Closing the Gap: A Data-Driven Priority Index for Safe Routes to School in Equity Priority Communities

31 March 2026

CP 255 Urban Informatics and Visualization
Final Project Report

[MTC Project #1: Proximity to Neighborhood Amenities]

MAP’N Team 11 (MTC Team)

Anwar Baroudi | Phoebe Chiu | Noelani Fixler | Michael Huang

Department of City and Regional Planning
University of California, Berkeley
230 Bauer Wurster Hall
Berkeley, CA 94720

Jump to Section:

Context and Motivation For Research | Methodology | Limitations | Initial Results and Visualizations

1. Context and Motivation For Research

The Metropolitan Transportation Commission (MTC), the regional planning agency for the San Francisco Bay Area, is leading efforts to prioritize safety for walking and cycling trips to schools through its Safe Routes to School (SRTS) programs. Initiatives comprise infrastructure- and non-infrastructure-based strategies funded by MTC’s One Bay Area Grant (OBAG). Projects have historically included public education or awareness campaigns, pedestrian or bicyclist training for K-12 students, and infrastructure audits near schools¹. This traditional focus on pedestrians and cyclists aligns with federal- and state-level SRTS programming but often overlooks students who live beyond a walkable distance, typically defined as 0.5 to 2 miles^2,3. Despite SRTS global success of improving active transportation safety for school commutes, it often fails to address strategies to close gaps in the transportation network outside of the active transportation context. Gaps in school access are critical areas to investigate within the context of SRTS, as students who use transit to get to school may encounter unique access needs, especially if transit stops are not near school grounds.

SRTS was founded in Denmark in the 1970s with a simple goal of reducing traffic injuries experienced by children on their way to school. This initiative resulted in a 40 percent reduction in child traffic-related injuries, and has since been replicated in countries around the world⁴. Since its introduction to the United States in 1997, the program has expanded its scope to improve safety, encourage physical activity, and foster enjoyable commutes⁵. Thus, including the active transportation considerations/needs of transit riders within SRTS would further promote the program’s goal of enhancing physical health. Research indicates that first- and last-mile walking distances from transit trips provide moderate-to-vigorous physical activity (MVPA) levels similar to dedicated walking trips, indicating how improving conditions that enable students to access schools via transit is a viable strategy for increasing youth activity levels⁶.

In addition to the safety outcomes facilitated by traditional SRTS programs and policies, the inclusion of transit access within the SRTS framework to support children’s trips to school is essential for promoting student academic success and socioeconomic equity. Students relying on public transit often face “time poverty,” where every additional 10 minutes of commute time correlates with measurable declines in core subject scores and increased psychological fatigue⁷. In urban settings the typical travel time by public transit is significantly greater than by car, resulting in a major imbalance of transportation reliability as access to a vehicle often provides more predictable and reliable travel outcomes.

Furthermore, access to higher quality schools is increased with household access to a vehicle⁸, while the logistical burden of long bus commutes for transit dependent students can often result in students transferring out of high-quality schools due to the lack of reliable transit options⁹. These patterns in transportation disparities are correlated by race: Black students travel further to school than their white or Hispanic peers, often navigating less efficient transit networks to reach desired campuses¹⁰. Students in Equity Priority Communities (EPCs), defined as Census tracts that have a significant concentration of underserved populations¹¹, often reside in what are regarded as “essential service deserts” where school access is highly correlated with limited healthcare and food resources¹². Equity Priority Communities (EPCs) are a designation used by the Metropolitan Transportation Commission (MTC) to identify census tracts with a significant concentration of underserved populations. In the San Francisco Bay Area, a tract is designated as an EPC if it meets specific thresholds for at least two of the following eight indicators: people of color, low-income households (below 200% of the federal poverty level), limited English proficiency, zero-vehicle households, seniors aged 75 and over, people with disabilities, single-parent families, and severely rent-burdened households. For the purposes of this report, the “Zero-Vehicle Household” metrics and racial distribution are the primary drivers for analyzing school-travel disparities.

Conversely, studies show that students with reliable transportation, such as school bus service, are significantly less likely to be chronically absent¹³, and bridging gaps in financial access to transit by providing free transit passes to high schoolers has been shown to reduce excused absences by as much as 27.5%¹⁴. We suggest that by focusing on how students who use transit finish the last portion of their trip to get to school, we can facilitate access and promote resilience in the transportation system by creating redundancies in route choice to schools: when one mode of transportation to school, such as walking or biking, or even access to a vehicle, is unavailable, access via pedestrian paths from transit networks can allow sufficient connectivity to educational institutions.

