The Impact of Predictive Bias in Transportation: Navigating Data Limitations and Assumptions for Effective Mobility Planning

Prediction models are integral to modern infrastructure planning, guiding daily decisions that impact millions of lives. However, when predictive models are biased, the outcomes can be costly and misaligned with public needs. Just as we have seen bias impact political polling, similar limitations and assumptions in predictive data can lead to issues within the transportation sector. These biases—whether rooted in data limitations, modelling assumptions, or socioeconomic disparities—can cause inaccuracies that misinform policy and investments, affecting everything from traffic congestion management to new infrastructure developments.

This article explores the biases within transportation prediction models and the consequences of these biases. By examining data limitations, socioeconomic factors, geographic representation, and assumptions about technology adoption, we can better understand the need for accurate, inclusive, and adaptable models in creating effective and equitable transportation systems.

1. Data Limitations in Transportation Modeling

Data is the cornerstone of predictive models in transportation, whether for traffic forecasts, transit ridership, or emissions estimates. Yet, even the most extensive datasets can be limited in ways that hinder predictive accuracy. One of the most visible examples of this limitation arose during the COVID-19 pandemic, which caused significant and lasting shifts in commuting and transportation patterns worldwide. According to the International Association of Public Transport (UITP), ridership dropped by as much as 90% in some cities during the height of the pandemic, a shift that models based on historical data failed to predict (UITP, 2020).

Moreover, urban centres dominate transportation data collection, while rural areas are often underrepresented. This imbalance can lead to projections prioritising urban needs and failing to account for the distinct transportation issues facing rural communities (Kockelman et al., 2018). For example, while urban predictions may emphasise public transit options, rural communities may rely more on personal vehicles due to limited access to public transit options. By focusing disproportionately on urban data, models may need to pay attention to essential factors that drive rural transportation needs, resulting in underfunded and poorly planned resources in those regions.

Continuous and inclusive data collection that captures a wide array of geographies and demographics is necessary to address these data limitations. This allows models to be more adaptable and reflective of actual transportation trends.

2. The Role of Modeling Assumptions and Bias

Transportation models are based on assumptions about travel patterns, technology uptake, and economic trends. While assumptions are necessary for building models, they can introduce significant bias when they do not account for dynamic societal changes. For instance, many models have historically assumed a consistent rise in private vehicle ownership; however, recent trends indicate a growing interest in car-sharing, biking, and public transit, particularly among younger urban populations (Sperling & Gordon, 2009).

Another prominent example is the optimistic timeline assumed for autonomous vehicle (AV) adoption. Many transportation models project that AVs will become mainstream within the next decade. Still, these models often need to pay more attention to potential barriers, such as regulatory delays, high costs, and public reluctance toward autonomous technology (Litman, 2020). As a result, cities may invest prematurely in infrastructure for AVs, with more evidence to support the likelihood of widespread adoption.

By updating models to account for the realities of behavioural change and regulatory uncertainty, transportation planners can make more informed decisions about infrastructure investments that meet current needs while remaining adaptable for future shifts.

3. Socioeconomic and Geographic Biases in Predictive Models

Socioeconomic and geographic biases are prevalent in transportation models and often result in inadequate planning for marginalised communities. Predictive models sometimes underrepresent lower-income communities, where transportation needs differ significantly from those in affluent, urban areas. Research by Pucher and Renne (2003) highlights that these communities face unique transportation barriers, such as limited access to affordable transit, higher dependence on public transportation, and longer commute times due to inadequate transit infrastructure. When models fail to consider these factors, the resulting policies and projects may cater disproportionately to higher-income areas, further marginalising underserved communities.

Geographic biases also affect transportation planning across different regions. Rural areas, where residents often rely on personal vehicles due to sparse public transit, are frequently overlooked in data collection and subsequent planning. Kawabata and Shen (2007) found that rural residents face distinct mobility challenges that need to be adequately addressed by models prioritising urban needs. For example, the assumption that people will easily access public transit networks is only held in areas where such networks are limited or non-existent.

