population, employment, and area of the zones to estimate the density measures. More sophisticated GIS systems can estimate these density measures using more disaggregate data.
Density measures have been used in two different ways: as a continuous variable or as a cross-classification variable, with the density measure having several classes (ranges). If possible, using density as a continuous variable is preferred over using density as a series of discrete ranges. The models that use density as discrete ranges tend to have "cliff" problems, since each range has a set of precise limits. For example, if a density step function is used in a mode choice model, then two adjoining TAZs could produce very different transit shares because the density measure changes from one group to another. This is especially true when the density measure is a binary variable (the zone is either a high density zone or it is not). If discrete ranges are used, there should be more than seven ranges and they should have a fairly minor effect on the dependent variable.
Parking cost has long been a major cost component of mode choice models and some studies estimate the cost of parking based on employment density. The reasoning is that density is an indicator of the cost of the land and, therefore, the building rent; as the cost of rent goes up so must the actual cost of parking. Typically these models are either non-linear equations or piece-wise regression equations. For example, in Atlanta, the parking cost was estimated as follows:
A travel demand model that includes non-motorized travel should have a method of estimating walk and bicycle trips in the mode choice model. The walk and bicycle model can be included as a separate mode( s) in a full share model, usually a nested logit model. In some cases, the non-motorized trips are removed from the universe of trips after trip generation:
If this type of procedure is followed, residential and employment density are possible independent variables in the model to remove the non-motorized trips. One method of using these densities would be to have a model similar to the area type model shown on Table 3.2.7-1 (except the residential and employment density combinations would be directed toward the estimation of non-motorized travel).
For example, a possible model structure would have a cross-classification similar to Table 3.2.7-1, with the independent variable being the percent of non-motorized trips (instead of a simple area type designation as in Table 3.2.7-1). This type of model would be used for the model structure that removed productions and attractions after trip generation.
For the model structure that separated person trips into motorized and non-motorized trips (simple model not a full nested logit), a similar set of cross-classification tables could be used but the model would have to include a distance component. For large urban areas, a state-of-the-art travel demand model should include a mode choice model that contains a full set of modes, including non-motorized modes. The models that remove the non-motorized modes prior to applying a full mode choice model should be considered interim models and should be used only until a full mode choice model can be developed.
Accessibility measures are usually not used in mode choice models. However, there have been some exceptions. Early mode choice models in the Houston, New Orleans, Seattle and Washington, DC regions used the cutoff accessibility measures (number of attractions within a certain transit travel time) as independent variables. The reasoning for the accessibility measures in these mode choice models was that a person would be more inclined to use transit for a specific opportunity (attraction) if there were many transit opportunities available. An interesting reason for discontinuing the use of accessibility in these models was that it was difficult to explain to policy makers why alternatives had different patronage when the alternatives had similar times but different accessibility values. For example, the travel time difference between stops on a rail line and a busway, from suburban areas to the CBD, might be very similar but the rail line – because of the intermediate stations that provide the ability to travel within the route – tends to have a larger accessibility value (the busway goes directly from the subway to downtown).
There have been some suggestions that non-home based transit trips, especially those related to work, could be estimated using direct demand models, with a major variable being a cutoff accessibility measure, such as employment within 20 minutes of transit time.
Mixed-use policies may or may not increase the use of transit in an urban area. On one hand, it should allow workers to have a reasonable distance to travel to work places which, in turn, allows local bus service to be potentially competitive. On the other hand, the policies may decrease concentrations of employment which, in turn, would decrease the potential for high volume transit modes, such as heavy rail, with a subsequent decrease in the potential for transit. Micro-level mixed-use policies should increase the potential for walk trips, especially for non-work travel. Normally the effect of the micro-level mixed-use policies would be implemented – in the mode choice models – through the use of accessibility measures or some other variables that would specify the increased benefits and potential for walking.
