Quantifying the pedestrian access potential of suburban street network retrofits

Introduction

Automobiles, while incredibly functional as a mode of transportation, have significant negative externalities in many areas. For example, personal vehicles have large environmental consequences: they contribute around 17% of total greenhouse gas emissions in the United States and are responsible for large shares of airborne and watershed pollution (Climate Change Division, 2018; Zhang and Batterman, 2013; Hwang et al., 2016). In addition to the environment, automobiles negatively influence public safety, personal health, land use and congestion (National Center for Injury Prevention and Control, 2018a, 2018b; Zhang and Batterman, 2013; Hwang et al., 2016; Shoup, 2011; Lee and Sener, 2016). This mode of transportation is also not universally accessible. Large swaths of the human population cannot independently use automobiles (e.g., the elderly, those with certain physical impairments, and children) or would incur an unsustainable financial burden through their use (Marottoli et al., 2000; Mackett, 2002; Smart and Klein, 2015).

Unfortunately, the United States is a nation that is largely dependent on the automobile for personal transportation, as it has been for well over half a century, with over 85% of commuting trips made by car in 2019 (Burrows et al., 2021). While this might be a reflection of a preference for the many real benefits of automobility over other forms of transportation, the extent of this dependence is a consequence of many policy decisions over the 20th century, such as investments in highways. But, in particular, it is a result of the particularities of the American suburban form.

The suburbs are host to a majority of Americans today. More people live in them than urban or rural areas combined and, in the 50 largest metro areas, suburbs account for nearly 80% of the population (Terwilliger Center for Housing, 2016). Suburbs have increasingly diversified, in contrast to the racial homogeneity that prevailed for much of the 20th century. Increasingly, the spatial patterns of wealth that used to typify American cities during the same period have begun to invert, with poverty increasing more quickly in the suburbs than urban (or rural) areas (Parker et al., 2018).

Despite this dominance of existing residential suburban land in the modern built environment, the existing literature on reducing automobile dependence often focuses elsewhere. For example, there is significant literature on improving neighborhoods that predate automotive dependency, on building more walkable greenfield developments on the suburban fringe, and on retrofitting walkability and density into suburban greyfield or commercial lots (Aurbach, 2020; Nieuwenhuijsen, 2021; Southworth and Ben-Joseph, 1997; Tachieva, 2010; Williamson and Dunham-Jones, 2021). Still, the core problem of these suburban areas that inhibit pedestrians is well known. Distances to potential points of interests—such as grocery stores and schools—along safe paths are often too large for walking to be a viable mode, and thus pedestrian access is low. Attempts to solve this problem inform the design of newer greenfield and greyfield developments (Dill, 2004; Tal and Handy, 2012; Williamson and Dunham-Jones, 2021).

This lack of attention to retrofitting exiting residential suburban areas could have substantial negative consequences. To do so would commit the significant number of people who live in such areas to automotive dependence—an undesirable outcome given the aforementioned externalities. The changing demographics of the suburbs makes this increasingly fraught as well, given the mentioned economic mobility challenges associated with automotive dependence (Smart and Klein, 2015).

The excessive travel distances have two primary causes. First, the post-war suburban housing boom coincided with an increase in the use of Euclidean zoning (the separation of land uses by function, e.g., residential, commercial), and large single-use zones, in particular. This style of zoning placed jobs and services further away from the average residential use than occurred in the pre-war mixed land use paradigm. Second, the street networks of suburbs have largely been designed assuming the primacy of cars for everyday transportation and have ultimately disadvantaged non-automotive forms of mobility.