Research on first and last mile trips suggest that walkability improvements near transit are multiplicative; high-quality pedestrian infrastructure within a quarter to half mile catchment area of a bus stop act as a force multiplier for ridership and safety¹⁵. In EPCs, these network effects may be more pronounced because these areas often have more hostile pedestrian infrastructure (e.g., highway overpasses or wide arterials) that act as absolute, impassable barriers. By removing these weak links in these pedestrian networks, practitioners can effectively double or triple the functional Pedestrian Catchment Area (PCA) for students within a 0.5-mile radius, significantly reducing the barriers for student transit access¹⁶.

Despite literature that suggests a pronounced impact of street network connectivity on transit ridership, there remains a modeling gap in the context of student commute patterns: while workforce commute patterns are modeled, there are few formal, large-scale modeling efforts to evaluate how transit-dependent students access education. To begin bridging this modeling gap in access to schools from transit networks, we are investigating how this gap is distributed amongst three EPCs within neighborhoods in the MTC region. Tile2Net, an open-source platform using a semantic segmentation model to create digital representations of roads, sidewalks, crosswalks, and other pedestrian infrastructure, can provide detailed data on the material conditions affecting pedestrian access to schools from transit stops. Because Tile2Net’s first and only supported region in California is Alameda County, the three EPCs our project will evaluate will be in Alameda County to enable us to leverage Tile2Net. Analyzing at the EPC Census tract level will also allow us to test our model on a small, yet regionally impactful scale, in alignment with MTC’s support for SRTS programs in the Bay Area.

To improve access to schools for pedestrians, we will use detailed pedestrian path networks obtained from Tile2Net to take a disaggregated approach; we can map out existing pedestrian routes from bus stops to schools, and note specific gaps. By highlighting those gaps that have the greatest detriment to school accessibility, we can help prioritize projects that most improve said accessibility. We can also compare the accessibility by pedestrians to schools to the accessibility for drivers to highlight the regions with the greatest disparities.

Below, we develop and visualize our methodology to show a precise relationship between pedestrian routes that serve schools and bus stop locations, to quantify the transportation barriers that could lead to chronic absenteeism and educational inequity.

A) Stakeholder Considerations and Anticipated Impact

The motivation for this study evolved from a broad examination of pedestrian network connectivity and access to amenities to a targeted analysis of connecting pedestrians to schools from the transit network in the context of SRTS, specifically focusing on MTC’s EPCs. Earlier project drafts included a wide array of amenities, such as grocery stores and healthcare facilities, but feedback from MTC and academic advisors to adjust the project scope in favor of a more narrowly defined research question encouraged us to turn toward SRTS. We selected schools as our focus since trips to school are high-frequency, non-discretionary daily trips, made by vulnerable populations who often can’t and don’t drive. Moreover, SRTS is an explicit priority within MTC’s key planning interests. This selection allows us the flexibility to test our model in the context of differing street networks and urban typologies. In the future, we could select different areas of focus for this analysis as relevant for further research.

B) Unit of Analysis, Measures and Metrics

To most accurately model the student pathways and origin-destination (OD) pairs, we have explicitly established high-frequency transit stops as the primary origin point for the last mile of school commutes. We are defining high-frequency transit stops as stops with service every 20-minutes or less (three trips per hour) during the morning and afternoon peak service hours, in alignment with the AB 2553 redefinition¹⁷. Our model utilizes this standard to identify gaps where physical infrastructure barriers force transit-dependent students into commutes that exceed this legally-defined threshold for accessibility. Through our analysis, we aim to explore the differences between options for a driver, who may have multiple redundant paths to a campus, as compared to access for a transit-and-sidewalk-reliant pedestrian. To normalize our analysis across both travel modes, we will tether the origin point of vehicle and pedestrian trips from a transit node, in this case, bus stops. Though we recognize those who travel by vehicle will often have greater flexibility in route choice as they are not limited by a fixed-route system, our project aims to specifically examine disparities in access that remain for pedestrians, considering first-last mile, even after completing the transit portion of their trips.

If a path in the pedestrian network is missing compared to the vehicle road network, we can identify it as a spot where pedestrian access is disproportionately lower than auto access. Drawing from studies on pedestrian connectivity referenced above, we aim to identify locations where multiplier effects can be applied, by making relatively small yet impactful connections in pedestrian networks.

2. Methodology

A) Data Description

The project’s data includes sociodemographic information about schools and Census tracts within MTC EPCs, as well as spatial information about the transportation network surrounding these locations, including the street network, pedestrian network, and bus stops.

