Accessibility: measurement and application in transportation planning

The primary role of a transportation system is to provide people and businesses with access to other people and businesses so that they can physically engage in spatially and temporally distributed activities of all kinds (social, economic, etc.), and so that they can physically exchange information, goods and services. Given this fundamental relationship between transportation and accessibility, it is surprising that accessibility remains a rather illusive concept in transportation planning and modelling, with a number of issues still existing concerning its definition, measurement and, most importantly, usage in practical applications.

Starting from first principles, there is arguably general, axiomatic agreement that accessibility:

• Varies from point to point in space.

• Is activity (trip purpose) specific.

• Combines the concept of travel impedance (i.e., the ease/difficulty to reach or interact with different points in space) with that of attractiveness and/or magnitude of opportunities (i.e., the desirability of/opportunities for interaction at a given point).

• Is a measure of the potential to interact. Mobility, on the other hand, is the realisation of this potential in terms of actual travel through the transportation system.

• Involves the integration or summation over the space of opportunities, weighted by the ease of interaction. If there are many attractive stores near my home, my accessibility for shopping clearly is higher than if there only a limited number of stores available that are of poor quality and/or located very far away.

• Given the needs/wants-based nature of travel, the concept of ‘interaction potential’ (i.e., accessibility) is clearly tied closely to that of travel demand, as well as the location choices of households and firms.

Translating these intuitive but loose properties into concrete measures of accessibility requires a number of further assumptions. The first involves defining how to measure how ‘near’ or ‘far’ one point is from another and what the ‘cost’ of travel from one point to another is. Many accessibility measures simply use distance (either straight-line or network path-based, with the latter being generally preferable) as the metric. For walk accessibilities, distance is a suitable measure since it readily maps into travel time and is also a direct measure of the level of effort involved in accessing a given point. Otherwise, however, travel time is generally a much better measure of perceived impedance than distance. Travel time for any trip, however, depends on both the mode of travel used and the time of day when the trip is made. Thus, travel time-based measures of access (and hence accessibility) must either be mode-based or else ‘integrate over’ the potential to travel by the various modes available for a given trip in an appropriate manner. Further, to the extent that travel times by both auto and transit vary by time of day, this means that accessibility also varies within and across days.

Travel time is also a transportation policy-sensitive variable since it will change in response to changes in both the demand and supply sides of the system; whereas distance-based measures generally will only change in response to changes in the attraction variable levels in the measure.

Similarly, many factors might influence the attractiveness of a given location for a given activity. Conventional factors that are often used include location ‘size’ measures (floorspace, number of employees, population, etc.), but a wider variety of variables is clearly conceivable. A major practical challenge, however, in virtually all operational accessibility measures is the use of overly simplistic and aggregate ‘size’ measures as very crude measures of the actual attractiveness of a given location for a given activity. This challenge in measuring attractiveness is compounded in any practical application by the inescapable significant heterogeneity that exists in human activities and the associated attributes of activity locations. ‘Shopping’ is often an activity episode/trip purpose category for which we may wish to compute an accessibility measure. But shopping is an extremely heterogeneous activity, encompassing everything from picking up a litre of milk at the local ‘corner store’ to buying major durables such as cars, furniture, etc. How does one represent location attractiveness appropriately in the face of such variation in activities? Additional categorisation of activity/trip types helps, of course, but there are inevitable practical limits to achievable levels of detail.

Further, perception and evaluation (weighting) of impedance and attraction factors vary from agent to agent, depending on the agent’s cognitive abilities, tastes/preferences, personal experiences and personal (and household) constraints. The ability of people to exploit a given physical context of access and attraction will also vary depending on their financial, cognitive and physical capabilities. As obvious examples, persons without cars and persons with physical and mental disabilities all generally have quite different accessibilities (potential for interactions) than persons who do not face these same challenges. Therefore, accessibility ultimately varies among persons as a function of both their individual preferences and capabilities.