One of the major positive points in design features should be the design of good transit access. Covered walkways, sidewalks and direct access should be elements of the urban design, especially for TODs. The analysts preparing the travel demand models should take care to properly specify the walk impedances for "good" design areas and poor design areas. This would be in sharp contrast to the present procedures of estimating walk times based on spatial areas and assumptions of uniform densities. This type of analysis would probably rely on good GIS data sets and procedures and therefore may still be a few years in the future for many urban areas. Given that this information is available, it could be used to specify more accurate walk times and to separate the walk times into types of "walks" (e. g., covered walkways, separate walk paths, sidewalks on streets, no sidewalks). Each of these different walkways could be either assigned a different "weight" or be used in the statistical analysis to produce different coefficients. With the development of GIS data sets allowing the analyst to perform the above analysis is in the future, it may be an extremely useful analysis in the estimation of market shares.
Design factor rating schemes can be used in mode choice modeling. In this case the rating schemes would be used as a set of dummy variables in the logit utility expression. The Maryland National Capital Park and Planning Commission performed such an analysis for the Washington, DC region and found that the design factor ratings were significant variables (Replogle, 1990, 1991, 1995).
Once travel patterns have been established and modes have been chosen, the final step in travel forecasting is assigning trips to the networks to determine where congestion levels will occur. The trip assignment models start with trips that have specified origins and destinations and select a path based on minimizing the generalized costs – usually a combination of time and money. The impact of MSD features on this process is not well documented and may not be a particularly strong relationship. For repetitive trips (e. g., work trips) and medium-to long-distance vehicle trips, the path choice of the vehicle is generally not seriously affected by the MSD features at either end of the trip. There may be some impact on the path chosen if the site design characteristics of the development along the right-of-way influence roadway or transitway capacity. As with the time of day models, the impact of various MSD factors on trip assignment procedures represents what we think is an interesting and critical research topic.
Density measures have been used in the assignment phase of travel demand modeling. Typically density measures are used as one of the variables in the determination of highway capacity and free-flow speed. Density measures represent urban congestion that impacts the design of highway and performance, which is difficult to measure in the base year and probably impossible to estimate in future-year forecasts. These are items like driveways, on-street parking, truck loading areas, and design standards. In this case, density measures are typically used in a cross- classification model with density and the type of highway being the two independent variables, and the capacity per lane or the free-flow speed being the dependent variable.
Since residential density and employment density may have different, but related, effects on highway capacity, some studies have specified an "area type" that is developed as a cross-classification model. Here, residential density is used as one independent variable and employment density as the other dependent variable. In most cases, the transportation analyst classifies TAZs into a fairly wide range of "area" types; ranging from very commercial urban with substantial curbside interference to rural areas. The employment and residential densities are then reviewed to ascertain the cross-classification model that will estimate the area types accurately. An example of a cross-classification model to estimate area type is shown on Table 3.2.7-1.
The area type and a definition of the type of highway (facility type) are then used as indicators for free-flow speed and capacity. Some transportation planning packages allow this type of table to be directly input in the assignment procedure, while in other cases the analyst has to update the highway link records with the free-flow speed and capacities. An example of this type of model is shown on Table 3.2.8-1.
|Area Type||Facility Type||Free-flow Speed||Capacity Vehicles per Lane per Hour|
All the data used in travel demand modeling are stored either on networks or in polygons (called zones in most MPO models). With this information, we can describe the demographic and environmental characteristics of a point in space occupied by a single traveler and then the characteristics of a journey to another point in space. These two points are typically called "the origin and destination of trips," or the "venue for activities," or simply places. Ideally, for MSD purposes, we would like to be able to express the activity and travel-related information at the smallest practical area level, such as a single dwelling unit, plot of land, or commercial place. We also would like to be able to calculate and interpret all of the opportunities and constraints available to a traveler wishing to move from one place to another.
It is clear that each step in enhancing travel demand models leads to requirements for more data, not only a larger quantity of data, but also more kinds of data that include greater levels of intimacy. Adding stated preference surveys to the revealed preference surveys adds another level of complexity and uncertainty to the underlying data used to construct choice models.
Collecting the desired household interview survey (HIS) and site design data to include MSD features in models results in increased costs and, unfortunately, a certain level of intrusiveness into people's lives. The results of this increased intrusion are likely to include omission of activities or trips that the respondent considers private, or even rejection of the survey by some households, leading to concerns about non-response bias and something we could call "mis-response" bias. Forecasting these variables may be the most daunting task of all. MPO staff members have frequently been attacked for what is seen as invasive.