The MTC/ABAG dataset on Plan Bay Area (PBA) 2050 Equity Priority Communities includes information for 339 EPC Census tracts as a GeoJSON file, based on the 2014-2018 5-Year American Community Survey (ACS). Relevant information for this project includes each EPC’s Census GEOID and geometry for spatial visualizations and triangulation with other resources providing more recent data for research analysis. The 2020-2024 5-Year ACS shapefile from the U.S. Census Bureau provides updated estimates of the share of households without a vehicle within each Census tract. The California Department of Education has separate data sets for all public and private schools throughout California for the 2023-2024 academic year available via GeoJSON. Both data sets include information about school location, as well as the total number of enrolled students for each school. Only the data set on public schools also includes student counts by several demographic groups, including race/ethnicity, socioeconomically disadvantaged, and students eligible for free or reduced meals. We selected the 2020-2024 ACS estimates for households without a vehicle and demographic information for the 2023-2024 academic year for schools within Alameda County to ensure consistency within the data collection period for both data sets.

Given potential changes or complete reconfiguration of Census tract boundaries, we conducted a brief spatial analysis for our study areas in Alameda County by overlaying the 2018 5-Year ACS PBA 2050 EPC Census tracts over the 2024 5-Year ACS Census tracts with matching GEOIDs to the PBA 2050 EPCs. We found that only 4 out of the 101 EPCs in Alameda County no longer existed in the 2024 5-Year ACS, and only 3.3% of the combined area of all Alameda County EPCs is missing in the estimates for the share of households without a vehicle. Examining the overlaid map, the Census tract boundaries of the EPCs that remained in the 2024 5-Year appeared to be relatively similar to the original 2018 5-Year ACS boundaries.

Information about the transportation network in EPCs in Alameda County originates from one public-facing data set from AC Transit and two open-source data processing platforms, OSMnx and Tile2Net. The AC Transit General Transit Speed Specification (GTFS) CSV file provides information about transit service throughout the AC Transit service area in Alameda and Contra Costa counties. Specifically, GTFS provides the latitude and longitude of bus stop locations, bus stop names, and a link to conduct spatial joins with bus stop locations with other attribute data for each bus stop (e.g., information about the bus lines served by the bus stop). Our analysis will use August 2025 GTFS data, following the AC Transit realignment. Though the data collection period for students and EPC Census tract demographics lags behind the AC Transit GTFS data, we reason that the inclusion of the most recent GTFS data in our final deliverable will provide a more accurate representation of how students travel by foot or on wheels from transit stops to schools, given the most recent data we have on transit stops, schools, and neighborhoods from the data available. Map 1 below was generated to visualize the spatial distribution of schools and bus stops within the study areas.

OSMnx and Tile2Net are two Python packages that generate data about street and pedestrian infrastructure. OSMnx produces network graphs that include spatial data for nodes and edges, including location and distance. Tile2Net provides satellite imagery data for Alameda County to generate polygons of pedestrian infrastructure, including roads, sidewalks, crosswalks, and other pedestrian infrastructure (e.g., curb ramps). The computer vision and semantic segmentation model will parse through satellite images of Alameda County at zoom level 19, or 0.3 meters per pixel, to estimate data about the active transportation network. Our research team is still examining the metadata of these platforms to determine the data collection period of these resources.

</div>

Map 1: Schools and Bus Stops in the Oakland Area

B) Data Cleaning and Model Limitations

Our data cleaning process included the following steps for the following categories:

California Open Data Portal: The raw school data included both public and private school data so we first standardized this by manually adding a “school type” categorical variable into both public and private datasets prior to merging. A subset of this data was created next by filtering the raw data specifically for Alameda County to ensure local relevance for our analysis. Because the school records did not contain EPC identifiers, a spatial join was required. We conducted sanity checks for whether to consider all schools or only public schools since the California Department of Education was missing demographic data for private schools. By producing figures to determine the relevance of private schools to EPC Census tracts, we chose to focus specifically on public schools, which may require future adjustments to our accessibility analyses and chosen EPC Census tracts.
EPC data: Regarding the MTC EPC data, we used a spatial “within” predicate to join school coordinates with the MTC PBA 2050 EPC census tract polygons. We may have to coordinate with MTC about whether their priorities are to run the most current analysis versus the most accurate analysis to the 2014-2018 ACS Census tract boundaries of the Plan Bay Area 2050 EPCs.
ACS data: The raw ACS 2020-2024 data provided raw counts rather than percentages. To normalize this, we calculated the proportion of zero vehicle households per tract by dividing households with access to zero vehicles by the total number of households.
AC Transit data (bus stops): open-source data was downloaded from the AC Transit Open Data Portal, and clipped to only include the selected EPCs. Remaining data cleaning for bus stops includes joining the stops with bus lines and schedules to filter out buses that don’t run frequent service during peak hours.
Tile2Net (pedestrian pathways): we set the pathway geometries into Tile2Net by obtaining the rectangular envelope of selected EPCs and running Tile2Net, we then clipped the output to our EPCs. Remaining data cleaning for pedestrian pathways includes distinguishing between greenpaths, sidewalks, and crosswalks, and between gaps that are model noise versus genuine service gaps. Manual cleaning may be necessary to distinguish these data.
Oakland Speed Limits data: As per our understanding of the dataset, any street for which an OMC informed speed limit wasn’t included, we substituted in the CA state default of 25 for residential/commercial streets.