Thus, the ‘simple’ notion of accessibility becomes surprisingly difficult to operationalise. In particular, if we believe that travel time is the appropriate measure of access, this leads to a veritable ‘hyperspace’ of accessibilities varying from person to person by mode, purpose and time of day, in addition to spatially. Note that not only does this pose significant analytic challenges in terms of computing myriad accessibilities, it creates serious challenges to accessibility-based policy-making, as is discussed further below.

These conceptual challenges, of course, have not stopped analysts from asserting and constructing a wide variety of accessibility measures for a wide variety of applications. Four of the most commonly used measures in the literature are (Handy & Niemeier, 1997; Kwan, 1998)

1. Distance (or perhaps time by mode M) to the nearest subway stop, freeway interchange, school, hospital, etc.

2. Cumulative opportunities within an access distance or time threshold (isochrone method).

3. Gravity/entropy model denominators (‘Hanson’s measure’; Hansen, 1959).

4. Expected maximum random utility-based measures (e.g., logit model ‘logsums’; Ben-Akiva and Leman, 1985).

Detailed discussion of the relative merits or appropriate uses for these four approaches is not feasible in this editorial. The only point that will be made is that the first three are simply asserted measures that have no solid basis in theory. Their usage is based on pragmatic grounds that they: (1) are generally consistent with the accessibility axioms; (2) are easy to compute and (3) (may) generate statistically significant and plausible parameter values in location choice models, etc. Logit logsum type measures have some theoretical foundation in random utility theory, given that the logsum expected maximum utility measure can be interpreted as a measure of consumer surplus, and that ‘it seems reasonable’, therefore, to assert that accessibility is defined as this consumer surplus term. Use of such logsum terms to ‘roll up’ into, for example, logit residential location choice models is also a practically convenient approach in many models.

This lack of a compelling, robust theoretical foundation for measuring accessibility results in many conceptual and practical problems in using accessibility in transportation planning and policy analysis. Taking worker access to employment opportunities as a common accessibility measure of policy interest, examples of issues include the following.

First, there is no objective, normative standard for what constitutes ‘good’ or ‘acceptable’ accessibility. My employment accessibility may be 20% below the average accessibility for people in my occupation group in my city, but the value of this average accessibility (and hence the value of my 20% deficit) is difficult to establish. Using an isochrone measure, for example, what is the value of access to one more job within a 30-minute threshold – either to me individually or to society as a whole? Further, this average accessibility is the emergent outcome of the current transportation system and the current distribution of workers and jobs, will change over time and has no intrinsic meaning except as a relative, within-group benchmark at one point in time. Are accessibilities in a given urban region getting better or worse over time? Does one urban region have better average accessibility than another? How should governments set standards for ‘minimum acceptable’ accessibility levels? How much money should be allocated to improving accessibility (for whom, where, in what way)?

Second, the subjective (person-based) nature of most accessibility measures makes comparison of accessibility levels across groups difficult. If I have a lower accessibility score based on my tastes and preferences (perhaps defined on the basis of my occupation group and income) than you do based on yours, can a policy-maker actually assess my ‘need’ relative to yours for more accessibility?

Third, as noted above, all accessibility measures are ad hoc in derivation to various degrees. This results in different measures and specifications being used in different studies and urban regions, making ‘learning’ over time and across applications concerning what are ‘best’ specifications difficult. Reasonable correlations between observed location and travel choices and an accessibility measure may be achieved, but these usually do not provide strong ‘hypothesis tests’ of the assumed specifications. Isochrone-based measures are particularly arbitrary and behaviourally suspect (Xi, et al., 2018), but gravity/logit model-based measures have their own issues, as discussed below.