As an alternative to household surveys, there is growing interest in using global positioning satellite (GPS) tracking. Some believe that this results in more precise locations, although the individual locations still need to be tied to the activities that are taking place. As a result, there would still be a need for a fair amount of diary collection, even using GPS.
Another data issue concerns those variables, particularly at the parcel level, that may provide spurious information since their current status may be quite different from their future status. This may be particularly true where future land use patterns depend upon private actions that are kept secret for fear of commercial or economic difficulties if they are made public. For example, a developer assembling parcels of land for a major shopping mall may be reluctant to divulge ownership (frequently hidden under dummy corporations) for fear that divulging the information may raise the cost of assembling the property. In a similar vein, detailed plans by central planning boards that pinpoint specific parcels for future development may create friction between the planners and the current owners of the property.
Planners have expressed concern about the TIGER file and its lack of accuracy and precision. Apparently the longitude and latitude values result in errors of one to several hundred feet. E-TAC files, which are developed at the county level, are considered to be superior. Several analysts in New York have suggested that we warn users about match problems and discuss which GIS regional consortia have used. The MPO or transit agency can coordinate the GIS work with the local jurisdictions, working at the lowest level of detail imaginable, since it's always easier to aggregate to larger land areas or polygons than it is to subdivide at some later date. A challenge for the GIS would be to provide highway and signal data, currently lacking in GIS at most jurisdictions.
GIS data seem to be a natural for use in MSD. We recommend starting with census data, particularly the TIGER line file, as the least expensive accessible data that could be used for MSD analysis. By manipulating publicly available data it is possible to develop indexes that distinguish among MSD features among different transportation planning zones.
One of the new approaches in travel forecasting is the use of activities rather than trips to describe a household's daily pattern. The use of activities leads to a need for better land use data and in fact is virtually meaningless without it. The major difference between activity analysis and trip analysis is the inclusion of trade-offs between in-home vs. out-of-home opportunities. Coupled with these activity trade-offs, we also are moving towards including walk trips and bicycle trips or other non-motorized modes in addition to autos and transit. Not only is the neighborhood pattern important in the home end of a home-to-work couple, but the land use pattern at the work place has a major impact on the set of activities and the tours that will be made by residents of the household.
The trip decision process asks the following questions:
In the past, trip generation models ignored trips that did not take place in a motorized vehicle such as an auto or a bus, thus giving no credit for settlement patterns or policies that reduced travel by increasing satisfaction through walk trips instead of auto trips.
The data required for activity analysis is quite similar to the data collected for travel or activity diaries. The essential ingredient for activity analysis is to have enough information to estimate the trade-offs between activities taking place in the home and out of the home. In general, household surveys do not include information about MSD features surrounding the venues listed in the survey itself. This information is readily available, although not always inexpensive, through GIS.
Activity modeling as opposed to trip analysis is the wave of the future according to planners in Portland, Oregon; Berkeley, California; and other fast-growing cities. This study will not deal in depth with activity modeling. Planners will continue to deal with trips until the activity models are ready. It is important to collect activity data sets today even if we don't include them in the models. This is because:
It is essential that we recognize the value of activity data and associated data collection issues. Activity data surveys require more intrusive and probing questions that may increase the non-response rate and data collection costs (more interview time per household). There is also a trend towards multi-day surveys to capture trade-offs around activity schedules, and greater use of stated preference surveys to estimate traveler response to new, not previously experienced, stimuli. Each of these innovations and enhancements increases the costs of data collection, but since the next HIS may need to support 5 to 10 years of model development and application, careful survey design and aggressive collection seems imperative. Any other course appears to have serious long-range disadvantages.
Three prejudices exacerbate the phenomenon of overemphasizing vehicular trips and under-emphasizing the role of non-motorized travel as part of gaining satisfaction and fulfillment of needs and desires. First, travelers tend to be students, young teen and elderly demographic groups that are not considered as important as vehicle-using workers and shoppers. Second, short trips that don't consume roadway space for vehicles are seen as less interesting and of less importance; reductions in costs are not viewed with the same importance as increases in costs and, consequently, saving a trip lacks the impact of causing an additional trip. Third, trips that begin and end at home with no intermediate stops registered, such as a walk around the block, are not considered trips at all, in spite of the fact that they may be providing an important benefit. An alternative would be driving to a health club, which would be considered a trip and another activity.