We include a map in Section 4, Initial results, to provide a visual representation of EPC boundaries, socioeconomic status, and school location with regard to EPCs [above categories 1, 2, and 3]. For this map, we gathered the following data:

Car-ownership metrics were derived from the 2018 American Community Survey (ACS) 5-Year Estimates (Table B25044). We calculated the ‘% Car 2. Free’ metric by aggregating households with zero available vehicles across both owner and renter tenures, providing a proxy for transit dependency.
Building geometry was downloaded from Dryad (https://doi.org/10.7280/D16387) and filtered for Alameda County using QGIS due to its large file size.
The EPC layer was ingested via MTC API.
School locations were mapped using datasets from the California Open Data Portal.

C) The Shortest-Path Route Analysis

To understand how students actually navigate their neighborhoods, our model moves away from simple straight-line distances (“as the crow flies”) and simulates real-world walking behavior. We accomplish this by using otherwise-rare sidewalk data instead of the more commonly used street networks.

Building the Digital Network

Our spatial model is like a digital map of the neighborhood. Using Python data libraries, we convert physical geography into a Graph, with the two core components:

Nodes (The Intersections): Every street corner, crosswalk entrance, and bus stop is mapped as a specific geographic coordinate point, as well as points where a sidwalk or crosswalk changes direction.
Edges (The Pathways): The actual segments of sidewalks or crosswalks that connect these points are mapped as pathways.

By cleaning and linking these pieces together, we create a mathematically precise network grid where every path knows its exact length and location. Cleaning the pedestrian networks after pulling the data from Tile2Net was done through QGIS in order to fill or interpret any gaps that was missed by the program.

Weighting the Graph

With edges created, we applied weight to enable better routing. After all, traversing 30 feet of sidewalk and 30 feet of crosswalk are not the same and should be treated as such.

This weighting considered three key factors: Distance: the length of the edge Facility Type: sidewalk, crosswalk, or (new) midblock crossing. Which facility type determined an exponential modifier applied to distance. For sidewalks that modifier is 1, so no modifier, and we applied a smaller modifier for new midblock crossings than for crosswalks (and existing midblock crossings) with the assumption that new midblocks made with safety in mind would receive the necessary traffic calming treatments needed to make them safer than a crosswalk is capable of being. Speed limit: Only applies to crosswalks and midblocks, increased the weight modifier based on the speed limit of the street being crossed. No penalty applied for a speed limit of 25, and a linearly increasing (or decreasing) penalty added for each mile above (or below) that baseline.

Simulating Student Walking Logic

With the weighted graph established, we use a routing algorithm to calculate the most efficient walking paths from local high-frequency bus stops to school gates. Specifically we apply Dijkstra’s Algorithm, the best choice for a weighted graph.

We run this routing simulation under two distinct scenarios:

The Baseline Condition: We calculate the best path a student can take using only the neighborhood’s current, existing sidewalk network. If a critical sidewalk is missing, the model recognizes it as a dead end and forces the simulated student to take a longer detour.
The Remediation Counterfactual: We rerun the simulation after digitally fixing the missing connections. This allows us to see how a new sidewalk segment might unlock a much more direct, safer path.

By comparing the distance of the winding current path against the streamlined remediated path, we isolate exactly how much unnecessary walking distance (circuity) is forced upon students by broken infrastructure. Gaps that cause the largest detours or block access to vital transit nodes are flagged as our highest priority for municipal repair.

3. Limitations

The following section outlines the boundaries of our analytical framework. While our methodology provides a data-driven approach to prioritizing sidewalk improvements, it is important to recognize that a spatial model is an approximation of reality. Our analysis primarily focuses on the physical proximity and infrastructure potential of school-to-transit corridors. It does not capture real-time behavioral data, individual student route choices, or the micro-scale quality of sidewalk pavement (such as cracks or narrowness) beyond what is available in the public record.