Early twentieth century planners in America, such as the Regional Planning Association of America (RPAA), sought to dispense with the street grid topology that had dominated in the prior century. The RPAA charged that grids unnecessarily dispersed the harms of the automobile into residential neighborhoods. Their proposed alternative was a hierarchical street network in which the harms of car traffic would be concentrated onto larger arterials. This concentration was effected through the use of curvilinear paths, loops, cul-de-sacs, and three-way intersections on side streets. During the Great Depression, the Federal Housing Administration (FHA) and the Institute for Transportation Engineers (ITE) incorporated these ideas into their respective design standards. These standards, due to the large influence of the FHA (mortgage insurance) and ITE (local street regulations), were very influential for the rest of the century. By 1970, nearly all greenfield developments were characterized by street networks of “loops and lollipops” (Southworth and Ben-Joseph, 1997).

There is evidence that this transition succeeded at concentrating automotive externalities (Aurbach, 2020; Southworth and Ben-Joseph, 1997). However, it has had the additional consequence of disadvantaging walking (and public transit) as a mode of transportation by increasing distances to potential nearby destinations, thus furthering automobile dependency. This problem, ironically, had been foreseen by the RPAA, who originally suggested that pedestrian shortcuts be provided to bisect longer blocks. However, these suggestions have since been seldom incorporated, with the exception of a few New Urbanist developments (Southworth and Ben-Joseph, 1997).

The limited efforts toward retrofitting residential developments in existing suburbs have generally proposed two similar solutions. The first is to reconnect cul-de-sacs with entirely new streets, accessible to all modes including cars. In addition to the benefits for pedestrians, these sorts of improvements are usually promoted for how they might improve first responder response times as well as water and waste removal services (Tachieva, 2010). However, this class of solutions is likely to be impractical in the near term as retrofits, as they incur real costs to existing suburban residents. The construction of new street segments would likely require extensive use of eminent domain and may require the demolition of existing buildings (Tachieva, 2010). Cul-de-sacs additionally have real safety benefits and increased land values. (Southworth and Ben-Joseph, 1997). The second class of proposed retrofits is mildly narrower in scope. It proposes the construction of comparatively narrow pedestrian and bicycle paths along lot lines to connect cul-de-sacs or divide longer blocks in half. Of course, this class of retrofits would likely face some amount of opposition, as it would also require the use of eminent domain. However, it seems more feasible and worth exploring in further depth.

A first step in this exploration is to quantify the potential impact of these pedestrian shortcut retrofits, both for individual shortcuts and a comprehensive retrofit plan including multiple shortcuts. Unfortunately, to the best of our knowledge, no such work existings to systematically quantify this potential. Such quantification is also necessary to measure the relative effectiveness of retrofitting pedestrian shortcuts versus modifying land uses away from the single-use status quo. Moreover, given the potential political difficulties in proposing shortcut retrofits, it is critical that we have a method to prioritize certain shortcuts and understand the relative importance of such prioritization.

Here we attempt to address these shortcomings of the literature by developing methods that quantify the disruption that the street network imposes on residential suburban access to nearby points of interest (POIs). These methods build on past work on graph detour/dilation and route directness from the planar graph and street network communities, respectively. We also consider a class of planar graph augmentation problems similar to those discussed in Brelsford et al. (2019) to model the addition of pedestrian shortcuts to the street network that follow existing property lot lines. We quantify the decreases in what we term the parcel mean accessible POI circuity (MAPC) that result from implementing a random set of interventions in a case study neighborhood in the Seattle suburb of Bellevue, Washington. Interestingly, we find that a relatively small number of random interventions consistently reduces the median parcel MAPC by a significant margin, suggesting that prioritizing shortcuts is relatively unimportant given that a small but sufficient quantity of shortcuts are implemented. Given this finding, we also develop a method for identifying the blocks that are most topologically disruptive toward potential pedestrian routes. This method could allow planners to more intelligently choose where to target initial retrofits given a limited budget or political capital.

Measuring impacts to access

There are a number of conflicting definitions of access in the transportation literature. In this paper, we use the definition of pedestrian access as the ability to reach goods, services, and activities on foot. This is often operationalized as the number of relevant destinations a pedestrian can reach within some fixed time or distance from a set location (Levinson and King, 2020). We use this operationalization as the base atop which we measure the disruption that the street network imposes on pedestrian access. In particular, for a given residential parcel, there is some set of POIs that could theoretically be accessed by pedestrians within a reasonable straight line walking distance while ignoring the street grid. We introduce a metric to quantify how much harder the street network makes it to access these POIs in practice. We term it the “mean accessible POI circuity (MAPC).”