Concurrently with the interest in reflecting MSD issues and attributes in the modeling process, a number of efforts are under way to enhance various components of the modeling process and to transform it from the essentially n-step sequential models used for the last several decades. Several of these efforts are critical to the incorporation of land use and MSD variables in the planning and travel forecasting modeling process. In particular, efforts should be increased in the area of activity-based models and the development of GIS capabilities for describing the environmental and cultural attributes of any given point in the region based on evaluation of a number of data layers or coverages that include infrastructure, environmental, and cultural information.
The neighborhood or settlement pattern around a residence has been seen to have a major influence on travel and the activities undertaken by residents. For purposes of this discussion, we divided households into five cohorts that influence the type of residential neighborhood chosen and the type of accessibility to jobs, shopping, and entertainment that are needed or desired:
Each of these cohorts has different requirements for space, costs, educational facilities, cultural activities, accessibility to jobs, and walkability and safety, both personal and security. The mix of attributes of different settlement patterns or neighborhoods is important because location of a residence in an outer suburb may preclude a number of activities and decrease accessibility by foot or bicycle. As a result, this type of residential location may not be attractive to young people or retirees. On the other hand, educational facilities are not of great importance to young families without children, whereas they are extremely important to families with school age children. These phenomena have two significant ramifications:
It is important to recognize the interactive nature of the transportation and land use models even though the changes may be difficult to represent in travel forecasting models. Long-term residential locations may be chosen as a function of short-run visions. This is another way of saying that the choice we make may not be applicable very far into the future. It may be that a life event such as a birth, death, divorce or marriage raises the question of whether a household will move or stay or experience a job change. The decision may frequently revolve around friends or other members of the family cohort and the location of that cohort in the life cycle. For example, a decision to move or stay when the household is composed of adults and children who are within a year of leaving for college or their own life may be quite different than the same family's decision made when the children were pre-school age. The residential location and the location of job and other activities will have a major influence on short-term or daily decisions.
In the short term, we deal with decisions that may include impulse or habitual patterns. A number of retail choices, such as fast food restaurants, non-prescription drug stores, and other relatively ubiquitous retail establishments attract customers who are making other journeys and include a stop on impulse. A number of travel decisions are the result of habitual patterns; for example, the route taken on the way to work that may or may not be the least congested on a particular day but is chosen on the basis of historical data and habit. This choice may be "efficient" in terms of the energy needed to make decisions on alternate paths.
MSD features will influence whether or not an activity will take place at home. Certain activities must take place at home or be foregone if the only other alternative requires automobile use (too far to walk and no transit available). If the only transportation alternative is auto, then to reach the activity out of the home will involve a journey that is part of a longer tour. MSD features at the origin or production point determine the modes that are available for the first journey and whether or not autos are available for the rest of the tour. Likewise, they determine whether or not child care requires a diversion from the typical path, and whether or not it is located near the home or workplace. The mix of land uses, another MSD feature, will also be important in setting up the sequence of trips within a tour or journey.
We see that MSD features play an important role in making long-term residential and job location decisions, that in turn have important and, in fact, critical impacts on short- term (daily) or impulse decisions. Another impact of this process is that life style changes that tend to take place slowly may mean that neighborhoods respond in a slow fashion to changes in MSD or urban design features. Therefore, introducing good transit service and sidewalks may not show impressive benefits or changes in lifestyle for as much as a decade, principally because the residents have organized their lifestyle around different patterns.
Another factor in the choice of residential location appears to be the neighborhood and its visual and structural components, how close the houses are to each other, how walkable it is, how well the houses and surrounding grounds are tended, and how attractive the walkways are. These may be appealing to individual family taste more than to economic decisions about transportation. In examining decisions and models, it would appear that travel time or commuting distances are more a constraint on the other choices than a real element in the residential location decision. Put another way, any location within a tolerable commuting distance is a candidate and is not significantly more attractive because it's closer rather than farther from the center of employment. Of course, it appears that urban dwellers would prefer to be closer to their jobs and not have to deal with traffic congestion.
2 These are models which, given aggregate data and system data, estimate the number of households (or the market share of households) by various attributes of the household including household size, income, workers per household, etc.