Our findings are grounded in two primary assumptions: first, that enrollment is a valid proxy for the maximum potential impact of infrastructure improvements; and second, that students prioritize the shortest walking path between transit and school. To ensure the integrity of our results, we have validated our outputs against the MTC EPC definitions and performed internal pipeline audits to ensure that spatial joins correctly handled the coordinate reference system shifts across various California state and local datasets. Despite these validations, our model is subject to several forms of Dark Data. These limitations are detailed below.

Missing Data Analysis

While our routing model integrates standard transit, educational, and demographic datasets, any spatial model is an approximation of reality. To evaluate what our methodology cannot capture, where it makes assumptions, and how those factors affect our prioritization, we apply David J. Hand’s framework from Dark Data: Why What You Don’t Know Matters. By identifying where data is missing, hidden, or structurally obscured, we transition from a simple spatial join to a validated, high-utility network utility model.

Category 1: Highly Applicable Dark Data Gaps (Directly Affecting Results)

A) Selection Bias and Target Population Gaps

Our school data, drawn from California Open Data Portal (2023-2024), may omit non-traditional learning institutions such as “learning pods”, unregistered micro-schools post-pandemic, and transitional kindergarten and pre-K programs. This omission can be classified as both selection bias and a mismatch of the target population and the sample population. In other words, the data we are relying on only identifies schools officially registered by the State, leaving out non-traditional and unregistered schools. This bias can result in our conclusions also leaving students omitted by the data, resulting in an incomplete analysis of the impacted target population.

Since there are no clear cut avenues to mitigate the absence of this data, we will emphasize in our findings, discussion, and implications that the results of our study can only be applied to schools registered with the California Department of Education.

B) Self-Selection Bias

Hand’s Self-Selection Bias applies because our model assumes students select the school closest to their home. In reality, students from outside the Equity Priority Community (EPC) may self-select into an EPC school. They rely heavily on the bus-to-school path, but their residential data is dark to us because they live outside our study tracts. Conversely, students living in the EPC may attend the local school by walking, meaning our focus on bus stops may over-estimate the necessity of those specific transit paths for the immediate neighborhood.

We focus on the bus-stop-to-school link as a universal improvement. Enhancing this path benefits the highest-need users (those who have no choice but to commute) without harming the local walkers who share the same sidewalks.

C) Proxy Bias

Our analysis uses school enrollment data as a proxy for the potential pedestrian load on a given sidewalk. Hand defines Proxy Bias as using a substitute metric that may not perfectly reflect the phenomenon of interest.

We are not forecasting daily ridership or pedestrian counts; rather, we are ranking infrastructure priority. Chronic absenteeism, a form of dark data, means the actual daily usage may be lower than enrollment suggests. However, for the purpose of sidewalk improvement recommendations, we treat enrollment as the maximum potential “service burden” of a path. A sidewalk must be safe and accessible for the entire student body, regardless of whether every student uses it every day.

Instead of scaling down for absenteeism, we acknowledge enrollment as a ceiling for potential impact. This ensures our recommendations prioritize paths with the highest potential to serve the student population.

D) Intentional Hiding, Fraud, and Information Asymmetry

For intentional hiding, fraud, and information asymmetry, we assume that the California Department of Education and local transit agencies are reporting in good faith for public record.

E) Technical Limitations and Incomplete Records

While minor gaps likely exist in any large-scale portal, our data consists of static infrastructure and enrollment records rather than real-time sensor streams. Therefore, we do not expect random technical glitches to create the systemic darkness that would invalidate the spatial model.

Tile2Net, on the other hand, does have its limitations, especially as it was trained on dense urban cities on the East Coast of the US, not the Bay Area. We addressed this model noise with hand checking and cleaning of the output.

F) Experimental Conditions and Cognitive Biases

Because this is an observational study of existing infrastructure rather than a controlled trial, Experimental Conditions do not apply. Similarly, Cognitive Biases are minimized here by our reliance on algorithmic spatial joins rather than qualitative surveys or subjective human reporting.

G) Survivor Bias

While historically interesting, our study is a point-in-time diagnostic of current equity. We are measuring the existing physical relationship between today’s stops and today’s schools. We are not trying to explain why the network looks the way it does, but rather how well the surviving network serves the current student population.

H) Exhaust vs. Designed Data

Our study primarily utilizes Designed Data (census and enrollment records created for a specific purpose). While Exhaust Data (digital footprints like GPS pings from student phones) could theoretically provide higher resolution, its absence is a choice of scope rather than an unintentional gap. We rely on the designed nature of official records to ensure a consistent baseline across all EPCs.