There is a sizable body of literature that has quantified various attributes of street networks and their potential impacts on pedestrians. The most relevant measure for our purposes is variously termed route directness or circuity in the mobility literature and more generally as detour or dilation in the planar graph theory literature (Luo and Wulff-Nilsen, 2008; Stangl and Guinn, 2011; Boeing, 2019). When applied to pedestrian networks, it is known as pedestrian route directness (PRD). This measure is defined as the ratio between the route distance and the Euclidean distance between two nodes in the street network. A ratio of 1 means that the distance required to travel between two points along the street network is no greater than “as the crow flies”; that is, the street network does not inflate the travel distance. Larger ratios indicate the street network requires increasingly indirect pedestrian routes.

This has been applied to study a number of different issues. At the urban scale, some studies have used route directness to understand the extent to which individual suburban developments function as barriers to through movement across all modes (Stangl and Guinn, 2011). Boeing (2019) performed a comparison of circuity between the pedestrian and automotive networks across several cities to identify common network patterns that ultimately favor one mode. Several studies measure the average PRD at a neighborhood or development scale as a proxy for how amenable particular neighborhoods are to internal pedestrian circulation. These studies have primarily focused on a US suburban context, though a limited body of work has studied other global examples. A few studies focus on the impact of the street network on PRD between specific land use pairings, for example, between a residential parcel and a school. Randall and Baetz (2001) measure the potential impact of retrofitting the network with pedestrian shortcuts, albeit in a limited ad-hoc and hand-drawn fashion. Lahoorpoor et al. (2022) conducts a similar study that ranks expert drawn retrofits based on their ability to reduce circuity to two major POIs.

All these studies have used a variety of strategies to deal with the still significant challenges with computing a large number of shortest distances. The most common solution is to limit the size of the study area to the size of a single neighborhood or residential development (Stangl and Guinn, 2011; Paul Hess, 1997; Randall and Baetz, 2001; Alawadi et al., 2021). The alternative is to take a subsample of every potential origin/destination pair. This is done in one of three ways: random selection, selecting a small number of points along a perimeter some distance from the study area centroid, or by limiting to specific land use pairings (Dill, 2004; Boeing, 2019; Stangl and Guinn, 2011; Hess et al., 1999; Randall and Baetz, 2001). These techniques are often combined, though this has become less necessary with improvements to computational power over the last two decades.

The MAPC metric we define shares many similarities with prior work on circuity and PRD. We define it as

t h = 1 | A h | ∑ p ∈ A h ( d g ( h , p ) d ( h , p ) )

This formulation has a number of benefits. For one, it can be applied to parcels across a larger study area, as the accessible POI requirement has the secondary benefit of limiting the growth of the number of distance calculations required. Second, it narrowly focuses on the impact of the street network on utilitarian walking trips that could feasibly happen today in the absence of other interventions (e.g., land use changes). Finally, it is easy to interpret as the penalty that the street network imposes on accessing theoretically accessible POIs. For example, residents at a parcel where t _h = 1.5 would experience 50% longer trips, on average, to nearby POIs that would be accessible to them in a featureless plane.

Case study: Seattle suburb

We collected parcel, street network, and POI data for a subset of the Seattle suburb of Bellevue, Washington, USA. Parcel data was collected from King County and street network data from OpenStreetMap (OpenStreetMap Contributors, 2023). Shortest paths between parcels and POIs were calculated using the Pandana package (Foti et al., 2012). POI data was collected from the SafeGraph Places dataset, a comprehensive set of non-residential POIs. This dataset describes the physical location of, and provides a North American Industry Classification System (NAICS) code for, each POI.