Similarly, global predictive models often need to account for regional differences in infrastructure and transportation needs. European cities, which typically have robust public transit systems, require different predictive models than U.S. cities, where urban sprawl and car dependency are more common. Applying a one-size-fits-all approach to these varying contexts can result in misguided policies and funding misallocations (Litman, 2021).

4. Overreliance on Quantitative Data: The Need for Qualitative Inputs

Quantitative data, which is easier to collect and analyse, dominates transportation modelling. However, this reliance on quantitative inputs can create blind spots by ignoring qualitative factors, such as community sentiment, cultural norms, and individual preferences, which can significantly impact transportation choices. For example, individuals’ decisions to use public transit versus private vehicles are often influenced by perceptions of safety, cleanliness, and convenience—factors not easily captured by quantitative data alone (Ewing & Cervero, 2010).

A bias toward quantitative data can also obscure critical social issues. Policies promoting biking in urban centres, for instance, may not consider the need for more safe biking infrastructure or storage facilities in lower-income neighbourhoods. Studies by Talen and Koschinsky (2013) show that when models do not account for these qualitative factors, they may unintentionally perpetuate inequities by promoting policies that cater to a narrow demographic.

Incorporating qualitative data through community surveys, focus groups, and interviews can help bridge these gaps and provide a more nuanced understanding of transportation needs. By balancing quantitative data with qualitative insights, transportation models can better reflect the real-world diversity of human behaviour and preferences.

5. Technological Bias: Assumptions about Innovation Adoption

Predictive models often exhibit bias toward optimistic timelines for technology adoption, particularly in areas like electric vehicles (EVs) and ride-sharing services. Many projections assume rapid EV adoption due to decreasing battery costs and environmental policies; however, real-world EV adoption rates vary greatly depending on factors like infrastructure availability, regional incentives, and consumer willingness to adopt (Hannisdahl et al., 2013).

Ride-sharing is another area where predictive models sometimes assume widespread adoption without considering cost barriers, especially in lower-income and rural communities where such services may be too expensive or unavailable. A study by Clewlow and Mishra (2017) found that ride-sharing usage is heavily concentrated in urban areas, with rural residents needing more access to these services. When models assume equal access across demographics, they need to account for the social and economic divides that limit technology adoption, leading to projections that may not hold in diverse contexts.

Models must account for these disparities by recognising that not all regions or demographics have the same access to technology and infrastructure. Adjusting assumptions about adoption rates based on regional and socioeconomic realities can make predictions more accurate and actionable.

Case Studies: Real-World Consequences of Predictive Bias

Several real-world cases highlight the impacts of predictive bias on transportation projects and policies:

• New York City’s Second Avenue Subway Extension: Initially justified by predictions of high commuter volume, the extension has underperformed due to overestimated ridership projections. These projections failed to consider neighbourhood demographic changes and commuting preferences, leading to an expensive project that has yet to meet expectations (New York Times, 2019).
• Decline in U.S. Public Transit Ridership Post-Pandemic: Many transportation models anticipated a quick rebound in public transit ridership after the pandemic, assuming a return to pre-pandemic commuting patterns. However, remote work, increased vehicle usage, and persistent safety concerns have kept ridership levels lower than expected (American Public Transportation Association, 2022).

These cases illustrate the importance of continuously updating and calibrating models to reflect better-evolving realities and complex, multi-layered factors influencing transportation usage.

Conclusion

Predictive bias in transportation modelling can have widespread consequences, shaping policies, funding allocations, and infrastructure projects worldwide. Addressing these biases requires an inclusive approach to data collection, balancing quantitative and qualitative inputs and considering diverse geographic, socioeconomic, and cultural contexts. Furthermore, models must adapt to reflect changing trends, technological realities, and behavioural patterns to develop transportation systems that genuinely serve the public.

Inclusive, adaptable models are essential to creating equitable and resilient transportation solutions in an era of rapid change and rising demand for sustainable mobility. By recognising and mitigating predictive biases, transportation planners can foster systems that reflect real-world needs and improve mobility for all.