I) Missing Data Types (Missing at Random [MAR] vs. Missing Not at Random [MNAR])

In statistical terms, most of our gaps are Missing Not at Random (MNAR). For instance, the omission of “learning pods” (Section A) is not a random glitch; it is specifically linked to the nature of those institutions. Because these values are MNAR, we cannot use standard imputation to “fill in” the gaps. That said, the focus of our project is in public schools where most students are represented.

K) The Modifiable Areal Unit Problem (MAUP)

Our analysis utilizes Census Tract boundaries as the primary unit of analysis to define EPCs. We selected Census Tracts because they are the smallest geographic unit for which the high-quality longitudinal demographic data required by the EPC framework (such as income-to-poverty ratios and vehicle ownership) is consistently available.

However, we acknowledge that Census Tracts are modifiable areal units; they are administrative boundaries rather than natural or behavioral ones. Hand’s framework warns that data can be “darkened” by this aggregation, a phenomenon known as MAUP. If these boundaries were redrawn—for example, shifting a border by one block—certain school zones might lose or gain EPC status, fundamentally shifting our priority recommendations.

4. Initial Results and Visualizations

The map below layers building-level footprints, MTC designated EPCs, and schools in EPCs. The EPCs highlight areas requiring focused MTC investment and support; users can toggle these layers independently to explore how school proximity overlaps with socioeconomic priority zones. The sections below include a detailed analysis of school access equity, using metrics such as Zero Vehicle Households (ZVH), socioeconomic status, race/ethnicity, and school types.

</div>

Map 2: Zero Vehicle Households and Schools Below the Poverty Line

A) Equity in School Access

The primary justification for our analysis on bus stop connectivity in relation to school access is the higher degree of transit reliance found within EPCs. As shown in Figure 1, census tracts within EPCs show a higher median proportion of households without access to a vehicle compared to Non-EPC tracts. Most Non-EPC tracts cluster below 10 percent, EPC tracts see vehicle-less rates between 10 and 20 percent with outliers as high as 75 percent.

This data establishes that for a significant portion of families in EPCs, walking or public transit is not a choice but a necessity when it comes to accessing mandatory daily school attendance. Thus, gaps in pedestrian networks or a lack of bus stop proximity or frequency represents a direct barrier to educational access for students. Consequently, SRTS programs should prioritize EPCs for infrastructure investment and policy intervention. Addressing these specific transit-to-school gaps is a pressing point of equity that allows policymakers and planners to direct resources where the ‘time-tax’ is most prevalent on disadvantaged households and their youth. The following figures provide an overview of our initial results regarding EPCs and public transit access disparities.

Figure 1: Distribution of Share of Zero Vehicle Households in Alameda County Census Tracts

B) EPCs as a Focal Point of this Analysis

The following data analyses of school demographics within EPCs compared to Non-EPCs further reinforces why SRTS projects should prioritize EPCs and takes a deeper look at who is being impacted by these transit gaps.

As shown in Figure 2, schools located within EPCs have a median socioeconomically disadvantaged (SED) population of approximately 92 percent with schools serving a population that is at least 80 percent disadvantaged. This variable was already defined by the California Public Schools dataset and this visualization draws on that variable. In contrast, schools in Non-EPC tracts show a much wider distribution and a significantly lower median of approximately 37 percent. This indicates that transit-dependent infrastructure in EPCs serves the most economically vulnerable students in Alameda County.

Figure 2: Distribution of Socioeconomically Disadvantaged Students in Alameda County Schools

The SED gap is also a racial one, as shown in Figures 3 and 4. The charts show that schools in EPCs serve a student population that is approximately 88 percent BIPOC with a 95 percent confidence interval. Schools in Non-EPCs serve significantly fewer BIPOC students, averaging approximately 72 percent of their student body. This further highlights that when students in EPCs face service gaps at bus stops, students of color are disproportionately impacted.

Figure 3: Average Share of BIPOC Students in EPC and non-EPC Tracts with 95 Percent Confidence Intervals

Figure 4: Boxplot of Average Share of BIPOC Students in EPC and non-EPC Tracts

According to the breakdown of school types in Figure 5, EPCs are more heavily served by public institutions at 81 percent than the countywide average, which is 74 percent. Thus, the burden of providing safe, accessible transit falls primarily on public infrastructure and public school district coordination.

Figure 5: School Type Breakdown (EPCs vs Alameda County)

While we map the demographics of schools within EPCs, a notable caveat remains as discussed earlier in the missing data analysis regarding students in the EPCs who may be commuting out to schools in Non-EPC areas. These students are not captured in this analysis and their travel paths may be more complex than the transit-reliance being analyzed in this report.