The study area (shown in Figure 1) is slightly larger than 10km² in area, and roughly bounded by I-90 to the north, I-405 to the west, Coal Creek Parkway to the southwest, 150th Avenue to the East, and Highland Drive to the southeast. It primarily corresponds to the Factoria, Eastgate, and Somerset neighborhoods. The area is spatially dominated by single-family housing built in the latter half of the 20th century, though the area around Factoria Boulevard is dominated by a typical post-war shopping mall, various strip malls, and some multifamily residential uses. We selected this particular area as it reflects typical post-war suburban development patterns and has nearly complete sidewalk data in OpenStreetMap that we validated by hand. As a technical note, we include POIs up to 800m outside of the study area to ensure the accuracy of statistics measured at its boundary.OPEN IN VIEWER

We calculate the MAPC for every feasible parcel in the study area. Since we are primarily concerned with residential access to POIs, we limit the statistics we present here to parcels labeled as residential by the county. However, we do include all POIs in our calculations as this provides a general understanding of how much the street network hinders access to all potential sources of pedestrian travel demand. In some sense, every POI, even those that duplicate the service of others, matters as they increase consumer choice or labor market thickness for would be pedestrians.

We show the MAPC for every residential parcel in our study area in Figure 2(c). Values range from near 1 in a few cases, to over 4 at the most extreme. Most cases correspond to places that demonstrate curvilinear street patterns. We note how the lack of through streets along the perimeter of the Somerset neighborhood appears to correspond to dramatically larger MAPC for parcels within the neighborhood. This general pattern is replicated in a few other places in the study area as well. In general, we note that the accessibility costs in this neighborhood seem extreme—pedestrians are sometimes required to walk up to several times the straight line distance to nearby POIs. Even residents at the median parcel see a penalty imposed by the street network of around 90%.OPEN IN VIEWER

As described earlier, we take the set of all property lot lines as space upon which we could augment the pedestrian network with a pedestrian shortcut intervention. This is very similar to the work done by Brelsford et al. (2019) on reblocking informal settlements. More precisely, we define an intervention to be a sequence of property lot lines that connect to the existing street network at both ends of the sequence. Note that some interventions may overlap for some set of property lot lines, by virtue of this definition. In total, we identify 48,398 potential pedestrian shortcut interventions in our study area. Due to the setback requirements in the study area, the vast majority of these shortcuts should not require building demolition. However, additional work is needed to fully assess the technical feasibility of each shortcut and narrow the intervention space.

To assess the limits of the accessibility benefits that this class of interventions could provide in our case study, we augmented the network with every possible intervention in a “maximal intervention” scenario. While a comprehensive set of interventions is unrealistic, if not fully impossible, this scenario provides a baseline against which to compare less intrusive and more feasible intervention scenarios. The MAPC for each parcel in this scenario is presented in Figure 2(d). The differences with the original “no intervention” scenario are presented in Figure 2(b). Figure 2(a) plots the frequency of MAPC across parcels in the non-intervention and maximal intervention scenarios. As is evident in these two maps and the histogram, pedestrian shortcuts along property lot lines, in the limit of all possible interventions, significantly reduces the average disruption that the street network presents to residents in accessing nearby POIs. For a significant number of parcels, the reduction in MAPC is greater than 2. Overall, the median MAPC across all parcels fell from 1.90 to 1.36. The variance in MAPC across parcels also noticeably decreased.