Fourth, all conventional accessibility measures are static/cross-sectional in nature, based on a snapshot in time of population and employment distributions and associated transportation service levels at that point in time. The estimation of model parameters from cross-sectional data inherently assumes a system at equilibrium (which the entropy formulation underlying both gravity-based and logit-based measures makes clear). But if residential and labour markets are not ever in equilibrium (which very arguably is the case) how can we impute the role that accessibility plays in housing and employment location decisions, among others, from cross-sectional models? A specific example of this concern is residential location choice models, which typically are estimated from cross-sectional data and then applied in a quasi-dynamic integrated transport – land use model system. ‘Integration’ usually occurs through a logsum accessibility term feedback from the travel demand model to the land use model. But does this logsum, derived from modelling an arbitrary ‘typical’ day actually reflect the way in which households assess their accessibility and the trade-offs between accessibility and the many other factors affecting their long-term location choices?

Fifth, the relationship between accessibility and land value is not as well established as is needed for rational transportation infrastructure investment and urban planning. Spatial economic theory tells us that our location choices should be the outcome of trade-offs between different levels of accessibility and other amenities/considerations in residential and firm location choices and building stock development decisions, in addition to activity/travel location choices. The value of land and buildings (their ‘bid rents’) should reflect these trade-offs and the competition among agents to locate at points of greater or lesser transportation advantage (accessibility) (Alonso, 1964). Despite this theory, and despite decades of analysis and modelling activity, we generally lack robust relationships between accessibility and land value that can be confidently used in policy analysis. This is a particularly important gap for transit infrastructure investment decisions, whose benefit–cost evaluations often critically hinge on the land development and land value increases expected from such major investments.

Further, the role that accessibility plays in the associated issue of agglomeration economies and processes is, again, assumed in economic theory, but not as explicitly elaborated as one needs to truly understand the feedback processes at work. The emerging ‘science of cities’, and, in particular, explanation of the empirically observed ‘scaling’ of city inputs and outputs is largely based on assumptions of increasing agglomeration effects (tied to social and physical network interactions) with growth in city size (West, 2017). As with so many aspects of the accessibility discussion, this assertion is plausible, but not yet well demonstrated.

Finally, the complexity of the concept and its measures makes accessibility very difficult for the public, politicians and even many planners to understand and use. The simpler ‘distance to the closest transit stop’ and ‘number of jobs within a 30-minute drive are intuitively understandable, but often theoretically dubious’. Model-based measures improve upon many of the failings of the simpler measures, but often generate a massive amount of data that are difficult to digest, while the methods themselves are generally incomprehensible to the non-modeller.

If accessibility is to ‘take its rightful place’ as a central concept in transportation planning, we need to, first, firm up and standardise our theoretical concepts and operational methods, and, second, find much better ways to communicate the usefulness of these concepts and measures in clear, compelling and credible ways to the public and decision-makers. The latter can be addressed in part through well designed visualisation tools (including interactive displays which allow people to explore the impacts on their neighbourhood accessibilities of various transportation policies) and thoughtful, non-technical, context-sensitive ‘story-telling’ that explains the way in which transportation policies might (or might not, in the case of poorly conceived policies) improve a neighbourhood’s accessibility (and, hence, opportunities for improved quality of life).

The much more challenging problem, however, is to establish useful, robust standards for ‘acceptable’ levels of accessibility and methods for valuing the benefits of accessibility. In transportation planning, we are used to computing travel time changes due to infrastructure investment and other policies and to attaching an economic value to these changes via value of time calculations. This represents a mobility benefit due to increased efficiency in movement that ‘saves time’ that can then be used productively in other ways. We do not currently have a comparable, standard way of measuring the accessibility benefit of the opportunity or potential to interact at various levels of access. Defining accessibility in terms of consumer surplus is conceptually very attractive, since it associates accessibility with a very well established measure of social welfare/benefit. It also facilitates monetisation of accessibility, which can be useful for benefit–cost evaluations. Significant improvement in our representations of the actual attractiveness of alternative locations and in our specifications of location choice sets, and extensions to include a wider range of activities and services (education, health care, etc.) are all required if this approach is to be broadly useful in transportation planning and decision-making.