C) Generated Maps of EPCs

The following analysis details the spatial relationship between pedestrian infrastructure, high-frequency transit nodes, and K-12 educational institutions across three representative EPC tracts. By layering Tile2Net-extracted pedestrian geometries with bus stop and school data, we can identify specific barriers that contribute to educational inequity. As noted in the methodology section, 3 maps were generated for each of the 3 EPC study areas. The maps showing all gaps for the Nimitz and West Oakland EPCs are excluded from this section due to redundancy and relevance to analysis. These maps can be found in the appendix.

Oakland Airport EPC Accessibility Analysis

Figure 6 maps the complete inventory of structural and geometric deficiencies identified within the localized pedestrian network. These missing connections (highlighted in orange) represent missing sidewalks, crosswalk gaps, mid-block disconnects, or unpaved segments that disrupt contiguous pedestrian paths. Rather than analyzing these breaks in isolation, the spatial network pipeline evaluates them as systemic barriers that introduce artificial circuitry, forcing students to take significantly longer, less direct routes to their destinations.

Figure 6: All Identified Gaps in Pedestrian Network Oakland Airport EPC

Figure 7 models the pedestrian routing behavior from the nearest high-frequency transit hubs to the destination school campus.

The baseline condition (blue lines) reflects the current network routing constraints. Due to the infrastructure breaks identified in Figure 6, student pedestrians are forced onto circuitous paths, increasing exposure to traffic conflicts along major thoroughfares like International Boulevard and Hegenberger Road.
The optimized alternative (red line) simulates a counterfactual scenario where key infrastructure gaps are remediated. By synthesizing these broken links back into the pedestrian grid, the network model uncovers a highly direct, efficient, and lower-stress corridor that minimizes overall walking travel time and avoids high-injury corridors.

Figure 7: Adjusted Routes in Oakland Airport EPC

Figure 8 isolates the highest-utility interventions. By intersecting the theoretical optimal path (from Figure 7) with the complete gap inventory (from Figure 6), we pinpoint the exact conflict points that unlock the greatest network efficiency. The highlighted callouts indicate the specific, high-priority projects that yield the maximum reduction in student pedestrian circuitry per dollar spent. Remediating only these designated segments bridges the critical network divide, successfully converting a fragmented system into a safe, contiguous, transit-to-school corridor.

Figure 8: Most Important Gaps in Pedestrian Network Oakland Airport EPC

Nimitz EPC Accessibility Analysis

Figure 9 reveals a stark failure in the current pedestrian network. Currently, there is no viable contiguous route from the nearest high-frequency transit stops to the destination school. This structural void forces students into a connectivity desert, where public transit access is physically decoupled from the school. While the model identifies a potential new route after addressing primary gaps, the resulting path is exceptionally long and circuitous. This creates a “Logic of Avoidance”, a scenario where even a fixed route is so inefficient that no student would logically utilize it. The path is dictated not by the destination, but by the extreme detour required to navigate around the freeway’s physical footprint.

Figure 9: Adjusted Routes in Nimitz EPC

Figure 10 highlights the single most important project (highlighted in red) required to at least establish a physical connection across the divide. These maps visually capture the severance effect of the Nimitz Freeway. The highway acts as a massive physical barrier cutting through the heart of the community. Remedying this specific gap (Figure 10) technically yields a contiguous route (Figure 9), but the sheer length of the path proves that simple infrastructure fixes are often insufficient to overcome legacy planning decisions that prioritized vehicle throughput over community-scale pedestrian safety.

Figure 10: Most Important Gaps in Pedestrian Network Nimitz EPC

West Oakland EPC Accessibility Analysis

Figure 11 captures a shift in optimal path modeling. In the current network, student pedestrians traveling from the south must navigate a lengthy, rectangular loop around the neighborhood blocks to access the school campus from the north. This circuity occurs because the southern perimeter behaves as a closed boundary, blocking direct northern progression.

When the critical missing link is resolved, a linear path is established. Crucially, this does not just streamline the walk, it unlocks an entirely new high-frequency bus stop origin to the south. Students arriving via this transit point can now enter the walking network immediately adjacent to their transit drop-off, bypassing the circuitous perimeter walk entirely.

Figure 11: Important Gaps in Pedestrian Network West Oakland EPC

Figure 12 maps the structural gaps across West Oakland, highlighting the single project (highlighted red) that acts as the “gatekeeper” to this entire transit hub. While Figure 12 shows several minor missing connections (orange lines) throughout the neighborhood grid, the highlighted project at the southern terminus is the highest-leverage intervention.

In classic infrastructure planning, projects are often prioritized based on linear feet of sidewalk laid or simple gap lengths. However, this case study proves the value of a network-utility approach. Remediating this single, short segment yields exponential benefits because it bridges two distinct systems: the local pedestrian grid and a high-frequency transit node that was previously functionally isolated from the school.