As the case of maximal intervention is impossible and undesired, we also study the impact of a limited number of random, or untargeted, interventions. The untargeted nature of these interventions reflects a baseline for how much improvement could be expected assuming that political or land acquisition difficulties prohibit the implementation of an informed approach. We created an ensemble of 20 implementations of 1000 random interventions into the street network of the case study. All 1000 interventions are chosen from the 48,398 interventions identified earlier and added sequentially. The results are shown in Figure 3. We find that a large fraction of the maximum decrease in median MAPC is captured far before every potential intervention is included. In fact, we find that a relatively small 60 interventions captures over 20% of the maximal decrease in median MAPC.OPEN IN VIEWER

Smarter interventions

While we have shown that a limited number of interventions, even chosen at random, can have a significant impact on widely reducing MAPC, it would still be useful to give planners and other concerned parties a method to intelligently site interventions. Planners need to be able to give the public some justification as to how a particular intervention, or set of interventions, are chosen. In the abstract, this appears to be an optimization problem in which the objective is to minimize some function of the MAPC by creating new pedestrian shortcuts, subject to some budgetary, engineering, and political constraints.

There exists a class of problems in the planar graph theory literature that resemble this optimization problem. In particular, a number of studies detail algorithms that identify the best shortcut to minimize the dilation of the resulting graph—that is, the worst detour between any two nodes in the graph. However, to the best of our knowledge, even the most performant methods are quite computationally expensive, with complexity that is quadratic with the number of nodes in the graph (Luo and Wulff-Nilsen, 2008). Due to the aforementioned political challenges in siting interventions in the first place, a perfectly optimal solution is likely superfluous anyway. It would be preferable instead to develop a heuristic that roughly specifies where new pedestrian shortcuts would have a large impact.

Note that blocks, as the morphological dual to the street network, are one of the primary drivers of longer routes along the street network. They, along with highways and other uncrossable roads, function as obstructions to pedestrian through movement. In this paper, we focus on blocks, as the class of interventions we address—pedestrian shortcuts along property lot lines—are functionally modifications to blocks. In particular, pedestrian shortcuts along lot lines derive their impact from splitting blocks in two. Given these two attributes, one method to intelligently site new pedestrian shortcuts is to identify which blocks are most disruptive to pedestrian through movement.

We formalize this notion in a measure that we term a block’s “barrier score.” Barrier score is defined as the difference between maximum and minimum parcel MAPC on a given block. This is calculated over all parcels regardless of land use. Higher scores suggest that a block is more disruptive toward through movement and is thus a prime candidate for siting a pedestrian shortcut intervention. We can inexactly interpret this measure as the amount that an individual block is responsible for increasing the MAPC of impacted parcels. The intuition for why this measure is meaningful is as follows.

Imagine a rectangular block with a long side with length w and short side with length h. Imagine that there is a single POI directly across the road from the block in the middle of the long side. Assuming the existence of a crosswalk, a parcel on the near side of the block would have a MAPC of 1—the Euclidean and route distances to reach the single POI are identical. A parcel on the exact far side of the block would have a MAPC of w + h h (assuming 0 width streets). If the block were to be elongated such that the long side now has length 2w, the MAPC for the far side parcel would increase accordingly to 2 w + h h , reflecting that the block has become more of a barrier for parcels on the far side of the block. However, the near side parcel would still have a MAPC of 1. Thus, this increase in disruption gets encoded in a larger barrier score. Importantly, the increase in MAPC for the far side parcel reflects similar increases for parcels across the street from, or along a street that intersects nearby to, the far side parcel. And thus, the barrier score reflects the role a block plays in locally disrupting pedestrian through movement.

This contrasts with calculating the median MAPC over a block, for example. Such as a measure identifies blocks that are disadvantaged by the grid but does not provide information on whether intervening in the block with a shortcut would reduce the MAPC for parcels within the block or nearby blocks. For example, consider a block B encircled by a second block C, where the POIs are outside of C. The MAPC of each parcel within B, and thus the median MAPC of B will be high. However, some of the parcels on the outer edge of C could be adjacent to a POI, and thus have a lower MAPC. In such a scenario, C could have a lower median MAPC than B, even though C is the block that requires an intervention.

We apply this measure to each block in our case study area as depicted in Figure 4. We observe that a number of long, snake-like blocks have large barrier scores. A comparison of these scores with the Polsby-Popper score for polygon roundness demonstrates a strong correlation between the two measures: less round blocks are more likely to be barriers to through movement. A number of the worst blocks by barrier score correspond to noticeable breaks in the spatial distribution of MAPC in Figure 2(c). This suggests the role that these blocks have in causing those breaks in the spatial distribution of access. We note that a few blocks with large barrier scores identified here already have existing pedestrian trails that bisect them, and thus are not truly unified blocks in some sense. However, this emphasizes that they may require additional interventions.OPEN IN VIEWER

Conclusions and future work

In this paper, we performed a novel analysis of the potential for pedestrian shortcut retrofits to improve suburban pedestrian access to nearby POIs in a case study of a suburban area in the Seattle metropolitan area. We developed a new measure, that we termed the mean accessible POI circuity (MAPC), that extends previous work on circuity and pedestrian route directness to quantify the impact that the street network has in obstructing residential pedestrian access to POIs. This measure, applied to our case study, revealed a significant amount of obstruction to a significant number of parcels. To ameliorate these issues, we studied whether a specific class of interventions—pedestrian shortcuts along property lot boundaries—could meaningfully reduce parcel MAPC across the study area. We find that, in the limit of all possible interventions, this class of interventions can significantly reduce access taxes across nearly the entire study area, with the exception of highway induced obstructions. We additionally found that a relatively small number of random interventions could achieve a large percentage of the same reductions. Finally, we developed a measure that could be used as a heuristic for planners to more intelligently site pedestrian shortcuts given budgetary and political constraints.

The potential benefits of suburban street network retrofits presented in this research highlights the need for deeper investigation into a number of related problems. First, an argument can be made that certain POIs matter more than others for the purposes of reducing auto-dependency. This argument has three components: (1) certain categories of POI may matter more than others, (2) POI importance is some function of distance from the residential parcel, and (3) POI importance is a function of the number of similar POIs nearby. Future work could study how sensitive the results found here are to these weighting decisions. Second, we did not assess the technical, financial, or political feasibility of any of the shortcuts. Additional work is needed to capture limitations due to building placement, elevation changes, land acquisition costs, and political feasibility. Third, highways and other arterial streets without sufficient crosswalks also function as significant barriers; property line interventions do not address them. The barrier score developed in this paper for blocks could easily be extended to identify locations for overpasses or crosswalks. Finally, the barrier score developed here has room for improvement. In particular, it is limited in its ability to handle scenarios in which there are POIs on opposite sides of a block. While this did not have an impact in the case study used in this paper, more work is needed to address this limitation.

This work has implications that extend further than the direct pedestrian access impacts. In particular, since transit trips often require walking to or from a stop, pedestrian shortcut retrofits potentially have implications for access at the urban scale, not just the neighborhood scale. Suburban transit planning often faces difficult tradeoffs to address the “last mile” of trips, the journey the rider takes to and from the stop at either end of the trip (Braun et al., 2022). For example, in many suburbs, the arterial streets that bound residential developments intentionally have few connections to the interior of the developments in order to reduce automotive ingress while decreasing the number of cross streets on the arterial. However, this design pattern ends up increasing residential distances to bus stops located along arterials. While buses could instead be routed through the residential development to increase access, this would slow them down significantly, as local suburban streets are designed to be circuitous (Aurbach, 2020; Southworth and Ben-Joseph, 1997). The pedestrian shortcuts here help circumvent this trade off by reducing pedestrian distances to the arterials. The exact contours of this benefit merit further study.

Furthermore, a large benefit of reducing automobile dependency are the potential implications on greenhouse gas (GHG) emissions. Pedestrians and transit (and bicyclists, who may also benefit from shortcuts) emit far fewer GHGs per trip. Future research could extend this work to apply to other non-automobile based mobility and ascertain how much these sorts of retrofits could reduce GHG emissions. In general, the potential benefits of suburban street network retrofits presented in this paper reveal several extensive avenues for future work. These retrofits, could in turn meaningfully pave the way for increased accessibility and a low-carbon transportation future for suburban America.