Figure 12: Adjusted Routes West Oakland EPC

Comparative Case Study Synthesis & Regional Prioritization Framework

By moving from a regional macro-demographic analysis to network-level micro-routing, our localized assessment of Lockwood Gardens, the Nimitz Freeway Corridor, and West Oakland reveals that not all pedestrian infrastructure gaps are created equal. Equal physical lengths of missing sidewalks can yield vastly different impacts on real-world student mobility.

Airport (Path Optimization): This area represents the classic low-hanging fruit of pedestrian planning. The underlying street network is contiguous, but minor physical dropouts degrade pedestrian safety and routing. Remediating these high-utility gaps directly compresses student walking times and keeps children off high-injury corridors like International Boulevard.

Nimitz Freeway Corridor (Structural Severance): Nimitz serves as a cautionary tale of infrastructural barriers. The freeway physically isolates the school from neighboring transit corridors. While repairing the gatekeeper gap establishes a continuous link on paper, the resulting path is so circuitous that it creates a “logic of avoidance” for students.

West Oakland Corridor (Asset Activation): West Oakland showcases the power of a network-utility approach. The primary missing segment acts as a physical gatekeeper to the neighborhood’s southern border. Remediation does not merely improve an existing walk—it alters the network’s geometry, immediately activating a high-frequency transit node that was previously completely inaccessible to the school campus.

To maximize the return on municipal infrastructure investments, the Metropolitan Transportation Commission (MTC) should transition away from traditional prioritization metrics (e.g., funding projects purely based on linear feet of sidewalk laid or total gap length) and Focus on first- and last-mile in SRTS programming at a disaggregate, pedestrian scale and provide support for municipalities to build and maintain sidewalk inventories.

Appendix

Figure 13: Gaps in Pedestrian Network Nimitz EPC

Figure 14: Gaps in Pedestrian Network West Oakland EPC

References

Metropolitan Transportation Commission (MTC), “Bay Area Safe Routes to School.” https://mtc.ca.gov/funding/investment-strategies-commitments/climate-protection/bay-area-safe-routes-school.
N. Wexler et al., “Free Transit Passes and School Attendance among High School Students,” Transportation Research Record 2675, no. 10 (2021).
USDOT, Safe Routes to School (SRTS).
Safe Routes Partnership, “History of Safe Routes to School,” https://saferoutespartnership.org/safe-routes-school/srts-history/.
U.S. Department of Transportation (USDOT), Safe Routes to School (SRTS) Statewide Mobility Assessment Study: Phase I Report (2010).
Christine Voss et al., “School-travel by public transit: Rethinking active transportation,” Preventive Medicine Reports 2 (2015).
Sarah A. Cordes, Amy Ellen Schwartz, and Leanna Stiefel, “The Path to School: The Effects of School Commutes on Attendance and Academic Outcomes,” Educational Evaluation and Policy Analysis 44, no. 1 (2022): 45–68.
Kristin Blagg et al., The Road to School: How Far Students Travel to School in the Choice-Rich Cities of Denver, Detroit, New Orleans, New York City, and Washington, DC (Urban Institute, 2018).
Jon Valant and Jane Arnold Lincove, “Transportation Inequities and School Choice: How Car, Public Transit, and School Bus Access Affect Families’ Options,” Educational Researcher 52, no. 6 (2023): 335–347.
Blagg et al., The Road to School.
Metropolitan Transportation Commission (MTC). “Equity Priority Communities.” Plan Bay Area 2050.
Y. Guo and C. Brakewood, “Analysis of spatiotemporal transit accessibility and transit inequity of essential services in low-density cities, a case study of Nashville, TN,” Transportation Research Part A: Policy and Practice 179 (2024): 103934.
Michael A. Gottfried, “Linking Getting to School With Going to School,” Educational Evaluation and Policy Analysis 39, no. 4 (2017).
Wexler et al., “Free Transit Passes.”
UPPER Project (EU Transit Consortium), “Giving Catchment Areas a Reality Check: Incorporating Walkability into Transport Planning,” Mobility Strategy White Papers, November 2024.
California Air Resources Board (CARB), “Street and Network Connectivity: 2025 Policy Brief,” Sustainable Communities Research Series, Sacramento, CA, 2025.
Housing development: major transit stops: vehicular traffic impact fees, Assemb. Bill 2553, 2023-2024 Reg. Sess., ch. 275, 2024 Cal. Stat. (amending Cal. Pub. Res. Code § 21064.3).
David J. Hand, Dark Data: Why What You Don’t Know Matters, (2020).